• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Understanding and improving memriy bandwidth

N

nrgibson

Junior Member
Hello - I am new here and not very 'techie' so sorry if this is an obvious question but I have researched for ages and still do not understand.

I have an Athlon 64 3200 with 1.5GB of RAM. I think my RAM is underperforming but not sure if it is the RAM or the motherboard. Using SISOFT benchmarks my RAM is 1384 MBS on the 'INT' measure and 1385 MBS on float. This seems to be under half what it should. On Everest benchmarks it says the write test is 600Mbs, half the 1180 suggested for my processor. Similarly the read test is 1463 MBS compared to the 2900 MBS suggested.

I do not really understand the difference in the test but it is clear something is wrong. The RAM is, according to Everest, 3 DDR SDRAM 512mb with CL2 trcd 2 trp 2 tras 5 and trc 7.

Can anyone explain this to me - tell me if I should be worrying (games stutter a bit but that maybe my 9800 ATI graphic card). I need to know what needs replaced - the RAM, the motherboard or the graphics card.

Alo SIsoft says my theoretical bandwidth is 6400 MBS - is this true is my machine really running that slow? It does say that my bandwidth is at 87% efficiecny - again not sure what that means.

Sorry if this is a basic query - hope someone can help

Neil G
 
If you are running 3 DIMMs in an A64, can I assume this is a S754 A64 (single channel)?

I do not think your theoretical bandwidth should be 6400MB/s if you are using a S754 A64 (and I don't think the S939 A64 memory controller supports 3 DIMMs), I am thinking you should be at 3200MB/s max theoretical bandwidth, as it's single channel memory.

Is it true your machine is really running that slow? Yes and no.
SiSoft doesn't lie, so you are indeed running that slow!
However, as was mentioned, memory throughput is only one factor to overall performance, and it's not a huge factor. Overall system performance will degrade on the order of 10% (probably a little less) with max theoretical bandwidth dropping from 6400 MB/sec to 3200 MB/sec.

You may be seeing a comparison of your 3200+ (socket 754) to a different 3200+ (Socket 939). Both are called a 3200+ and the benchmark programs may have the S939 version in their database. The S939 version uses dual channel memory, which doubles the theoretical maximum bandwidth (S754 is 3.2 GB/sec and S939 is 6.4GB/sec). However the S754 3200+ runs at 200 MHz higher, 2.2GHz instead of 2.0GHz for the S939 version.

In most real world applications, the S754 version of the 3200+ performs slightly better than the S939 version, though there are some applications that really love the memory bandwidth and the S939 version wins out. This is why they both have the same "speed designation" from AMD (more or less a wash in performance).

Any application that specifically focuses on memory bandwidth, as the benchmarking programs you are using will obviously greatly favor the S939 version for it's increased bandwidth. However if you look at performance as a whole in applications you actually USE, I think you will find the S754 platform is slightly faster overall at the same 3200+ designation.
 
Make sure you have the testing options set to bufered memory test. If it is turned off you will considerably lower numbers (probably more realistic). Also you three dimms are probably running at PC 2700 (the 754 I think defaults to 2700 will all three slots filled but I am unsure). Take one dimm out and retest and see what the numbers do.
 
Thanks folks,

Not sure how to tell if it is a socket 754 or 939 - does SIsoft tell me? Also I have looked more closely at the SIsoft mainboard information and it says that the RAM is DID400 and the speed is 3.0-3-3-8 (at 200 mhz). This is not the speed settings in my BIOS - which is correct? It also says the timing is 2.0-2-2-5 at 133 mhz. On boot up my machines says RAM clock 200 mhz, CAS LAtency 2. I am confused which is it actually running at? Should I slow the RAM clock - will this help

Thanks in advance

Neil G
 
Go here
http://www.cpuid.org/cpuz.php#download

and download CPU-Z, it's very small and will help you identify the components in your system.

Under the CPU tab it will have the Socket Type
Under the memory tab, it will have the memory settings you are actually running, and the number of memory channels (dual or single)
Under the SPD tab, it will have the default factory memory settings for various speeds. This is what the BIOS will automatically set latencies to if you don't override them.
 
Thanks,

Have done that and it says that I have single channel memory and under the memory tab it says the timings are 100MHZ with CAS 2. I still do not understand why I get benchmarks at a level similar to PC1600 RAM according to SIsoft not the PC3200 I have.

Is 1.5GB with a low bandwith and read write times better than 1GB with fast read write? I may buy new memory. The motherboard manaul (K8T Neo) suggests 1GB in two 512mb slots. Not sure why that is the case.

NG
 
So can you set the memory to 166 MHz in BIOS? Likely it won't do 200 MHz in that configuration because of the A64 memory controller. It may do 200 MHz IF you set a "command rate" of 2T (some BIOS will label that as "CPC OFF" and some will have a Command Rate parameter than can be either 1T or 2T.)

You should be able to have higher than 100 MHz RAM, though you will probably not be able to maintain 2-2-2 timings. The SPD tab should show appropriate timings at 166 MHz and 200 MHz. Or there should be an option in BIOS to use SPD parameters that will automatically use these timings built into the RAM.

As far as 1.5GB of slower RAM vs. 1GB of faster RAM, it depends a lot on what you are doing. If you're doing anything that is swapping to the HD, that takes a LOT more time than a RAM access no matter how slow the RAM is. (on the order of 1000 times slower, as hard drive access times are measured in xx milliseconds and RAM access times are measured in x nanoseconds.)
 
it's because the pc3200 score you are looking at is a dual-channel score and 3 dimms will knock you down to single-channel

also, you didn't mention your cpc... is it 1t or 2t? it makes a big difference in your sisoft test

but regardless, you are getting worked up over a synthetic benchmark that has almost nothing to do with real world performance and you shouldn't spend money to deal with something like that
 
You are running at PC 1600? Oh boy, you might have a DDR 100 memory module mixed in with the three dimms. The memory will default to the lowest frequency/timed dimm to ensure stable performance. I would only use one dimm at a time and rerun CPU-Z to make sure you have all PC3200 dimms.

Edit: A PC1600 dimm would default to 100mghrtz
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top