review llano in overtime 1866mhz ram igp only power

Llano in overtime: 1866MHz RAM & IGP-only power

Between our reviews of the mobile and desktop versions of the chip, we’ve spent a tremendous amount of time recently with AMD’s new Llano APU. That doesn’t mean we haven’t had some lingering questions about it, though. We always have more questions about new technologies than we have time to devote to them, seems like.

One of our remaining questions about Llano has to do with its memory support. Uniquely, the desktop versions of Llano officially support DDR3 memory at clock speeds of 1600 and 1866MHz, higher frequencies than the 1333MHz memory currently widely available on the market, with the potential for higher performance.

Now, that official endorsement from AMD comes along with some big, honkin’ caveats. Among them: those speeds are only supported “in a single DIMM-per-channel configuration.” So you’ll have to forgo half of the DIMM slots on most desktop Llano motherboards in order to use those higher speeds. Also, only “true 1.5V, JDEC-spec memory” gets official support.

In other words, AMD’s stamp of approval for the use of higher-frequency RAM with Llano is essentially forward-looking. In preparation for our review, we pinged the memory mavens at Corsair about the availability of JEDEC-approved DRAMs rated for these new frequencies. We were told JEDEC has indeed ratified the spec for those parts. However, Corsair couldn’t give us any firm date for the market availability of such DRAMs. All we know at present is that they’re on the way. AMD’s advice to reviewers concurs with this estimate, stating, “that kind of memory will likely not be widely available until a later time” and conceding that most reviewers “will likely end up testing with 1333MHz DDR3 appropriate for the price tag of this platform.”

That’s not to say one can’t find DIMMs on the market, made with current DRAMs, that will operate at 1600 or 1866MHz. Firms like Corsair perform the invaluable service of sorting and binning DRAMs and packaging up the best chips into enthusiast-class modules. Such modules can sometimes perform amazing feats that would make a JEDEC committee gasp in unison, drop the monocles from their eye holes, and whisper-talk in exasperated tones about how scandalized they are. In fact, among enthusiast brands, DIMMs rated for 1600MHz are fairly common these days and don’t carry much of a price premium over 1333MHz modules. Many of them will even operate at 1.5V. One can find modules rated for 1866MHz, as well, although they tend to come with a fairly substantial price premium attached.

The complexities of this situation forced us to make some choices when formulating our review of the desktop Llano APUs. We took a multi-pronged approach. We started by testing with 1333MHz memory in our CPU benchmark suite, since that’s the type of memory most likely to be used in value-focused systems based on this APU. Also, the architecture of Llano’s CPU cores is largely familiar, and given what we knew about it, we didn’t expect major performance gains from incremental improvements in memory bandwidth. We added 1600MHz memory to the mix in our integrated graphics testing, because that’s where we expected additional memory bandwidth to make the biggest impact. We also requested some 1866MHz-capable DIMMs from Corsair, but those unfortunately didn’t arrive in time for us to test them and include the results in our initial review.

For reasons science is only beginning to understand, the omissions of higher-speed DIMMs in the CPU suite and of 1866MHz memory in the graphics tests led to a minor outcry from some quarters. Little did the complainers know that deep in the underground expanse of Damage Labs, we were taking the wraps off of these puppies:

These two Corsair Vengeance DIMMs are 4GB each and are rated for operation at a clock speed of 1866MHz at only 1.5V with 9-10-9-27 timings. That’s pretty sweet, but at $94 for a pair, they’ll set you back about 30 bucks more than the cheapest 1600MHz modules at Newegg.

We dropped these DIMMs into our A8-3850 APU-based test system and configured them to use Corsair’s recommended timings. That’s a bit of a step back from the 8-8-8-20 timings we’d used before with 1333MHz and 1600MHz DIMMs, but with two crucial differences: these modules are running at 1866MHz, and because we’ve dropped from four DIMMs to two, we’re able to lower the DRAM command rate from 2T to 1T. The end result should be a higher bandwidth, lower latency memory subsystem than anything we’ve tested with Llano to date.

The question is: what difference does it make? To find out, we ran a broad selection of benchmarks from our CPU and IGP test suites, which you’ll find in the following pages. If you want more info on our test configurations and methods, be sure to see our original A8-3850 review for that.

Memory subsystem performance

A couple of quick synthetic memory benchmarks confirm our 1866MHz memory config is indeed both higher bandwidth and lower latency than our default 1333MHz setup. The A8-3850 with 1866MHz RAM manages more throughput in Stream than any flavor of Phenom II, and its memory latency is comparable.

However, these results also serve as a reminder of how proficient Intel’s newer CPU architectures are at wringing as much as possible from their attached memory. The Core i3-2100 manages 5GB/s more throughput with 1333MHz DIMMs—at a 2T command rate, no less—than the A8-3850 does with 1866MHz DIMMs with a 1T command rate. The Intel processors hide access latencies better, too.

These differences are, in part, a result of Intel’s use of larger caches with smart data pre-fetching algorithms that appear to make good guesses about what data our tests will request next. Llano adds a somewhat larger 1MB L2 cache per core and a tweaked pre-fetch algorithm, but it has also shed its L3 cache. In the grand scheme, the changes don’t appear to add up to much.


Among our productivity tests, 7-Zip’s decompression routine benefits the most from the faster memory config, but even there, the gains aren’t terribly impressive.

Video encoding

Video encoding is perhaps the one consumer application most widely considered to be limited by memory throughput, yet our trio of tests reflects only modest gains from 1866MHz DIMMs.

3D modeling and rendering

On the other hand, 3D rendering traditionally hasn’t been sensitive at all to the speed of the memory subsystem, and it proves true to form here. The needle barely moves.

Source engine particle simulation

Scientific computing

More so than almost any other tests, our two scientific computing benchmarks have long been willing to take advantage of additional memory bandwidth to good effect, and they do show some modest improvements for Llano with 1866MHz RAM. Still, even here, the differences are fairly small.

Integrated graphics performance

Turns out Llano’s IGP doesn’t extract much more performance from 1866MHz memory, either. The most notable improvement with faster RAM is probably in Civ V‘s Leader benchmark, which works the pixel shaders pretty hard. Even there, 1600MHz memory captures most of the benefit to be had.

Power consumption and efficiency
Another question we’ve wanted to see answered about Llano is how its power consumption looks without a discrete GPU in the picture. We actually did the measurements for the A8-3850 and the Core i3-2100 prior to the publication of our initial review, but we ran out of time and weren’t able to include the results. Now we can remedy that oversight.

Note that the vast majority of the results below show total system power consumption using as similar a slate of components as possible, though the systems obviously varied by motherboard type as necessary. Just four of the result sets omit our standard GeForce GTX 460 video card—the two marked “(IGP)” and the two marked “(Brick PSU).” Additionally, the “Brick PSU” results involve low-power processors—the Atom and the E-350 APU—using a more efficient laptop-style brick power supply. We pondered putting the Llano and Core i3-2100 systems on the brick PSU, as well, but Llano’s 100W TDP probably wouldn’t mix well with that power supply’s 80W peak rating, so we stuck with our standard PSU instead. Out of necessity and because it’s only fair, we did install the Core i3-2100 in a smaller microATX motherboard, the Intel DH67BL, for the IGP power tests.

This is a pretty comprehensive win for Intel—not terribly surprising given that the Core i3-2100 has a 65W power peak. The transition to the DH67BL mobo allows the i3-2100 to achieve lower power draw at idle than the A8-3850, which is a bit of an upset. Even without a graphics card attached, the A8-3850 draws more power under load than the Core i3-2100 does with a GTX 460 installed.

Removing the graphics cards reduces the amount of energy the A8-3850 requires to complete our rendering task, but the efficiency gain isn’t terribly dramatic in the grand scheme of things.

We’re pleased Llano is capable of supporting the latest in DDR3 memory standards, and we’re happy to be able to demonstrate some performance gains when the APU is used in conjunction with faster RAM. However, I think we can say with some certainty that Llano’s overall competitive picture doesn’t change substantially when higher speed memory is added to the mix.

The CPU cores on this chip are based on, uh, a proven-technology microarchitecture, and that architecture doesn’t appear to extract much additional instruction throughput from a higher-performance memory subsystem. One might expect Llano, with no L3 cache, to benefit more than most processors from faster RAM, but these days, it seems CPUs with larger caches and aggressive pre-fetching make the best use of additional memory bandwidth when it becomes available.

More unexpectedly, even Llano’s IGP doesn’t churn out substantially higher frame rates in the presence of 1866MHz memory. At its relatively low core clock speed of 600MHz, the Sumo IGP appears to be getting most of what it needs from 1333MHz or 1600MHz memory.

Despite these results, we do expect memory bandwidth to become ever more critical as CPUs with integrated graphics—or APUs, if you must—mature. Recall that AMD has done little in the way of true architectural integration with Llano, instead opting to focus on reducing power consumption (especially at idle) and getting a product to market quickly. AMD’s next APU in this segment, code-named Trinity, should combine the brand-new Bulldozer microarchitecture with a Cayman-derived, second-generation DX11 Radeon IGP. That chip might also feature Sandy Bridge-esque architectural integration, with features like a high-speed internal bus and a shared last-level cache. Such changes should make this next APU much hungrier for memory bandwidth—and much more capable of taking advantage of it—than this first attempt.

0 responses to “Llano in overtime: 1866MHz RAM & IGP-only power

  1. id like to see some undervolting testing of the llano bunch…ive been hearing rumors its stable around 1.2-1.25v at stock speeds.

    could turn out to be a good htpc when undervolted.

    great article as always…will be interesting to see how 65w llano parts fair.

    and cant wait for trinity 🙂 seems like Amd will be doing a refresh rather quickly

    edit: found article from NordicHardware….


  2. Then that would be “something wrong”, because now you’re saying that “well you can pick higher speeds in the BIOS but it can’t really handle it”. I’d hardly call that efficient.

  3. My bad, sorry, I was for some reason convinced regular Llano was 65W and low-power 35W… I need to move all that up by 30.

    For nettops, I’m being cautious about heat, and constrained by power. For example, the cheaper nettop case is: [url<];[/url<] it includes 80W picoPSU, 60W powerbrick, and VESA mount. Notice the 60W power brick, which means Intel T-CPUs = OK, desktop Llano = not OK, so I need to upgrade at least the brick, possibly the picoPSU (gotta have some spare power for a few active USB ports, maybe an external HD or DVD...). And then I need a quicker fan to get rid of the extra heat (bigger won't fit, so it's got to be quicker = noisier). Apparently LLano can be undervolted by a lot, but I'm leery of tweaking PCs that I build for others. Hopefully OEMs will release mini-itx boards with laptop Llanos then, like they did with Brazos.

  4. I’m gonna say one thing and let it go. First CPU bound software has shown very little if any difference with RAM speeds. So there’s really no surprise there. DDR2/DDR3 has been more than enough for CPU bound tasks for ages– triple-channel i7’s have already proved this.

    That being said, testing most games at 4XAA almost (depending on game and type of AA) could unintentionally remove bandwidth out of the equation. Especially for any IGP that’s naturally castrated from discrete. Why wasn’t more than one AA mode shown here? This is normally done so that the reader can see not only what bar is higher than another, but to weigh workload / setting with the performance given. By showing just one AA mode, you’ve basically reduced the data points in precisely the area where a good GPU would matter the most. No one can tell what setting bandwidth matters because you’ve only showed one setting for the GPU tests. In previous reviews you have said that you’ve included more data points because the more data provided the better. What changed here?

  5. You can’t deny that over some GPU performance, we will go SOC way. Just have a look to this console cycle. It’s taking more than 7 years just because the GPU’s were “good enough” to have good graphics and CPU’s strong enough for almost everything the required.

    Next generation it’s required to have fullHD 1080p 3D games… but later…? We might be looking at a new generation of game systems that may have 10-15 years of lifespan… im not kidding. If that happens, you will see PC’s go SoC way. You won’t get higher performance in games by going discrete almost. Most of games will be targeting a console that it’s 10 years old and no one avoids SoC’s in desktops to use 200W and duplicate discrate level performance pairing the motherboard with 1Gb or 2GB of GDDR-N.

    It’s also likely that GPGPU will take off at some point. That means that owning stronger GPU’s inside the x86 SoC’s becomes important not just for gamers. I can see different SoC’s targeting different segments. Mainstream, Hardcore and Economy. Llano has targeted the Economy segment now. Look forward for Mainstream next year.

  6. [quote<]Remember, according to AMD TDP should be the theoretical max the part can draw, not what it normally draws under load.[/quote<] Actually, according to AMD, TDP is not the theoretical max the part can draw. "TDP is not the maximum power of the processor." page 61 -- [url<][/url<]

  7. I think Llano is just the latest way to build a budget box. If you consider history a bit, it’s not even that innovative. For example, nForce 1/2 were equally hailed as bringing very competent gaming to the IGP world. You can say the same for most of the ATI/NV IGPs. They could usually power contemporaneous games at decent resolutions and quality levels. People got excited by each one for a little while.

    With 780G and its offshoots we reached a point where the HT bus was a major bottleneck for the IGP getting access to RAM. In order to move performance forward, AMD needed to move the IGP closer to the CPU’s memory controller. That’s what Llano is.

    It is more of a SoC than usual but that isn’t unprecedented either. It’s similar to the old Cyrix MediaGX and some of the AMD Geode chips in many ways.

  8. Those are the 100W parts, the 65W parts are the A8-3800 and the A6-3600. Somewhat slower clocks but Turbo Core enabled (it was disabled for the 100W launch parts). They should be out soon; some sites have already tested them.

    AMD might launch some lower-TDP desktop parts later (45W? maybe dual-cores at 35W?), but I don’t think they’ll be putting out desktop quad-core Llano parts with 35W TDP. See the page where TR compared the A8-3500M to desktop processors for why- it’s just too much of a performance hit.

    The mobile parts have their own socket (FS1 instead of FM1). So unless you find motherboard manufacturers doing MoDT (unlikely as the mobile parts don’t really have a compelling advantage) there’s no use trying to get your hands on a mobile processor.

    I don’t see why you think the 65W TDP chips can’t compete with Intel’s parts. Power-wise they’ll be just fine for mini-ITX: [url<][/url<]. Remember, according to AMD TDP should be the theoretical max the part can draw, not what it normally draws under load.

  9. [url<][/url<] I can buy both the [email protected] and the [email protected] and get them tomorrow: they're in stock (the green bar in the "Dispo" column, "Dispo" = "Disponibilité" = Availability, green is good, as is greed, though soilent green is people !). Maybe Europe is getting them earlier ? I didn't know the 35W parts were specifically for notebooks. Hopefully they'll be available at retail, and be supported in upcoming FM1 mini-ITX boards. If not, AMD are leaving the door wide open to Intel's T-series (pentium G620T and Core i3 2100T), which are available at retail, and supported by all socket 1155 boards, at least the mini-ITX ones ! Brazos just can't compete against those (I know, I built both, and am getting ready to murder&rob the friend who got an i3T instead of my Brazos) ... back in 5....

  10. Llano’s Ram-Speed related gains seem to depend very much on the workload. No doubt, games can potentially gain more because the GPU benefits more from the higher bandwidth (reserve). Encoding for example shows absolutely no gain, within 1%. I also tested that.

    So I did some gaming/graphics tests on Linux:
    First some open source games, mostly non demanding GFX:
    [url<][/url<] Second is the same run, normalized against the 1333 MHz run: [url<][/url<] That's up to 31% gain, going down to just 8% for the result with the least gain. Second three Unigine Benchmarks: [url<][/url<] Normaized against the 1333-run: [url<][/url<] As good as 13%, 10% and as low as 6% are the gains here. Funny about these results seems to be that the more GFX intensive benches, and the Unigine stuff ist the best I've ever seen on Linux, seems to gain [i<]less[/i<] from the faster ram, but the reason might also be that they also utilize the CPU a lot more than the rather light open source games. I don't have much knowledge about how demanding each of these tests is for the CPU- nad GPU-part respectiely, so maybe somebody could shed some more light. System details can be found on the benchmark pages, Ram was G.Skill RipJaws-X DIMM Kit 8GB PC3-14900U CL9-10-9-28 (F3-14900CL9D-8GBXL) at standard timings, just using the mainboard's mechanism to set them to 1333 an 1866 operation respectively.

  11. The point of this system is low price. I can build an entire A8-3850 box that will play games quite well at lesser resolutions for less than many i7 processors.

    This is the point of the FM1 socket systems.

  12. Bulldozer should be the game changer for Hardcore people like you, and Llano is a good game changer for low-requirements people. We must wait for Bulldozer benchmarks to figerup if AMD will remain a good enough CPU to keep competing by price, recover some CPU market and status getting closer to intel, or totaly fail to keep the track.

    For LLano, most of people I know would benefit much more from 2-3 times faster GPU than 50% faster CPU. So I will strongly recommend to avoid intel for people wanting to get a desktop or low-budget laptops.

  13. Yes, the CPU side doesn’t benefit… so if llano was just a CPU the current controller is sufficient.
    But we are not talking just about a quad core version of the 2003 athlon, this controller also feeds a decent GPU.
    From what we see, if AMD had a better controller it would get PC2100 performance out of PC1333.

    What I see is that AMD is falling farther and falling behind Intel.
    Its a concern because 2011 was supposed to be a come back and so far AMD is doing the opposite.
    Well Intel is not help by delivering state of the art design and manufacturing…

    AMD should just sell its CPU division and focus on what its good at GPU… And maybe change its name back to ATI 🙂

  14. Hardcore gamers won’t buy Llano… no one will ever try to deny this fact with using common sense. I understand it’s not a problem for you to write an article that I didn’t expect to… I just pointed that most of people IMHO was expecting more in-deep analysis of Llano GPU. Techreport it’s my computer-tech related news place, hope you understand you don’t need to argue that you are doing things trying to be coherently, I already know, that’s why I read this website everyday. But even trying to argue and be coherent, sometimes you may miss a point.

    Anyway, I’m glad we are starting to have a discussion about Llano usage model and what we should be looking forward in Llano reviews.

    Llano from my point of view should be targeted as light gaming machines and budget tight systems with Low gaming requirements (people that wants to play, but don’t really care the level of details and are up to tweak the details to get playability at the lowest cost possible) or budget OpenCL systems. The later should be developers more than customers right now looking forward to answer themselves: “What can I do with a low-profile GPU integrated into a CPU”, and targeting Llano as the base system for OpenCL testing.

    Some of us are looking at Llano instead of Core i3 also because we believe that GPU will gain applications over time. Windows8 points to have an store with HTML5/Javascript applications. With HTML5 being GPU accelerated and targeted to tablets also, we might see Windows8 apps like performing much better and the UI being much smoother and snappier than with Intel systems that rely solely in the IGP.

    And the last point, I expect AMD Llano line with trinity or further incarnations to start to go upper in the traditional GPU segments. Now we are talking to require at least 70$ to get a benefit. Next year we may start to talk at 100$ or 150$, that’s why I believe it’s worth to start to develop Llano benchmarks as any other GPU and not like a CPU with IGP.

    PD: What we are looking at with Llano is an x86 SoC. It’s just like smartphones and I strongly believe that the desktop market will go into the SoC way just as Smartphones. Dedicated GPU’s will be depreciated overtime and we will need to be benchmarking CPU and GPU for different SoC’s from AMD, Intel and maybe NVIDIA, Qualcomm, Samsung, etc. It might happen if from now in the future, Windows8 game developers and companies target both, ARM and x86 architecture with their software releases. Dedicated graphic cards benchmarks will banish and you should look forward for SoC benchmarks style, fusing your traditional analysis of CPU’s and GPU’s.

    Edit: PD added.

  15. I think we’ve already established that you wanted to see a different article than the one I wanted to produce. That’s really not a problem for me. Just wanted you to understand that we did things a certain way for a reason and that our results are sensible and rationally coherent, even if they don’t meet your goals.

    To take things a step further, I’m pretty sure disagree with you about the likely usage model for a $135 desktop CPU with a 100W TDP. I don’t think such a CPU is likely to spend its life running games at the lowest possible quality settings at 1024×768 or such–at least, not here in the U.S. I expect such a CPU to end up in a desktop case, maybe a mid- or mini-tower, with a discrete monitor attached. The most common resolution these days for such a display is probably 1440×900. As I stated in the original review, I wanted to test at that resolution but was thwarted when the games and drivers involved didn’t offer that res as a choice. We picked 1680×1050 as a backup, with a few more pixels in each dimension and (for what it’s worth) a similarly-shaped viewport.

    You’re welcome to show me all of these 4:3 and 5:4 displays on modern desktops that hard-core gamers call home, but until that happens, I’ll continue to disagree about who has the Llano usage model wrong. 😉

  16. Damage, despite what you may think, I’m not saying that this review is manipulated or god knows what. I’m pointing that you are missing to focus the review around the important point: IGP performance. A detailed analysis of IGP performance and memory speed it’s what most of people wanted when we claimed for 1866 testings.

    I like you point that ROP’s may be what it’s taking down the peformance and adding AA may hit any improvement you can get from higher memory speeds, now we are discussing and you are starting to have a deeper analysis of the Llano IGP and memory speed, just what was expected from this review.

  17. It’s true that the difference between what your review is showing and the results others have gotten aren’t huge. But for a lot of games and a lot of common detail/resolution settings, Llano is right on the border of playability, and 20% could make the difference between playable and unplayable. It’s not worth paying a huge premium for faster RAM, but it does reintroduce memory speed as a decision variable in the cost allocation problem of building a cheap desktop again after a couple of years of total irrelevance (very very few things have benefited from >DDR3-1333 speeds in previous tests).

    Cheap is relative; you’re used to getting review samples of $500 GPUs sent your way but that’s not how it works for everybody. The cheapest discrete card which will be a major upgrade over Llano is the 5670, which goes for about $70. For a serious gamer building their primary machine, getting a discrete card is still a no-brainer, and if you’re spending that $70 you may as well spend another ~$30 and get a 5770/6770 and much better performance. But at the end of the market desktop Llano is aimed at (<$500 systems), $70 is a huge part of a system’s total cost.

  18. Two responses:

    1) The system setups and settings we used for IGP testing are not “unknown.” They’re very clearly documented in our desktop Llano review and referenced in this article. Start here:

    [url<][/url<] 2) Versus your cited examples, we tested at relatively higher quality levels, with AA where possible, largely at 1680x1050, using an entirely different set of games. To the extent that we can hazard a guess about the likely sources of scaling differences, I'd say the information we have supports my theory about the Llano IGP being more ROP-bound than shader-bound--thus, it scales less well when drawing more pixels with more edge AA. And thus, it scales less well under our test conditions than at 1024x768 or what have you. What I don't see in your examples: direct contradictions of any of our test results using similar setups, games, and settings.

  19. Could be that going from 1333 to 1600 unleashes most of the improvements you can get from higher frquency memory…. However:

    “More unexpectedly, even Llano’s IGP doesn’t churn out substantially higher frame rates in the presence of 1866MHz memory. At its relatively low core clock speed of 600MHz, the Sumo IGP appears to be getting most of what it needs from 1333MHz or 1600MHz memory.”

    “Turns out Llano’s IGP doesn’t extract much more performance from 1866MHz memory, either. The most notable improvement with faster RAM is probably in Civ V’s Leader benchmark, which works the pixel shaders pretty hard. Even there, 1600MHz memory captures most of the benefit to be had.”

    Techreport points that 1600 to 1866 seems less exciting, but they are failing to show the improvement itself.

  20. What I see is 1333 likely being a tangible IGP bottleneck compared to 1866. However, going from 1600 to 1866 is generally less exciting, as should be expected. I think TR shows these aspects in their results.

    I thought it would be interesting to see AMD come up with a GDDR5 SODIMM spec. But I believe that GDDR5 is considerably more expensive and thus it’s unrealistic for Llano’s intended market segment.

  21. The one everybody’s been focusing on is [url<][/url<]. There are plenty of others, here's a few:[list<][*<] [url<][/url<][/*<][*<][url<][/url<][/*<][*<][url<][/url<][/*<][/list<]

  22. Going from 1333 to 1866:

    [url<][/url<] Resident Evil 5, from 50.9/28.7 (1280x720/1920x1080) to 62.3/35.2 -> ~ 22.5% improvement both resolutions [url<][/url<] STALKER Call of Pripyat, from 37.2/23.6 to (1280x720/1920x1080) to 45.0/27.9 -> ~21%/18% improvement respectively [url<][/url<] Crysis: Warhead (1280x1024 performance quality) from 63.5 to 79.6 -> 25.35% improvement Crysis: Warhead (1024x768 mainstream quality) from 38.4 to 46.2 -> 20.3% improvement Metro 2033 (1280x1024 DX9 Low Quality 0xAA/4xAF) from 23 to 29.5 -> 28.26% improvement Metro 2033 (1024x768 DX9 Low Quality 0xAA/4xAF) from 31 to 40 -> 29% improvement Dirt 2 (1280x1024, DX9 Medium quality) from 42.4 to 50.6 -> 19.33% improvement Dirt 2 (1024x768, DX9 Medium quality) from 53.6 to 65.6 -> 22.23% improvement Mass Effect 2 (1280x1024 Minimum quality) from 36.2 to 45.4 -> 25.4% improvement Mass Effect 2 (1024x768 Medium quality) from 51.9 to 60 -> 15.6% improvement Is it enough references? Top 3 Maximum improvement found in techreport IGP analysis.... Civilization V Leader benchmark (Unknown Settings) from 33 to 40 ->21.2% improvement Starcraft II (Unknown Settings) from 26 to 31 -> 19.2% improvement Borderlands (Unknown Settings) from 18 to 21 -> 16.7% improvement Conclusions from Scott: "Turns out Llano's IGP doesn't extract much more performance from 1866MHz memory, either." Do you need anymore references or pointing to understand why I think most of people that are looking at Llano and likes Llano would like techreport to take it seriously and realize a review just like this one: [url<][/url<] but of course, not using the same kind of in-game details or setup. The contenders should be low-middle range graphic cards and intel IGP's, and should include the 1866, 1600, 1333 memory speeds. I hope scott understand the reasons at some point. This is my site of choice because I believe the editors usually are unbiased, detailed and have a pretty good knowledge about what they are testing. That's why I feel disappointed they seem to not get the point with Llano.

  23. Wait– are you claiming the baseline Llano system whose memory frequency isn’t noted on all these charts is DDR3-1600? If so, it really would make sense that the synthetic memory test shows gains of less than 20% and the games show gains of around 10%. I think most of us had assumed the baseline was DDR3-1333.

  24. [quote<] it begs the question why AMD even bothers to support newer RAM. [/quote<] Because newer RAM will be the cheaper RAM soon enough, and older RAM will be more expensive. AMD has to keep up on the DDR technology treadmill, because the treadmill is being spun for the industry's benefit, whether AMD can extract any performance out of it or not. If they don't keep up, they'll quickly be forcing their customers into buying more expensive, obsolescent RAM; for a company with platforms stuck in the "value" niche, that's suicide.

  25. Damage, it’s not my pre-conceived notions, it’s what other websites found. There is a difference between your results and other websites found. I’m pretty sure you didn’t do this review just because my “personal” complaining, but most of people is still wondering why you didn’t found the same results that everyone else did.

  26. You went from DDR3 1600 to DDR3 1866 – it’d be silly to think that there’d be a big difference.

    Why not go from DDR3 1066 or DDR3 1333 to DDR3 1866? That’d be a far better comparison.

  27. Llano DOES support 1600 with 4 dimms: all the motherboard manuals only have footnotes about 1866 ram, not 1600. (I read at least one from just about every vendor tying to find the exact setup I want for htpc)

    It doesn’t *officially* support 1866 with 2 per channel, but that doesn’t mean it won’t work. A couple sites already benched that way.

    Anyways back in regular land: the price difference between 1333 and 1600 really isn’t there if you are a DIY builder. There is no excuse to not get 1600.

    random tidbit about htpc use I thought was very interesting:


  28. Llano: Worthy as a very nice HTPC rig. Any sort of middlin’ gaming machine, you are correct!

  29. “Sure, faster RAM almost always mean higher latency.”

    Not really. The timings may have to be bumped up slightly (but also might not) for, say, a 266-300 MHz speed increase at the same voltage, but the overall latency has the potential to drop. At worst, it’s usually about the same.

    “…if AMD can’t extract more performance out of faster RAM, it begs the question why AMD even bothers to support newer RAM.”

    Because anything but Bobcat or Atom originates from server R&D, and servers need the most up to date RAM possible. The work is already done long before they ever consider bringing these things to PCs.

    “The L3 cache found in Phenom II wasn’t really something to write home about. According to Anandtech’s CPU benchmark tool, a same-clock Athlon II and Phenom II are sometimes neck-and-neck, suggesting the L3 cache doesn’t help much.”

    Same deal as any CPU with cache size variations [b<]for PCs[/b<]. Even without removing a cache stage and drastically altering latency and RAM access, the numerous L2 cache variations in Core 2s showed very little difference at the same clock speeds - except in games, which seem to need to keep more information on hand to keep minimum FPS up. That allows both Intel and AMD to kill two birds with one stone, and reuse spare server chips, which are going to be mass produced with large caches, anyways, for the small and ever-shrinking high end PC CPU market. This probably saves them more money than it costs them, contrary to what "enthusiast" dingdongs seem to believe. "Increasing the amount of faster L2 sounds like a better idea." And then you increase the L2 latency, but in Llano's case, cut the total cache size in half, and it's not so much faster anymore. You can't have your cake and eat it, too. That was done as a compromise so that the ginormous amount of transistors in the cache could be power gated in laptops. Going forward, they'll probably all need a shared L3 cache stage (though not necessarily a large one) just because of the increasing complexity of even mainstream PC CPUs. Sandy Bridge's ring bus and L3 connecting to the GPU is just a glimpse of what's to come at the SoC level.

  30. Believe it or not, this article wasn’t produced just because you personally complained, which is probably why it doesn’t meet your goals.

    You’ve made it pretty clear you’ll be pretty “disappointed” unless we can contrive a scenario in which your pre-conceived notions about Llano IGP performance and RAM speed are confirmed. With respect, we’d be letting a down a lot more readers if we went out of our way to contrive such results artificially and then pronounced Llano somehow better than we first thought–or now think–that it is.

    If you simply want more detailed Llano IGP results, we’ve already done that work with the mobile part. Start reading here:


  31. In an article coming from the outcry of your readers to properly define the IGP improvements coming from using 1866Mhz memory, don’t you think would have more sense to drop all CPU related benchmarks(productivity, video encoding and scientifical calculation) and focus yourself to figerup why there is an improvement in GAMES, and look forward for the settings that can be raised and where we hit a wall that doesn’t allow us to get higher performance instead of doing some random testings and claiming that the outcry should be done with it?

    I’m not sure what you think, but what we all were expecting is a banch of graphic test, just like you would test any new DEDICATED GPU coming to the market, with different resolution and detail levels for each memory speed.

    I know I’m being a bit harsh with this topic with all of you in all your posts about it, but really, I think that you are miss-rating Llano and miss-identifying the use-case for such systems and consecuently,not paying enough atention to the benchmarking of the IGPU. We want to exactly know how good can be a system that exclusively relies in the IGPU of Llano for light or moderate gaming needs. How much resolution hits the peformance, how much AA, how much 3D modeling and texture details, etc.

    Please, I apology to you if you think I’m downrating your articles and reviews, but you are still missing the point about Llano, but I think it’s for the best to keep pointing you what I think most of people interested in Llano want to know.

    Take Llano and do your usuall long and detailed GPU reviews! That’s what we really want! No one is caring about CPU performance when we asked you for 1866 results!

  32. They adopt new technologies for reasons other than just speed. Power usage, for example.

  33. I’ve always expected that Llano will benefit from faster RAM because CPU and GPU contend for bandwidth, but it seems AMD historically doesn’t benefit from faster RAM. It happened with the switch to DDR2, and more recently, DDR3. Sure, faster RAM almost always mean higher latency, but if AMD can’t extract more performance out of faster RAM, it begs the question why AMD even bothers to support newer RAM. Will faster AMD CPUs start to slow down with DDR2 instead of DDR3? Someone should try to find that out.

    The L3 cache found in Phenom II wasn’t really something to write home about. According to Anandtech’s CPU benchmark tool, a same-clock Athlon II and Phenom II are sometimes neck-and-neck, suggesting the L3 cache doesn’t help much. Increasing the amount of faster L2 sounds like a better idea. I imagine losing the L3 isn’t such a big deal. I’d trade my Phenom II’s L3 cache for about 200MHz more.

    Here: [url<][/url<]

  34. It was self-delusional to think this was going to make any meaningful difference.

    As for those “edge of playability” arguments – you’d be better off spending less on the CPU and RAM and getting a “real” GPU.

  35. The correct timings were forced in the BIOS in our test system. The BIOS menu reads 1866MHz, the POST screen reports DDR3 1866, and CPU-Z (new version with Llano support) reports 933MHz (1866MHz) DRAM frequency.

    One possible source of a difference between our IGP results and others could be the resolution and anti-aliasing settings used. We tested mostly at 1680×1050, oftentimes with 4X AA enabled. Those settings will stress the IGP’s ROP throughput. If you tested at lower resolutions with no AA, you might find you were more purely shader-bound. As the Civ V “Leader” test shows, this IGP’s pixel shaders have a little more room left in them. Just a theory, but that could account for any difference with other sites, but it’s hard to say.

    Bottom line is that we’re really splitting hairs here, talking about a small percentage of performance difference in a small amount of total performance. We could spend a lot of time sussing out the differences more finely, but all we’d be proving is a slight imbalance in the IGP’s resource allocation–and we’ve known this generation of Radeons is somewhat shader-heavy for ages. Plus, it’s clear a cheap discrete GPU would easily be superior, which is what folks really need to know, regardless of their display resolution and the shader complexity of their current games.

  36. The 35W chips are mobile parts. What comes next for desktop Llano is the release of 65W parts, which aren’t out there yet, AFAIK.

  37. @Scoot Wasson & Crew, Can you increase the NB speed on these units? Memory speed barely makes any impact on most apps, and only mild improvements on Grfx. But in my experience the biggest improvmesnts you can see on a AMD system are

    Core speed
    NB speed
    FSB speed

    it may fall out of the “oem spec” testing methods, but bumping the NB from 2000mhz to 2500, 2700, 2900mhz can yield some serious very measurable benefits…

  38. If the bandwidth isn’t high enough, then you would see a fairly linear increase in FPS. There’s a problem with doling out memory bandwidth to the IGP, I think.

  39. If you’re looking to play any games at all, it’s worth spending those few extra bucks on a dedicated video card. It was true before Llano and it’s true after.

  40. By the by, ssk, just wanted to make sure you read my late reply to your Friday tirade about exchange rates and tech prices

  41. In other news:
    *CPUs don’t overclock any more, Turbo takes care of that for us
    *RAM timings are set by SPD
    *sk 1366 processors use DDR3-1333 memory (at best) since this is what Intel decided to validate
    *JDEC hasn’t approved any DDR3 speed bin faster than 1333, 1600+ memory isn’t safe to run

    /end sarcasm

  42. Brutal, to be honest. An extra 50% bandwidth and almost nothing to show for it. I figured it might help in games. How does a Redwood-based card do in these benchmarks? If a 5570 or 5670 is super-duper faster (the 6670 has more shaders and higher speeds, so of course it’d be faster on its own) then my guess is there’s something wrong with how memory is handled.

  43. I suppose this means the GPU is so slow it can’t even utilize faster memory. I don’t think I would’ve expected much more from a integrated GPU…

    Of course this is putting aside all the levels of wrong of spending $30 more a pop on memory and discarding extra memory slots to pimp out your integrated GPU.

  44. The price premium is dropping quickly, and 1600 is fast becoming the most common speed. I imagine I wasn’t the only one who hopped on the 8GB of 1866 for $70 incl. shipping deal at Newegg last week.

    It’s unsurprising that faster RAM does little for Llano in non-gaming tasks. But other sites, several of which I trust just as much as TR, have consistently come up with 20% gains in games from moving to 1866; your results are less than half that, and that smells fishy. It could just be vagaries of which games and what settings people are looking at, but this is still well beyond just the “measurement noise” level.

    Are you totally sure your sticks were running in 1866 mode? Your results in the memory bandwidth test seem to match what others are getting with DDR3-1600, and I wouldn’t be at all surprised if they were running at 1600 through all of your testing. From what I understand it’s more frequent than not that people have to force the correct frequency and timings in BIOS for higher-speed memory these days.

  45. I do wonder why the added bandwidth matters so little with the IGP stuff… Maybe it’s not enough?

  46. Read the review closer. Llano doesn’t support 1600 and 1866 with two sticks per channel.

  47. Agreed, it’s weird to find some websites that had pretty nice improvements with 1866 and IGP performance. Something doesn’t really match. I think that the point is that Llano improves performance when you ask higher quality textures to be rendered. If I’m not wrong, they used the lowest quality details into all games when they benchmarked the games. I think the point is that while using high resolution textures the memory becomes a bottleneck taking down both GPU computing and CPU performance.

    Given that, it’s normal that games bounded just by CPU or GPU Computing performance doesn’t get a high improvement. Other websites have tried with medium or high settings in game settings for Llano benchmarks. That may explain the different results from techreport and others. Anyway, it’s hard to say since they didn’t place any header about the benchmark setup (neither details, neither resolution levels).

    Scott, I’m sorry, but I’m still disappointed with you. Your results try to point that other websites “lie” when they report higher IGP performance with higher speed memory. They found it, and you should look forward wich performance the higher speed memory unleashes. What’s wrong with you an AMD? Please, add different detail levels to the benchmarking and you may be able to notice something different in your testings than you did today. I’m pretty sure.

  48. I’d hardly call Bobcat a fail!

    Maybe the memory controller is set up to not allow the CPUs to hog all of the memory bandwidth, and AMD have it more balanced between the GPU and CPU. Certainly I recall the reviews stating that the memory controller was optimised to handle both types of workload.

    So in synthetic benchmarks it looks bad, but in reality it doesn’t matter. Even if that’s because Llano is still using K10 cores and thus couldn’t make the most of it anyway.

  49. Interesting results, quite unexpected. I’m saying this as other reviews have shown quasi-linear increase of performance in games when they replaced the DDR 1333 memory with 1600 and 1866.

    I know there are maybe different testing methodologies and you guys can’t know what the others have done in their testing, but it would be interesting to see what are the differences between the setup used by other sites that have obtained a performance boost of 20-40% and TR that had at most 10%.

    What could it be? Timings? Command rate? Or maybe the use of two RAM sticks per channel in the DDR 1333 testing and only one stick per channel in DDR 1600/1866?

    One review (fail to remember which) posted something like this, quoting from my head: “AMD’s memory controller is able to efficiently interleave memory access so that if you put two sticks per channel (thus 4 sticks for dual channel) you get more performance than what you would get by using only 2 sticks, one per channel”.

  50. Thanks for the extra testing. I was expecting a bit better scaling, so at least I know not to splurge on 1866 if I buy LLano.

    BTW, any news on the 35W chips ? or the A6 ? It seem the A6 is available for purchase, I’d love to know if I should get a [email protected], [email protected], Pentium G620T or i3 2100T for all those nettops people have me build for them.

  51. Calm down. The benches showed that higher bandwidth and lower latency together didn’t do much for performance anywhere. Further, C2D beat various flavors of A64 even with a disadvantage in latency and bandwidth.

  52. This is nutty. Intel get 19GB out of DDR3 1333 while AMD gets 15GB out of DDR3 1866mhz.
    How did AMD managed to design and release a memory control in mid 2011 that is *** 75% *** slower?

    And Bobcat doesn’t look any better…. what are the chances bulldozer follow ?

    AMD, what is going on with your CPU team ?!?! its fail after fail this year …

  53. A few FPS is significant when you’re straddling the edge of playability. It looks like, if you plan to play any games at all with the IGP, it’s worth spending a few extra bucks on 1600MHz memory.

    Thanks for humoring us.

  54. Given the target price point for Llano systems, it looks like the 1600MHz speed is the best option when considering performance and price.

  55. With those timings on the 1866 ram ofc it doesn’t scale much, would have been nice to see cas 8 for all 3 speeds and also cas 7 for 1333 and 1600 to have a real answer on what is best for Llano.(and some ram OC with a faster kit?)