TR Forums

I'm facing low RAM performance and weird BIOS restrictions on an abit motherboard. I will outline the details below.
The topic is relevant to RAM timings and latencies, advanced BIOS switches and the like. If you're illiterate in these areas, don't read further.

Important system details:
abit AN-M2 motherboard using the latest BIOS
- nForce 630a chipset
- 4 GiB of DDR2 in a 2+2 configuration in DIMM slots 1 and 2
- onboard FDD and HD audio both switched off, IDE and SATA both enabled, no RAID, SATA ports 2 and 4 occupied, 1 and 3 empty (sometimes swapping around seemed to do differences)

Less important system details
- AMD Athlon64 X2 4000+ overclocked - RAM overclocked too
- nVidia GeForce 8800 GT overclocked
- Creative Audigy ES overclocked (no, kidding ) - don't worry, I don't buy Creative, I got this used for 9 dollars because of the features - hate them or not, those features are golden for this money with Daniel K's drivers and unlockage

Other information:
I've made another forum post about several Creative cards crackling used with this motherboard - many people online blame nForce chipsets and I seem to believe that now. Crackling and general problematic behaviour was tracked back to RAM itself - so I'm hoping that this thread will allow me to hit two birds with one rake. Specifically, abnormal and/or high RAM usage lead to buffer starvation on Creative cards and this is slightly mitigated when running dual-channel - examples include scrolling in Firefox (worst of all, so I assume the program is systematically raping my RAM) and certain conditions in Crysis, mostly many flame/fire/smoke effects on screen (still much less severe than Firefox).

Finally, the issue itself:
LOW MEMORY PERFORMANCE. Why, oh tell me why, does this board perform like a load of crap? Seriously, it's severe. Let me outline it in detail:

1. Throughput is low. By that, I mean F- DAMN SLOW. I've used a much more budget AsRock board before with an nForce 4 chipset and onboard audio only - using overclocked 667 MHz DDR2, memory throughput was knocking at the doors of 8000 MiB/s when reading and exceeded it when copying (latest Everest). THIS BOARD HOWEVER, oh please, this could only reach 6500 MiB/s read times when I packed it with my new 800 MHz DDR2 modules which I overclocked all the same. Overclocked 667 performs faster in a low-end motherboard than overclocked 800 does in a sort of almost-mainstream motherboard. Why? (Also, the AsRock board was stressed with 4 modules of 1 GiB each, while this board only has to hold up 2 - need I say more?)

2. BIOS behaves oddly. No matter what I set for some timing numbers, they stay high. I am able to configure the first 3 but the last 2 are locked at abnormal values - RAM boots up using 5-5-5-18-26. Strangely enough, nTune under Windows can physically change them to more normal values once I've finished booting, but that's tedious to do every time. I'd like to know where to kick this piece of poop to make it behave. (Note: I'm considering buying a better brand like Gigabyte next time, or maybe even MSI, and contemplating opting for AMD's 7 series of chipsets.)

3. There are BIOS values that I know nothing about. For now, I'd like to know about latency. Here goes:
The BIOS set DIMM latencies to these values (or something like it - typing from memory): DIMM1: 127.5 ns -- DIMM2: 75 ns -- DIMM3: 75 ns -- DIMM4: 75 ns
I found this odd, but google revealed it's not an isolated case.
Experimentation: setting all the latencies to 75 ns creates an unbootable, unrecoverable system that will not POST and allow the recovery keyhold ("Insert" key for this board) to make it POST, meaning I need to flush the CMOS.
Setting both latencies to 127.5 ns does work, however, but I feel no differences yet. What is the purpose? Is it better to keep modules in sync? I would guess: yes. But I honestly don't know what this does. Note: between the two values mentioned, I can set 105 ns as well, but I have yet to try whether it will POST.

There are a number of points above which need advice/answers, and I would appreciate all the help I'd get.

1). Seems like there are problems with the mobo making the RAM run slower than normal (bad BIOS, failing northbridge??). Also test the RAM modules on different boards and see if you still get the same throughput to make sure if the RAM modules themselves are bad or not.

2). Those numbers indicate that the mobo can't handle the RAM and the BIOS's auto-adjust isn't working properly or the BIOS is auto-adjusting to a setting that the mobo can handle.... What RAM modules do you have? brand - model number. The tRAS # at 18 is normal for some corsair modules. but I havent seen any others with 18... usually 15 or lower even.

3). OK... i have no idea what that means. For DDR2 800 the typical total latency should be way less than 75ns and definitely lower than 127.5ns.... that's like in the realm of old old RAM modules like the classic EDO RAM in the early 90's... so what the heck is that value supposed to signify?

Thanks for your reply. Here are some additional details:

I've used this abit board before with two RAM configurations - the previous one was a dual channel mixture of different brands, including and not limited to Kingmax and Nanya (Samsung) running 4 modules of 1 GiB each. They were all 667 MHz modules, from my old setup using the AsRock as lined out above. Low performance and inability to time them right (all rated for 5-5-5-15-20 but the board produced the above timing issues) was observed.

Currently (as above) I'm using two GeiL Value modules which are also rated for 5-5-5-15-20 and are 800 MHz modules. Everest says I'm running them around 840 MHz, nTune says I'm running them around 940 MHz. Either way, they still don't reach the performance of my old rig and my motherboard doesn't time them right by default - even when nTune can set the timings and Everest confirms the change in hardware, so it's workable. As for the latency settings, I believe many motherboards today have something like those values but I also don't know what they do.

Meadows wrote:

Currently (as above) I'm using two GeiL Value modules which are also rated for 5-5-5-15-20 and are 800 MHz modules.

I'm just wondering... Is that RAM module brand and type on the compatibility list for the mobo? If not, that might explain some of the weird problems you are having.

CB5000 wrote:

Meadows wrote:

Currently (as above) I'm using two GeiL Value modules which are also rated for 5-5-5-15-20 and are 800 MHz modules.

I'm just wondering... Is that RAM module brand and type on the compatibility list for the mobo? If not, that might explain some of the weird problems you are having.

What about the old ones? Were Kingmax, Kingston and Nanya also off their compatibility list?
What sort of damned compatibility list DO they have?

Meadows wrote:

CB5000 wrote:

Meadows wrote:

Currently (as above) I'm using two GeiL Value modules which are also rated for 5-5-5-15-20 and are 800 MHz modules.

I'm just wondering... Is that RAM module brand and type on the compatibility list for the mobo? If not, that might explain some of the weird problems you are having.

What about the old ones? Were Kingmax, Kingston and Nanya also off their compatibility list?
What sort of damned compatibility list DO they have?

I dunno if they are on the compatibility list or not. A lot of manufactures have a list on their manual about which brands have been tested and which haven't been. You can always call them.... Also you can try wiping the CMOS memory and see if that helps. I think you can do that by shorting the CCMOS1 jumper between pin 2 & 3.

I know how to do it, and trust me, being an overclocker I have done it more than enough times already.
On the compatibility list, 1 GiB GeiL modules have been tested, but these ones are 2 in size. Regardless, it shouldn't produce such hilarious issues. I will look into the issue when I can using a replacement motherboard and/or memory, but I'd really not like to spend much just to fix a sound card that uses the PCI bus like a pig's feeder.

Meadows wrote:

I know how to do it, and trust me, being an overclocker I have done it more than enough times already.
On the compatibility list, 1 GiB GeiL modules have been tested, but these ones are 2 in size. Regardless, it shouldn't produce such hilarious issues.

You never know. If they haven't tested the module then there is really no guarantee. You said you are running 2+2, what are the other 2 modules? Brand and model number please. It seems odd that in your OP you did not think the RAM stick brands are important information.

Also, can you test something other than synthetic benchmarks? May be get some WinRAR numbers?

Flying Fox wrote:

Meadows wrote:

I know how to do it, and trust me, being an overclocker I have done it more than enough times already.
On the compatibility list, 1 GiB GeiL modules have been tested, but these ones are 2 in size. Regardless, it shouldn't produce such hilarious issues.

You never know. If they haven't tested the module then there is really no guarantee. You said you are running 2+2, what are the other 2 modules? Brand and model number please. It seems odd that in your OP you did not think the RAM stick brands are important information.

Also, can you test something other than synthetic benchmarks? May be get some WinRAR numbers?

2+2 means it's 4 GiB of dual channel by two sticks. I'm not a big fan of writing it as multiplication.
Synthetic benchmarks in Everest are good enough, they do highlight actual problems. I'm using my new board now and it doesn't produce weird stuff. RAM is also faster than ever before, throughput going past 9000 MiB/s without breaking a sweat and average latencies as low as 43 ns.

But!

I still don't know what the refresh cycle time is (I got that name from Overdrive on my new board - it's the weird nanosecond setting that baffled me and some others).
Here's a screenshot: http://img408.imageshack.us/img408/7124/tempok3.jpg (don't look at the rest of the timings, it's not tuned to the sticks' actual ratings yet)
Two questions:
- What does it do?
- Why do all boards behave like they need the first stick's refresh cycle limit to be larger?

Thanks for any more advice/thoughts that might be dropped here.