Exploring the performance impact of memory latency

THESE DAYS, SEEMINGLY every major memory module manufacturer is producing fancy low-latency DIMMs. These DIMMs are equipped with tricked-out heat spreaders and come in a variety of different colors, making them easy to pick out in a crowd. There’s more to them than funky cosmetics, though. The most exotic low-latency DIMMs are rated to run at extremely tight 2-2-2-5 timings at 400MHz. Unfortunately, you’ll pay a small fortune for the privilege. Low-latency modules cost close to twice as much as more pedestrian DIMMs, if not more.

Lower latencies are a good thing, of course, but how much can they really improve system performance? Are exotic, low-latency DIMMs worth the price premium? Read on as we explore the effects of memory latency on Athlon 64 performance in synthetic memory benchmarks, games, and real-world applications.


Low latency DIMMs: Worth the premium?

Memory latency?
Before diving into our benchmark results, it’s worth taking a moment to go over how memory access works and where the various latencies come into play. Memory is organized like a spreadsheet, with data stored in cells that can be identified by a corresponding column and row. Spreadsheets can also be made up of multiple sheets, and similarly, memory can be made up of multiple banks. If we want to access a specific cell of memory, the system must first activate the sheet, or bank, containing the desired row. Next, the system sends an active command to the desired row. Once the row is activated, the system can issue read or write commands to specific columns in the row. When reading or writing has been completed, a precharge command is sent to close the row.

There are delays between each of the steps in memory access. These delays are referred to as latencies and expressed as a number of clock cycles. Here’s a brief explanation of some of the most common, and important, memory timing parameters that affect access latencies:

  • RAS-to-CAS delay (tRCD) — The RAS-to-CAS delay occurs between the time a row is activated and when the first read or write operation is performed.
  • CAS latency (CL) — CAS latency refers to the delay between when a read operation is issued and when the data returned by that read is considered valid.
  • RAS precharge (tRP) — The RAS precharge is the delay between when a precharge command is issued to close a row and when the next active command can be issued.
  • Active-to-precharge delay (tRAS) — This latency actually spans several steps in the memory access process. The active-to-precharge delay refers to the minimum number of cycles that must elapse between an active and precharge command.

Of course, no discussion of memory latency would be complete without mentioning the DRAM command rate. The command rate is the delay between when a memory chip is selected and when the first active command can be issued. The factors that determine whether a memory subsystem can tolerate a 1T command rate are many, including the number of memory banks, the number of DIMMs present, and the quality of the DIMMs. Some memory manufacturers claim that their DIMMs are rated for operation with a one-cycle (1T) command rate.

Since latencies refer to delays, lower is better. That doesn’t mean you should hop into your motherboard’s BIOS and set each memory timing option to its lowest possible value, though. Memory modules are rated for a specific set of latencies at a given clock speed, and they’re generally not stable with lower latencies. A DIMM’s latencies are usually expressed as a series of four hyphenated numbers corresponding to the CAS latency, RAS-to-CAS delay, RAS precharge, and active-to-precharge delay. Low latency DDR400, for example, is generally rated for 2-2-2-5 timings at 400MHz. That refers to two cycles of CAS latency, RAS-to-CAS delay, and RAS precharge, and five cycles of active-to-precharge delay.


OCZ’s Enhanced Latency Platinum Rev 2 DDR400 rated for 2-2-2-5 latencies
 

Our testing methods
We’ve tested several different memory configurations to illustrate the performance impact the key memory timings settings, including DRAM command rate. Tests were conducted with a set of low-latency OCZ DIMMs rated for 2-2-2-5 timings at 400MHz. We also tested with 2.5-4-4-8 timings to simulate the performance of more affordable “value” memory. Some budget memory is rated with CAS latencies as high as three cycles, but since CAS 2.5 memory is already quite affordable, we’ve limited our testing to CAS 2 and 2.5. In addition to testing system performance with 2-2-2-5 and 2.5-4-4-8 memory timings, we’ve also tested each configuration with both 1T and 2T command rates.

All tests were run three times, and their results were averaged, using the following test system.

Processor AMD Athlon 64 FX-53 2.4GHz
System bus HyperTransport 16-bit/1GHz
Motherboard DFI LANParty UT NF4 Ultra-D
BIOS revision N4D623-3
North bridge NVIDIA nForce4 Ultra
South bridge
Chipset drivers ForceWare 6.66
Memory size 1GB (2 DIMMs) 1GB (2 DIMMs) 1GB (2 DIMMs) 1GB (2 DIMMs)
Memory type OCZ PC3200 EL Platinum Rev 2 DDR SDRAM at 400MHz
CAS latency (CL) 2 2 2.5 2.5
RAS to CAS delay (tRCD) 2 2 4 4
RAS precharge (tRP) 2 2 4 4
Cycle time (tRAS) 5 5 8 8
Command rate 1T 2T 1T 2T
Hard drives Western Digital Raptor WD360GD 37GB SATA
Audio nForce4/ALC850
Audio driver Realtek 3.75
Graphics NVIDIA GeForce 6800 GT with ForceWare 77.77 drivers
OS Microsoft Windows XP Professional
OS updates Service Pack 2, DirectX 9.0c

Our test system was powered by OCZ PowerStream power supply units. The PowerStream was one of our Editor’s Choice winners in our latest PSU round-up.

We used the following versions of our test applications:

The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.

All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

 

Memory performance
We begin with some synthetic memory subsystem benchmarks that should easily expose any performance differences between the various settings.

Command rate has a profound impact on memory bandwidth in both Sandra and Cachemem. The difference in performance between 2-2-2-5 and 2.5-4-4-8 timings is much more subtle, though.

Moving to Cachemem’s latency test, we see that the command rate again has a bigger impact on performance than the other memory timings options.

 

Cinebench 2003

Memory latency doesn’t have much of an impact on Cinebench 2003 rendering performance, but our tighter 2-2-2-5 timings are a little faster in the shading tests. This time around, the command rate’s impact on performance is less than that of the other memory timings. 2-2-2-5 isn’t all that much faster than 2.5-4-4-8, though.

Sphinx

Sphinx is a sucker for fast memory subsystems, so it’s no surprise that quicker latencies translate into better overall performance. Here, our 2-2-2-5 timings are close to 13% faster than more relaxed 2.5-4-4-8 latencies. Moving from a 2T to 1T command rate only improves performance by about 7%, though.

 

WorldBench overall performance
WorldBench uses scripting to step through a series of tasks in common Windows applications. It then produces an overall score. WorldBench also spits out individual results for its component application tests, allowing us to compare performance in each. We’ll look at the overall score, and then we’ll show individual application results alongside the results from some of our own application tests.

Only a single point separates our overall WorldBench scores, with the 2-2-2-5 configurations just edging out our 2.5-4-4-8 timings. Let’s break down WorldBench’s overall score into individual tests results to see if we can find a breakout performance.

Multimedia editing and encoding

MusicMatch Jukebox

Windows Media Encoder

Adobe Premiere

VideoWave Movie Creator

Tighter 2-2-2-5 timings improve performance in several of WorldBench’s multimedia editing and encoding tests, but never by more than a couple of percentage points.

 

Image processing

Adobe Photoshop

ACDSee PowerPack

WorldBench’s image processing tests don’t see much benefit from either lower memory latencies or a more aggressive command rate.

Multitasking and office applications

Microsoft Office

Mozilla

Mozilla and Windows Media Encoder

Mozilla does show a difference between the settings, both on its own and when paired with Windows Media Encoder. Still, the differences in performance between 2-2-2-5 and 2.5-4-4-8 timings, and between the 1T and 2T command rates, are only a couple of percentage points.

Other applications

WinZip

Nero

Neither Nero nor WinZip shows much preference for quicker timings or a 1T command rate.

 

Gaming performance
We conducted our gaming tests with two sets of in-game quality settings. First, we tested at low resolutions with medium quality levels and antialiasing and anisotropic filtering disabled. We then tested at higher resolutions and detail levels, with antialiasing and aniso, to better reflect how most users would play games on a system of this caliber. The latter settings may bottleneck performance at the graphics card, but that’s how things are with the vast majority of today’s games.

3DMark05
Since you won’t find anyone playing 3DMark05, we’ve limited our testing to the app’s default settings.

Although 3DMark05’s overall score is GPU-bound on our test system, the CPU rendering tests show some preference for a faster command rate and tighter timings.

Far Cry

Lower memory latencies give Far Cry a nice boost at a low resolution and detail level, but there’s virtually no difference in performance at higher resolutions and detail levels.

 

DOOM 3

As we saw in Far Cry, the measurable performance impact of memory timings and the DRAM command rate in DOOM 3 seems to be limited to lower resolutions and detail levels, where the graphics card isn’t a bottleneck.

Quake 4
Timedemos in Quake 4 don’t appear to render all of the game’s eye candy effects, but since we’re only changing memory latencies and command rates, that shouldn’t impact our results.

Again, we only see memory latency and command rates having an impact on performance at lower resolutions and detail levels. Here, CAS latency and its cousins have a more significant impact on performance than the DRAM command rate.

 

Unreal Tournament 2004

Unreal Tournament favors lower latencies ever so slightly at higher resolutions and detail levels, but the biggest performance impact remains at lower resolutions and detail levels. Even then, there’s only about a 7% difference in frame rates between 2-2-2-5-1T and 2.5-4-4-8-2T.

Splinter Cell: Chaos Theory

Memory latency doesn’t have much of an impact on Splinter Cell: Chaos Theory performance, even at a lower resolution and detail level.

Battlefield 2

Battlefield 2 performance was tested with FRAPS, and the differences in performance are only a couple of frames per second. I suspect that’s within the margin of error of our manual tests, which were conducted at least five times before the results were averaged.

 

Conclusions
Although tighter memory timings and a 1T command rate can certainly improve the performance of the Athlon 64’s memory subsystem, that improvement doesn’t always translate to better application performance. In fact, with the exception of the Sphinx speech recognition engine, moving to tighter memory timings or a more aggressive command rate generally didn’t improve performance by more than a few percentage points, if at all, in our tests. Lower latencies only improved WorldBench’s overall score by a single point, and performance gains in games were generally limited to lower resolutions and detail levels.

So how much does the modest performance improvement brought by tighter memory latencies cost? Close to twice as much. As I write, a single 512MB stick of OCZ Value DDR400 memory rated at 2.5-4-4-8 sells for between $45 and $52 online, while a 512MB Platinum Rev 2 2-2-2-5 DDR400 module sells for between $81 and $94. Looking at dual-channel kits, a pair of 512MB OCZ Value DDR400 DIMMS rated for 2.5-4-4-8 timings sells for between $91 and $103 online, while a pair of 512MB Platinum Rev 2 sticks rated for 2-2-2-5 costs between $155 and $191.

OCZ isn’t the only DIMM maker charging that sort of premium for ultra-low-latency modules. In fact, it’s common. To cite another example, a pair of 512MB Corsair Value DDR400 DIMMs rated for 2.5-4-4-8 will set you back between $80 and $159, while a couple of the company’s 512MB TWINX1024-3200XL 2-2-2-5 DDR400 modules run from $189 all the way up to $325.

For most users, the price premium associated with exotic 2-2-2-5 memory won’t be worth the relatively modest performance gains that it offers. Low-latency memory does have an ace up its sleeve for overclockers, though. Most low-latency modules are capable of running at much higher clock speeds if you back off on their latencies a little. We’ve had our OCZ Platinum Rev 2 DIMMs, which are rated for 2-2-2-5 latencies at 400MHz, cranked all the way up to 560MHz with more relaxed 2.5-4-4-8 timings. Overclocking success is never guaranteed, of course, but low-latency memory modules tend to use higher quality chips that respond better to overclocking.

At the end of the day, the appeal of low-latency memory modules may be limited to overclockers and enthusiasts intent on squeezing every last drop of performance from a system. More pedestrian “value” memory should be plenty fast enough for everyone else, especially since you can practically afford twice as much. 

Comments closed
    • absinthexl
    • 14 years ago

    I love all these arguments. Sweet, sweet cognitive dissonance.

    • Xenolith
    • 14 years ago

    Where can I download those Quake 4 timedemos (trhangar1 and trtram)?

    • volcano
    • 14 years ago

    Interesting, because I’ve found Athlon 64 systems respond VERY well to low memory latencies. It’s obvious you’re limited by some other component in some of your tests, and this has absolutely nothing to do with the memory performance. Props for taking the time to do the testing, but I think, like others have said, you’ve come to the wrong conclusion.

    • Krogoth
    • 14 years ago

    OMG it’s the ./ effect!

    • Chrispy_
    • 14 years ago

    I have GeiL 2.5-4-4-8-1T by the bucketload.

    Silky smooth performance from having lots of RAM is far superior to slightly higher peak performance but annoying lows from disk swapping.

    Unless you can afford 2GB of premium modules, I’d take 2GB of cheap RAM over 1GB of premium RAM any day of the week.

    • Hector
    • 14 years ago

    I have been saying the same thing for a year now.
    §[<http://forums.anandtech.com/messageview.aspx?catid=28&threadid=1475190&enterthread=y<]§ I am Zebo at anandtech and even witten a couple pieces there. Just one point of note: You can get Crucial Ballistix 512MB PC3200 for $59 (newegg/mwave) which is not only faster memory than OCZ used in this test when both running 2-2-2 but cheaper too. So "perfomance" ram may be worth it today. Used to be high performance mem like TCCD/BH5 etc was hundreds more than value ram.. Like $90 vs $250 for a 1G setup. Today it's $80 vs $120. Also, if you know what to look for and how to overclock you can get value ram which will run 2-2-2 and/or high bandwidth. For example Crucials value sticks use same Micron mem as ballistix (micron G).... course they are NOT speed binned so YMMV Finally I'm not sure why you ran 2.5-4-4 to illustrate cheap ram. Any values sticks today can run 2.5-3-3 and 1T CMD in 2 x 512 config... I've tried patriot, corsiar, Mushkin and or course crucial all successful.

    • Zenith
    • 14 years ago

    Oh, and nice article Geoff. Quality as always. I like your approach, which appeared to aim at a more real world affect, rather than what impact it has on a super-tricked out SLI rig with it’s own nuclear reactor. 😀

    Never a dull moment around here, I like that.

    • albundy
    • 14 years ago

    Whats the point when GDDR-4 will be released by the end of this year. Anywho, great read. I would have liked to see ECC ram in the benchmarks too in case enthusiasts want to go the opteron way.

    • Drewstre
    • 14 years ago

    Now that was a great explanation. I’ve often wondered exactly what the numbers mean. That was the best explanation I’ve read. I think I grok it now. Thank you TR. Good stuff.

    Drewstre

    • Porkster
    • 14 years ago

    Isn’t it bad advice to recommend investing in old technology? DDR1 is finished and AMD items have near zero resale value due to their budget no-name style branding.

      • Krogoth
      • 14 years ago

      IT’S BACON TIME!

      on a more serious note, DDR1 is going to be here for a while, but DDR2 will be the cheaper and more plentful solution. AMD already has Socket M2 on the way and it’ll be out by the time DDR2 becomes really afford compaired to DDR1 kits.

        • Porkster
        • 14 years ago

        That doesn’t make the advice good so as to invest in old DDR1.

        If you ask computer owners, most will say they would like some resale value on new hardware they purchase. DDR1 is more or less useless for resale, even more so if AMD shifts off the current old interfaces, ala DDR1, Socket 939, etc, which will be the case.

        AMD’s biggest problem currently is that you would be foolish to buy AMD whilst they are stuck in the year 2003.

          • tfp
          • 14 years ago

          Dude it doesn’t really matter they could have done test with PC133 or SIMS and they would have probable come to the same conclusion.

          If you can extrapolate the data you see there to other memory products with the same sort of latency setups you need to start thinking a little harder.

        • Porkster
        • 14 years ago

        Also I make valid points, so no need for spamming or flaming my nickname so a to distract from the possitive message I give.

          • tfp
          • 14 years ago

          Well I’m pretty sure you are making points how valid is questionable.

            • sigher
            • 14 years ago

            DDR2 has its issues and might actually be a curse on an AMD system.
            intel uses DDR2 and are their system that much faster? the answer is no.

            • tfp
            • 14 years ago

            What the hell does that have to do with figuring out if one latency step is with the extra money?

            Pretty much nothing. Like I said in my other post any kind of ram with comparible latency design will be affected the same way. Thats why they DON’T need to run the test for SD-RAM or DDR2 you should be able to figure it out on your own.

            I know that is alot to ask of from some people but hell give it a shot.

            (I don’t necessarily mean you sigher)

            • Shintai
            • 14 years ago

            You mean issues like:

            1. On die termination instead of on the motherboard.
            2. 1.8v instead of 2.5v = less heat, less power and higher reliability.
            3. Up to 8GB DIMMs without chipstacking compared to DDRs 2GB.
            4. Up to 12.8GB/sec with dual DDR2-800 compared to 6.4GB/sec with dual DDR400.
            5. Supports more than 2 dual-rank DIMMs per channel, unlike DDR333/400.

            I can´t see any advantage of DDR vs DDR2. As shown the latency isn´t an advantage.

            • Spotpuff
            • 14 years ago

            And those points translate into what performance at the moment?

            I don’t plan on upgrading to DDR2 until there is a clear performance advantage. There isn’t one now, nor will there be for the foreseeable future I would guess. Technology always takes a while to show its benefits. Even PCI-E at the moment has none over AGP; it’s mostly just a future-proofing step to buy a PCI-E video card.

            Having learned from my experience in early-adoption of tech, I will be hanging back and letting all this crap play itself out before I buy anything again.

            • tfp
            • 14 years ago

            More bandwidth is the main part, with the right controler. Latency will not be to different when AMD gets there on board control all set.

            The point here is DDR2 is not crap.

          • Zenith
          • 14 years ago

          Baaacccooooon, Baaccccooooooooooooooon.

          No, but really, DDR1 isn’t going to just disappear one day. It’ll be phased out.

          You may not care about AMD products, but we do. So neener neener neener.

          Seweeeeeeeeee piggy piggy piggy.

      • Convert
      • 14 years ago

      Actually if you followed the SDR to DDR1 transition you would have known SDR had resell value and hefty one at that. Check out the prices right now, pc133 sells for about the same or more than ddr1.

      In fact ram has always been one of the best parts for resell, it is just a little different from the rest of the pc parts. Take a video card for example that are originally sold for hundreds of dollars; they go down in price extremely fast and are quickly worth $30 a few generations later. Ram dips to liquidation prices when something new comes out, then after that dies down it jumps in price. So all you have to do is be like any other seller with half a brain and pick the right time to get rid of your parts.

      Even with ebay prices for used sticks of 256mb pc133 they don’t really dip below $20. Considering even pc3200 goes for $20 brand new it supports my previous statement. Pc3200 will jump in price after amd moves to ddr2 (in fact it has already started to rise in price). Just like it has done in the past time and time again.

      Sorry porkster but you are wrong once again.

        • Porkster
        • 14 years ago

        Convert, you were warned not to reply to my posts as you never agree, you always take the oppsoite approach even it’s it’s blatantly false.

          • Zenith
          • 14 years ago

          You were warned about posting here. If you’re not going to listen, why should Convert?

          Anyway, you will always get these kinds of responses because your reputation on the internet is being that of a troll to the highest degree.

          Seweeeeee piggy piggy piggy.

          • Convert
          • 14 years ago

          Porkster, you were warned not to post as you are a troll, you always take the opposite approach even it’s it’s blatantly false.

          Seriously though I have never been warned. Unless you mean warned by you, although you have no authority to warn me of anything.

          Interesting to see you didn’t even try to discount anything I said.

            • Porkster
            • 14 years ago

            I was the one stuggesting that AMD was behind when they didn’t have PCIe, and other Intel style features, look now all new AMD boards are following the Intel market. It’s the same with DDR1, it’s old and finished. The only reason why AMD wont go DDR2 now is that they will leave themselves behind some benchmarks at the cost of their fans buying into old interfaced hardware.

            It’s a pity AMD is keeping the market behind. Imagine if all the memory market was aiming at making DDR2 better, rather than investing time and resources in older format for the AMD customers.

            • Krogoth
            • 14 years ago

            Dude, when DDR2 first came out for the Prescott. It was far more expensive per MB then DDR1 that was just one of nails against the Prescott platform for ethusiast. At the current moment, DDR2 has finally made price parity with it’s DDR1 counterparts.

            • IntelMole
            • 14 years ago

            …. And look at all the PCIe devices now that AMD have let the market catch up.

            Yeah. Right.

            Same with DDR2. There was zero performance advantage at the time by going with DDR2, still very little now unless you use exotic RAM. And from the way that the AMD chip works (being more dependant on latency, which is increased by DDR2), they stood to lose in transitioning to DDR2 now. By the time Socket M2 comes about though, DDR2 should be a clearly better performer.

            Sounds like a good transistion to me. But my brain hasn’t been made into pig feed, so maybe I lack your perspective.

            Why do you still post here?
            -Mole

            • tfp
            • 14 years ago

            At the increased DD2 speed the latency will be farely close to the DDR1 timings.

            And did we learn nothing? When it comes down to it, it probable will not make THAT much of a difference.

            Here are some simple latency comparisions:
            DD2-667 5-5-5-12 is equivalent to DDR1-400 3.333 – 3.333 – 3.333 – 8

            But there is a hell of a lot more bandwitdh. With DD2-800 5-5-5-12 your down to equivalent to DDR1-400 2.5 – 2.5 – 2.5 – 6 timings with double the bandwitdh or you can go back to running single channel with the same bandwidith we have now.

            • Porkster
            • 14 years ago

            It’s maybe why AMD is struggling with new games like FEAR.

            • IntelMole
            • 14 years ago

            You know this because you have tested an M2 box?

            And since when were AMD ever struggling in gaming?

            Come back when you have a clue.
            -Mole

            • tfp
            • 14 years ago

            Dude what the HELL are you taking about? Do you even have a clue?

            • Convert
            • 14 years ago

            I am afraid you painted yourself into a corner my porky friend. So let’s simplify this: AMD is a no-name company, and I will throw in the fact they only command around 16% of the market. Yet they are slowing the memory market down? You do realize ram is still being sold to older intel systems as well right? In fact there are probably more ddr1 intel systems out there than amd, you can be sure of it when one side only holds 16%.

            Since you don’t follow the ram market, at all, ddr2 is making great advances. The memory is getting faster with lower latency and the best part of all? DDR2 can actually be had for less if you aren’t interested in timings. Now all they have to do is work on the lower latency parts.

    • Shinare
    • 14 years ago

    Another great article. Thanks for clearing up what I have suspected for a long time to be true about “enthusiast” or “Premium” memory. You money is better spent on buying a more powerful video card when talking gaming rigs. Or a more powerful CPU when talking about everything else. Unfortunatly I fell for it hook line and sinker and have 2x $100 512MB sticks of Kingston HyperX DDR500 ram in my system when $79 worth of ValueRam would have been sufficiant.

    :<

    • jutta
    • 14 years ago

    I have an AthlonXP 2500+ Mobile at 2.3Ghz, I have 3 sticks of DDR400 RAM that I run at 333, two of those are TCCD, one is not, so it’s 333 @ 2.5-3-3-11

    I play games fine.
    That is all I need to know.
    I was overclocking-innoculated after the wallet and dissatisfaction virus hit me.

    You’re never happy with any speeds when you try to overclock 🙁

    • a_non_moose
    • 14 years ago

    So, my question is:

    What about the 1G modules with the 2325 timings?

    I’d queried over at Ars, and got an ok answer as far as it working with 2225, but I could have sworn one of the ‘big things’ about the A64 was the memory controller luvin’ lower latencies.

    But honestly, I could not care less if a measly 7% loss in speed translated to FEAR, et al not hitching so damn much.

    Oh, and one question to Damage: Does the 560Mhz on the ram translate to a x3 multiplier for Hypertransport?

    IIRC from my foray into pushing my system a bit, FSB@210Mhz, 1T CR and 2225 wound up putting the HTX bus to 840Mhz.

    That seemed to be about all it could take, IME, but part of a problem I discovered was the PSU was adequate when I got it, but showed its age when other upgrades came along. (9800pro -> X800 pro, couple of cheetas and 300G SATA drive and 16XDVDRW drive) 16Amps on the 12V rail did not do it anymore.

    Bit the bullet and got an enermax with 36A on the 12V and things smoothed out quite a bit (no more BSOD’s when starting up the 9 internal drives, so far) plus the memory V is a lot more stable than the old PSU’s.

    So, I guess it boils down to this: For an A64, does the lower latencies or the higher Mhz make a diff?

    I would not mind relaxing the mem, if I could push the 2Ghz up a bit for CPU bound games, but the current crop of games wants 2G to be happy.

    <sigh>

      • FroBozz_Inc
      • 14 years ago

      I would also be curious to see the difference in the 512 versus 1G modules.

    • PerfectCr
    • 14 years ago

    I’ve never bought that super expensive stuff, just not worth it. I use Corsair Value Select 2.5 Dimms and they work great. Why pay more?

    • Beomagi
    • 14 years ago

    #33 is right.

    At higher resolutions you’re forcing the system into being videocard bound. So it’s no surprise that lower latency memory come in at the same speed just about. Hell i’ll bet that there’s little difference using a faster cpu as well at those high resolutions since the 6800GT would be bottlenecking.

    I’ve seen comparisons before, and in general budgetram at 2.5-3-3 is generally worth the oh, $10 over 2-4-4 slower ram.

      • Damage
      • 14 years ago

      That’s the whole point. From the article:

      y[

        • SnowboardingTobi
        • 14 years ago

        Just a tiny little nitpick — on the first page where the memory timing parameters are listed in bullet point, how about you switch the order of tRCD and CL? If you do then it’ll put it in order of how timings are read.

        edit: d’oh. This was supposed to be its own thread. ah well

    • Usacomp2k3
    • 14 years ago

    Great read, thanks TR.
    I would be interested in seeing if the timings have any effect on a dual-core system running your masssively-multithreaded-do-a-million-things-at-once test.

    Now the test will be if 2GB cl2.5 is better than 1gb cl2, with overclocking in mind *sigh*

    • kun.woo.lee
    • 14 years ago

    I think it would be interesting to see the effects on networking. Most of the benchmarks shown aren’t dependent on latency too much. Actually when you goto dual or more cores, I think the memory latency will become more important since there are more agents fighting for memory access.

      • Shintai
      • 14 years ago

      Thats why a shared L2 cache dualcore design is the future.

        • tfp
        • 14 years ago

        Or the past…

          • Shintai
          • 14 years ago

          You don´t really know the benefits of shared caches do you?

          Only reason one would need low latency memory is if you had to seek the main memory all the time. So when you have a multicore design with shared cache all cores can access the content in it. So say for dualcores, an AMD64 with 2x512KB caches will need alot more access and fast access to the main memory when the cache memory doesn´t hold it compared to a single 2MB shared cache dualcore P-M.

          The larger and smarter cache, the less need for low latency memory and in some cases less need for bandwidth too.

          low latency memory is basicly only needed by Celerons and Semprons, kinda rather ironic.

            • tfp
            • 14 years ago

            Whatever man my point shared caches have been done before and I agree that it is probably the way to go. But there are disadvantages to using a shared cache but from your response I can tell you already know so much I’m not going to bother.

    • Larc
    • 14 years ago

    it was an ok article… I have a problem with the test rig.
    I dont like the pairing between the CPU and GPU at all.
    How many people go with a middle of the road video card (6800GT) and an expensive processor (FX53).

    All that was proven in the gaming benchmarks is that with a powerfull CPU, the extra memory boost does not make a difference when a benchmark is GPU bound.

    in the case where you were not bound by the GPU the memory settings did have an impact that people would pay for.

    so if someone had a better video card should they buy the better memory?

    What if someone has a 3200, they dont have as much cache and their cpu is weaker… at what point does the system become CPU bound? and i bet at that point memory timings would make more of a difference.

    Reading the article is like watching the myth busters. their intentions are good but sometimes their logic makes me wince.

      • nstuff
      • 14 years ago

      I agree. At the higher resolution and quality settings, it became obvious that the memory latency was not the bottleneck and made little difference. Further investigation may include mixing the current config used with a FX57 and a GeForce 7800GTX. FX53 + 7800 vs FX57 + 6800 vs FX57 + 7800…etc. The results in the article simply begs the question… why? We know that to focus a game benchmark on cpu power, we use the lowest supported resolution, otherwise, the gpu takes over. Is that the case here? It sure looks like it.

      Otherwise, very enlightening article, as always.

        • Larc
        • 14 years ago

        exactly, i know it would have taken more time, but a different processor speed would have been interesting in all of the benchmarks.

        you would likely see the same fps during the games.

        so the question usually comes down to, is it better to spend an extra $100 on my memory or a faster processor. often a small boost in processor speed produces only a marginal improvement as well.

        My biggest complaint is that they are testing memory latency on a chip that has twice the cache of my processor. I fail to see how this setup gives me any insight to the performance difference on my system

          • Usacomp2k3
          • 14 years ago

          Actually, for gaming, you’d put that $100 into a better video card 😉

        • spuppy
        • 14 years ago

        That’s the entire point this article makes: in resolutions people actually PLAY, there is no performance impact.

          • Larc
          • 14 years ago

          you missed my point… the test is flawed and there is not enough info to make a conclusion. yet somehow people are using this article to say “see, i knew i was right!”

          maybe peoples conclusions are 100% accurate, i’m not saying anyone is wrong only that people need to learn to look for flaws in studys and tests. often they are missleading.

          points:
          1) i dont know anyone who has a fx53 with a 6800gt.
          upgrade the video card or put on a cheaper cpu with lower ghz and a smaller cache… now see what happens when you are at game playing rez.

          i’m not saying there WILL be a difference only that the test is severely flawed.

          to take this to an extreme, i bet i could make a review on the raptor and show that on my p2-233 that windows 2k doesn’t run any faster… therefore dont waste your money on a raptor. whatever the outcome, its not a good test.

          2) cpu speed?
          if you are going to test memory, use a cpu that is in the middle like a 3200. it only has 512KB of cache. its way more common than a FX processor and with the smaller cache it might make more of a difference.

          test it with a cpu just a bit faster. if you get a 3% increase from faster ram on some test that matters to you, and you only get a 3% increase from the faster cpu, then maybe memory makes sence for you.

            • swaaye
            • 14 years ago

            Pointless. The conclusion is certainly relevant. If you put the load on the video card, through high resolutions and detail settings, the effect of memory latency is utterly negligible. The fastest video card in the world isn’t going to change that. Unless you run at quality levels where the video card is not the bottleneck.

            • Larc
            • 14 years ago

            Here is an example…

            Lets say you are going to build a new computer… For argument sake, its a 3200+ and you bought a 7800GT. I think this would be a very reasonable choice for someone.

            in this case you have
            1) a CPU that is much slower than the tested one
            2) a CPU that is more dependant on memory latency/speed than the tested one (becuase there is less on chip cache)
            3) a much more capable GPU than the tested one.

            Each of the above items COULD make the memory timings much more critical.

            If they had ran the test on a weak CPU and a strong GPU and found that memory timings didn’t matter, you CAN then say that for anyone who has a weaker GPU, better CPU, or both, that memory timings dont matter.

            This is why i conclude that there is not enough info to conclude anything for most people, however I keep reading comments and my point is oblivious to them.

      • Xylker
      • 14 years ago

      Uh… I dunno, the FX53 also goes by the name 4000+ (~$350 US)

      As such, I think that there are pretty good odds that the 6800GT and 4000+ combination is fairly representative. No?

        • Beer Moon
        • 14 years ago

        Dunno, if you can afford a CPU that expensive you’ve probably got SLI or at least a top of the line GPU.

        I think it’s a good point to note that at some point between 640×480 and 1600×1200 4xAA 8xAF the test became GPU bound, but since I generally play most games at 1024×768 or 1280×1024 on my 6800GT with NO AA and occaisionally AF turned on, I would like to know exactly where the lower latencies came into play.

        Somewhere between 640×480 NoAA NoAF and 1600×1200 FullAA FullAF latency stops paying off. Quite frankly, the difference between my single stick of Patriot Value 1024×1 DDR400 CAS 2.5 and my EL Plat Rev 2 DDR400 was $70. $95 for the single stick single channel RAM and $165 for the EL Plat Rev 2. Why not toss in a single-channel CAS3 1024×1 stick in the mix and see where that fails you?

        My Patriot Value RAM single channel stick wouldn’t do more than 4Mhz over stock at any voltage. My OCZ EL Plat Rev 2 does 500Mhz at 2.5-4-3-7 at 1.7v and lets me get my CPU up to 2.6Ghz (Venice 3000+). I assure you the difference is measurable. Worth $70? For anyone? Maybe not. For me? No doubt about it.

      • Lazier_Said
      • 14 years ago

      The point isn’t which specific video card is or is not GPU limited, the point is that a reasonable CPU with the slowest available memory will already generate ridiculous framerates (166fps, 209fps, 112fps, etc) to the point that incremental improvements from expensive low latency memory are humanly indistinguishable even with a video card fast enough to render them at real world image quality settings.

        • Larc
        • 14 years ago

        My point is that the article proves nothing to a gamer.

        please read my post #63.

          • swaaye
          • 14 years ago

          LOL If you say so Larc.

          I think it proves it fairly definitively. You want to blow 2x the $$ for a 1% performance boost, go ahead. Me, I’ll save that money and buy a higher end CPU for a 10% boost with that saved money.

          I’d say the article proves without a doubt that unless you’re some wacky gamer than runs 1024×768 on his $350+ GF 7800 to make sure he’s CPU limited, low latency does basically zip for ya.

          Now, RAM latency may just have an effect on some specific apps, but it’s obviously not a global improvement. And for the price premium it sure as hell better be.

            • Larc
            • 14 years ago

            In a discussion you must present facts and either backup your position or discredit the other.

            You make no arguments only accusations. Please tell me where my logic is wrong or add new statements to back up your point. laughing doesn’t prove anything.

            I keep trying to make my point simpler, but I dont know how to make it any easier to understand.

            If the article wanted to _[

            • Lazier_Said
            • 14 years ago

            1. 2.4ghz, dual channel AMD64 is faster than the typical? Anyone who buys bling memory or reads a website like this one overclocks, in which case that processor setup is at best typical and more likely slow.

            2. The test should have included a 512K cache processor for comparison. However, other web reviews, eg. at T-break have tested memory timing impact on other CPUs with varying caches. None of these show a significant performance impact either. Perhaps there would be with a 128K Sempron.

            3. Benchmarks were also taken at non GPU limited resolutions. These showed that even the slowest 3-4-4-8 memory configuration was still generating framerates well over 100 and usually closer to 200fps in current games. If that 6800GT was swapped for SLI 7800s, the GPU limited and non limited graphs would remain exactly the same, the only difference would be the resolution and AA/AF eye candy levels for a given graph.

            Perhaps you could now run 1600×1200 with 4AA8AF at CPU limited speeds. And 2048×1536 with SLI AA at GPU limited speeds. The point remains that no change in memory timing would have any impact on your gaming. You would still have a choice between two stupidly high rates at one setting, and two identical rates at the other.

            • Larc
            • 14 years ago

            I can see your arguments. Finally someone tries to back up their statements. In general you are probably right

            You did however use info that was not presented in the article. My argument was never that anyone was wrong, only that the article proves nothing.

            take the Far Cry graphs for example… there is a 12% improvement at 800×600 and 0% at 1600×1200. I think we can all agree that this is expected.

            However, these are two extremes. What do the combinations of these graphs say about 1280×1024? nothing… you cannot look at the two endpoints and make assumptions about what is in the middle. its logically unsound.

            I know there was a lot of time spent on the review and testing, that is part of the reason that I felt I had to say something. I just hope that people think about what they are actually testing and come up with a setup that will test what you want.

            Re: /[

      • provoko
      • 14 years ago

      I agree. Plus the test should have been with a 7800 GTX. You would have seen a big difference between 2,2,2 vs 2.5,4,4 with a much better video card.

    • Spotpuff
    • 14 years ago

    How does the P4 fare in all this with its horrid northbridge memory controller? Aren’t things only going to get worse when they start doing dual-cores and their dual-link to the northbridge solution in the future?

    Pretty sure AMD has the upper hand here (as well as everywhere else) over Intel.

      • Krogoth
      • 14 years ago

      Firstly, Intel is killing off the Prescott Dynasty so, the upcoming Presters and Cedar Hills are the last of the P4s. Secondly, Intel is already developing a interface very silimiar to HTP for interboard commutications. It’ll be utillize with their next generation parts (Melom, Conroe and Woodcrest).

        • Shintai
        • 14 years ago

        That would mean Yonah got it too..since you can slip a Merom into a Yonah motherboard.

        • Spotpuff
        • 14 years ago

        No on-die memory controller though right?

        So, I don’t know how the latency figures look. I guess it’s all speculation at this point anyways, but I think P4 figures are still relevant considering that’s what Intel is selling now and for a while still probably.

          • tfp
          • 14 years ago

          It probably looks about the same as AMD in respects to performance improvement just add another 60 or so to the latency numbers…

      • Shintai
      • 14 years ago

      Well I don´t have a P4, but only a P-M. However it´s the same northbridge setup.

      Running dual DDR2-667 4-4-4-8 compared to 667 6-6-6-15 hardly do any difference. But running at 533 3-3-3-6 is quitre abit slower than running 667 6-6-6-15. So I guess like the P-M, the P4 will go for bandwidth over latency.

    • axeman
    • 14 years ago

    Of course ram latency won’t matter at high-res in games when the frame rate starts to be limited by the graphics card. I would be curious if ram latencies make any difference to video encoding.

    edit:whoops, I think I skipped that page, since it was called “world bench” Still, too bad the video encoding benchmarks aren’t software geeks use. Am I going to encode anything with WMV? Ever? I don’t think so.

      • Vrock
      • 14 years ago

      WMV is good for making small size high quality videos for playback on mobile devices. And Windows Media Encoder is free. Two viable reasons to encode with WMV.

    • Shintai
    • 14 years ago

    (pets his higher bandwidth DDR2) 😀

    • spiritwalker2222
    • 14 years ago

    I would have liked to see some 2-3-3-6 DDR400 1T ram in there. I was able to pick some up for the same price as value ram, and would like to see how it fares compared to 2-2-2-5 ram.

    Although I’m sure my cheaper ram doesn’t have as much head room (haven’t tried) as the more expensive stuff.

    • Logan[TeamX]
    • 14 years ago

    The responsiveness that the entire system exhibits at lower latencies, as well as the higher performance for F@H (especially 600-point Gromacs WUs) more than justifies the OCZ EL Plat Rev. 2 I have in my box.

    For basic users, yes I agree they don’t need Plat Rev. 2. But, on the other hand, they’re not the targetted audience. By the same token, not everything relies on memory bandwidth when they’re obviously CPU / other sub-system-bound (i.e. Nero – WTF!?! Seriously, Nero??? In a RAM test. Come on).

    Your memory latencies test shows this in all too-clear detail. CAS 2.5, 4-4-8, 2T command rate yields a ~50ns memory latency. Moving to CAS 2, 2-2-5, 1T command rate yields a ~41ns latency. That’s 20% more response just by shifting latencies, without overclocking / over-volting the RAM. Anyone who is a dedicated performance enthusiast can reasonably feel the difference in their system, from system startup times to game load times, to application responsiveness. That’s just common sense as far as performance in computing goes… lower your latencies, improve responsiveness and overall performance.

    I’ve run my EL Plat Rev. 2 @ CAS 2.5, 4-4-8, 2T, and I’ve run it at CAS 2, 2-2-5, 1T, with my Winchester 3200+ @ 2.4. There’s a marked difference, about the same as going from DDR400 to DDR480, CAS 2, 4-4-7, 1T command rate.

      • Lazier_Said
      • 14 years ago

      Of course, any dedicated enthusiast can differentiate that 9 nanosecond difference…. in startup times and game loading times… right…

      • absinthexl
      • 14 years ago

      Any dedicated audiophile can hear the difference between an $800 cable and a $1000 cable, too…

    • Beer Moon
    • 14 years ago

    I’d like to see those benches at 560Mhz.

    • silent ninjah
    • 14 years ago

    Great article, means I wont go insane over latencies next time I buy ram!

    I have 2x1Gb of samsung ram, not too sure what it is exactly as I ordered it from pcspecialist.co.uk and haven’t bothered checking the dimm’s themselves. But the SPD timings are 3-3-3-8 @ 200 mhz according to cpu-z. When bios was set to pick the timings it was the same but running at 2T command rate :/

    I’m running it at 220mhz w/ 2.5-3-3-8 timings and 1T command rate, and things seem perfectly stable. I’m, not damaging the ram am I?

    • gerbilspy
    • 14 years ago

    Beautifully writen article. the best, most succinct, understandable article I’ve ever read explaining memory latencies. Eloquent and pointed. Great job! TR at it’s finest. 🙂

    • mongoosesRawesome
    • 14 years ago

    Perhaps you should try using /[

      • mongoosesRawesome
      • 14 years ago

      no response?

        • Beer Moon
        • 14 years ago

        Dude any memory that runs at those same timings will get the same performance. Only variation between test runs would create a difference, and it wouldn’t be significant.

        Any stick of RAM running a given set of timings at a given Mhz always performs the same regardless of manufacturer. The reason people pay more for good RAM is that the chips used are capable of higher speeds and lower latencies, not because one manufacturer’s RAM is better at storing and transferring data.

          • mongoosesRawesome
          • 14 years ago

          That’s not true. Just take a look at any of anandtech’s memory reviews. They show marked differences between different manufuacturers memory.

          Here is one that handles the very same issue that this TR article covers: §[<http://anandtech.com/memory/showdoc.aspx?i=2392<]§ Even with the exact same timings, memory can give different results.

            • Beer Moon
            • 14 years ago

            Those tests aren’t using the exact same timings for all memories.

            Most of those memories have variations in SPD settings. Each RAM when installed will have different stock SPD settings, and that accounts for the stock speed differences. The higher quality RAM is using faster SPD settings. That is where the performance difference is. Anand only goes out of his way to (try) to set a few SPD settings, Command Rate and tRAS I think he mentions. Otherwise why list the differing SPD settings for each stick on the individual description pages (some is CAS2, some CAS2.5, etc..).

            If Anand used the exact same timings for every stick of RAM in those tests, there wouldn’t be any difference, and there wouldn’t be any reason to run the tests. Some RAM can’t do the better timings and higher Mhz. Out of the box, some stuff is better than others because of SPD settings. When it comes to overclocking, still other stuff might show very high value for the price.

    • Thresher
    • 14 years ago

    Thank you TR.

    I’ve been telling friends for years that high performance RAM makes very little difference for most users.

    I’d really love to see more articles like this that help dispel marketing BS.

      • Krogoth
      • 14 years ago

      I still love my 2x512MB sticks of G.Skill PC4400 which use Samsung TCCD chips. The memory is so darn flexable with timings and bandwidth settings that my only bottleneck is Venice 3200’s memory controller. The best part is that only needs 2.65-2.70V and no ramsinks are necessary.

      I do agree that mainstream users shouldn’t care about “performance” memory. It’s only geared towards overclockers and performance freaks.

        • mikehodges2
        • 14 years ago

        Performance freaks are people too!!

        I love my low latency RAM. Don’t care what anyone says, my computer definatley ‘feels’ slower when i relax from 2-2-2-5.

      • indeego
      • 14 years ago

      I’ve been telling friends for years that high performance /[

    • WebHobbit
    • 14 years ago

    And how does all this apply to DDR2 ram? I notice it tends to have 4s and 5s in it’s CAS specs!

    The two Mushkin 512s I just added to my Intel 925 mobo (2 gigs total now) is 4 4 4 4 12

    • WaltC
    • 14 years ago

    The way I look at it is that synthetic benchmarks are designed to show the differences in performance between discrete system components, and this usually isn’t reflective in so-called “real world” software when the bottleneck in performance for that software is located elsewhere in the system.

    For instance, benchmarks designed to to bottleneck in the system memory subsystems will indeed reflect differences in performance which occur there as the result of faster timings, but “real world” software which happens to bottleneck in, say, the 3d card or the cpu, instead of the memory subsystem, might not show any performance difference in terms of framerate between faster and slower ram timings, since the performance bottleneck for the software does not occur in system memory.

    So my opinion is that in all cases ram with lower timings is indeed faster than ram with higher timings, but the degree of performance increase, if any, will depend on the software one runs. IE, just because some 3d games are cpu-limited and show little if any performance benefit from a newer 3d card, that’s no reason to pass up a new card because there’s plenty of software out there which will benefit because it bottlenecks more in the 3d card than the cpu, etc.

    I think the wrong conclusion to reach in this case, however, is that ram with lower latencies isn’t any faster than ram with higher latencies, because software designed to place the performance bottleneck in the memory subsystem (synthetic benchmarks written to do that, etc.) clearly highlights the performance differences that lower latencies make. So the following points are both true:

    (1) System ram with lower latency always performs faster than system ram with higher latency

    (2) Some software will see a performance gain from lower latencies and some software will not

    • Spotpuff
    • 14 years ago

    This review echoes the findings of countless other sites. Super expensive premium RAM is just not worth it for Athlon 64 systems; hell, even 1T command rate isn’t really worth it. The only thing CAS 2 RAM is good for is benchmarking, and no one I know plays benchmarks.

    I recently went from 1GB 1T DDR 400 to 2GB 2T DDR 400. No difference in games whatsoever.

    Just one more reason to go AMD over intel: RAM savings. My OCZ Premier still has a lifetime warranty.

    y[

      • Anomymous Gerbil
      • 14 years ago

      y[

        • smallstepforman
        • 14 years ago

        I’m a fool with money, but back when I built my current rig, I payed careful attention trying to maximise bang for buck. I settled for slower CPU, which could have more aggressive timiings with equally agressive RAM. I’ve had my set of 5-2-2-2 Corsairs @400 for well over 3 years, and also run an overclocked AthlonXP @ 400 FSB. Even though the CPU is beginning to show its age, the system is extremelly responisve compared to a newer rig at work. Back then when I assembled the box, I tried 3DMark2003 at various settings, and got quite a bit of performance increase running at 5-2-2-2 and a FSB of 400 compared to running the stock 8-3-3-3 @ 233. System is stable as a rock (proper fans and heatsinks).

        In the end, I payed a bit more for memory and fans/heatsink, but this 3 year old system is still screaming ahead, and will probably last me more than any other system I’ve ever had, since I still dont intend to upgrade (not until x86-64 becomes mainstream).

          • Vrock
          • 14 years ago

          I agree that the Athlon XP is still plenty fast. The problem is that AGP is dead. They’re effectively forcing us Socket A guys to upgrade if we want to play our games. What I’d give for a AGP 7800GT! Oh well.

        • Spotpuff
        • 14 years ago

        Yep; fixed

    • Koly
    • 14 years ago

    Nice review, it’s interesting to see that the 2-2-2-5-2T is slower than 2.5-4-4-8-1T in syntetic benchmarks, but is usually faster in real world applications.

    OTOH, I thought that the most “value RAMs” were 3-3-3-8, that’s the JEDEC standard after all. And my guess is that 3-3-3-8 would be most probably faster than 2.5-4-4-8. Another marketing trick maybe? Buy CL2.5 instead CL3, but don’t tell anybody it’s slower?

    I have bought CL2.5 value Corsairs a while ago and they were 2.5-3-3-8 though.

    • absinthexl
    • 14 years ago

    I also like how 2.5CAS 2x1GB packages on Newegg are more expensive than two 1GB sticks bought separately…

      • Vrock
      • 14 years ago

      You’re paying for the “convenience” of having “matched” pairs of RAM “certified for dual channel operation”. Because it’s soooo hard to enter “2” instead of “1” in the shopping cart basket for single sticks. Heh.

    • Crayon Shin Chan
    • 14 years ago

    Back in the days of the Athlon XP, I remember latencies used to be a big deal back then. DDR266 was faster than DDR333 because of the lower latency (2 vs. 2.5)

      • spiritwalker2222
      • 14 years ago

      Crayon, I don’t see how the DDR333 would be slower as the cas latencies in time are the same. Both 7.5ns.

    • FireGryphon
    • 14 years ago

    So let me get this straight. Let’s say a memory module is spec’ed to 2-2-2-5 at 1T. That means that there’s five cycles between the initial ‘activate’ command and the precharge (closing) command. So, you send a row-activate command on the first clock cycle. We wait one clock cycle (up to 2 cycles now). On the third clock cycle we try and fetch data from a cell. The fetch continues for the next clock cycle (we’re up to 4 clock cycles now). On the 5th clock cycle, we send a precharge command that will eventually take two clock cycles, but we’re done with this ‘session’ since we’re on the 5th clock cycle. During the 6th clock cycle we wait for the precharge to complete. Now since this module has a 1T command rate, whenever we want to select the next cell from memory, we have to select the proper chip. So on the seventh clock cycle we select a memory chip, and on the 8th cycle we can begin this process all over again by sending a row-activate command. Is this correct?

    You deserve a round of applause for explaining latencies as you did on the front page. You did so much more clearly than just about any other site I’ve read. Overall, a great article. I still have some questions after reading, though:

    1. How could you not do overclocking tests? You even admitted in the conclusion that overclocking is the ‘ace up the sleeve’ of more expensive memory modules, yet you didn’t do a single overclocking test.

    2. From the article: “Next, the system sends an active command to the desired row.” Did you mean “…sends an ACTIVATE command…”? I’m not sure, since later you wrote, “The active-to-precharge delay refers to the minimum number of cycles that must elapse between an active and precharge command.”

    3. (2) leaves me wondering whether the there’s a special command called an ‘active’ command as opposed to an inactive one.

      • Dissonance
      • 14 years ago

      It’s active, not activate.

      As for overclocking tests, we didn’t include them for several reasons. First, this wasn’t a review of a specific set of low latency modules, we simply set out to determine what kind of performance impact different memory timings had at stock speeds. We also didn’t have a set of value DIMMs to overclock; we were just running our low latency modules at more relaxed timings.

      And then there’s the fact that overclocking success can vary from manufacturer to manufacturer and module to module. Just because we’ve had our 2-2-2-5 memory at 560MHz with 2.5-4-4-8 timins doesn’t mean that all low latency modules can overclock that high, or even that other OCZ Platinum DIMMs will hit those speeds. In fact, our DIMMs will only work at 2.5-4-4-8 and 560MHz on certain motherboards, and it’s that kind of variability that makes it hard to draw many conclusions from overclocking tests conducted with a limited sample size.

        • FireGryphon
        • 14 years ago

        Agreed, but is it not possible to get a general idea of how much overclocking power is in a set of standard DIMMs vs. a set of expensive DIMMs? Or is it entirely possible that a regular set of DIMMs could outclock an expensive set? I suspect not, but would like to be able to bak that up even a small set of numbers.

        • dragmor
        • 14 years ago

        Doesnt the best tRAS change from chipset to chipset?

        I know I get higher bandwidth numbers from running 2-2-2-8 on a NF3gb board than I do from 2-2-2-5.

        • Anomymous Gerbil
        • 14 years ago

        Without trying to be antagonistic, those reasons for not doing overclocking tests aren’t valid. If you had just said “we didn’t have time, and wanted to get the article out”, then *[

    • eon_blue
    • 14 years ago

    too bad they didnt do any tests on the o/c’d ram running at higher frequencies, that would have made it a bit more interesting.

    • DorXtar
    • 14 years ago

    Thanks for that!! I feel better for not buying the premium stuff! This is what keeps me coming back to TR. 🙂

      • Foolio
      • 14 years ago

      Too bad I can’t say the same thing…

      Half a year ago, I bought a pair of Rev 2 sticks however they would only run at 2-3-2-5 (if I manually set it to 2-2-2-5, BSODs all over the place). Increased voltage didn’t seem to help at all. So I have lived with it….until now. This article has me thinking: “If I have paid this much for the sticks and I’m not even getting their rated performance out of it, should I get replacements?”. Is it worth my time sending them back in to get new ones? Somebody tell me what to think!

        • FireGryphon
        • 14 years ago

        Is your system overclocked in any way? The conclusion of the article states that an overclocked system can benefit from heartier RAM, even if those benefits don’t show up outright in tests. If you’re not overclocking, it might be worth your while to see what warranty you have on the RAM and try to RMA them for a better-functioning set, or exchange them for cheaper modules altogether. I’d say it all depends on how important the money is to you and how much you’re willing to deal with whoever it is that issues the replacements/refunds.

        Anyway, at least you know that you don’t need to spend that much on future RAM.

Pin It on Pinterest

Share This