128bit vs 256bit for GPU folding

[superkingkong]

Hi guys,

I can't find any of this related to folding. My question is, does it matter if i fold?
I'm looking at a gtx460 card, one is with 128 bit and the other with 256 bit card. The difference is about $10.

Thanks.

[7im]

Memory on GPU makes little difference to folding performance. For $10, go big. Unlikely, but it might make a difference in the future.

Shader count is much more important.

[superkingkong]

Thanks. Hmm, your second note makes confuse me more

Well, the card i'm looking at. EVGA 01G-P3-1361-KR is cheaper by about $10 and has:
Core Clock: 823MHz
Shader Clock: 1646MHz
Memory Size: 1gb
Memory Interface: 192-Bit

The second one: EVGA 01G-P3-1370-KR which is more expensive:
Core Clock: 720MHz
Shader Clock: 1440MHz
Memory Size: 1gb
Memory Interface: 256-Bit

[7im]

GPU work units are small now, and use very little memory. GPU Work units will get bigger, and might use more memory. Bigger WUs *might* benefit from better GPU memory. For only $10, get the better feature.

However, the faster clock speeds make the cheaper card look like a better bargain. Have you read the reviews on each card? Do they overclock to the same speeds?

[superkingkong]

I couldn't find any reviews for these cards that relate to folding. As for OC, some says they can push it up to 925mhz. for the 1361, cheaper version. As for the 1370, one guy pushed it up to 840, staying stable.

[{RaW}Eagle1]

For folding (as 7im has said) the memory interface makes little difference, as the memory bandwidth is not the limitting factor. However, if you should care very much about using the card to play games in high resolution (especially DX11) you may experience low texture fill rates, and potentially lag due to data being bottlenecked when being accessed from the frame buffers on the card with the lower bandwidth. . .

Overall bandwidth is calculated as: memory clock speed x bitrate or width (192 / 256) x transfers per clock (2 in this instance [GDDR5: Graphics Double Data Rate version 5]) divide 8 (bits to bytes)

This is of little consequence if using the card soley for folding. The more important factors are how many CUDA cores each card has and what each card's maximum stable clock speed is. There is only one revision of the 460 that had fewer CUDA cores: NVidia released a GTX460 Special Edition (which wasn't particularly special due to performing worse than its predecessors). I can't find any evidence that either card is an EVGA edition of the NVidia SE.

In terms of overclocking; both GPUs should have the same maximum stable clock depending on the available cooling. Personally I'd go for the card with the highest cubic feet per minute air displacement (if you can find that information), or in this case the only card to have a support page on the manufacturer's website: EVGA 01G-P3-1361-KR.

FYI: The last 2 letters in the model number are irrelevant for hardware comparrison purposes, as they refer to warranty length and country of sale (according to EVGA's forum).

Hope it helps.

[superkingkong]

Thanks for the helpful info.

[Ivoshiee]

If the past is any measure then do not buy equipment for the folding only. You may find relatively soon that you decided badly and project needs and abilities does not "favour" you anymore (if it ever did). Just take ability to fold as an added bonus and choose your equipment according to your main computer needs. Then you haven't lost anything and need not to be angry at FAH.

[superkingkong]

The more important factors are how many CUDA cores each card has and what each card's maximum stable clock speed is.

If a card with lesser CUDA cores but higher core clock, will it fold better than a card with higher CUDA cores but lower core clock?

Thank you

[Ivoshiee]

The more important factors are how many CUDA cores each card has and what each card's maximum stable clock speed is.

If a card with lesser CUDA cores but higher core clock, will it fold better than a card with higher CUDA cores but lower core clock?

Thank you

Answer to your question is defined by FAH WU size (atom count and relations). Small WUs will benefit relatively more of the clock gain than any additional CUDA core count. With larger ones it is other way around, but you have to take account the memory (size and speed) as well. But optimal configuration is changing as new projects enter the pipeline and old ones drop out of it.

[JimboPalmer]

I have a GTS 450 with a 128 bit memory bus.
GPU-Z indicates I use 22% of the memory bandwidth keeping the 192 shaders 99% busy, in 8031 which works on larger proteins
so a 256 bit bus may not have any benefit to my card.

[superkingkong]

Thank you so much for the info