Bottlenecks when folding/mining.

[heineygoat]

A train of thought I had...

I notice that when folding I have option to do the heavy work on either the CPU, or what ever GPU's I may have, or both. And if I opt to say, fold on the GPU it will hardly effect CPU usage.

This implies the processors work interdependently when crunching numbers right? And there is therefore no real potential for bottleneck between the CPU and GPU whatsoever.

 

If you're folding on just a Titan, you wouldn't see any difference in numbers crunched whether you have a 3960x or an Athlon CPU because the Titan is going to run at 100% regardless, right?

Even if you fold on an Athlon and a Titan they're both going to run at 100%. Sure you will get significantly less work done from the Athlon but that's not to say its holding back the Titan.

 

I'm not going to pair a Titan with an Athlon, I'm just curious, maybe I'm mistaken on how folding and mining works.

[bradscoolio]

The work units folded by gpu's are coded specifically for gpu's. Therefore a cpu couldn't do them if they tried, they're just there to manage the data flying through the PCI bus.

A cpu and a gpu folds 2 separate work units.

Disclaimer: I didn't research that at all so it likely isn't 100% true, though this is my understanding as to how gpu and cpu WU's are separated ;) @Whaler_99 please correct me if I'm wrong.

So to answer your question. No, an Athlon wouldn't bottleneck a titan when only folding GPU WU's

[Mentalguy]

Amd cards need a little CPU power to run at 100%. As long as you can leave one thread open for GPUs to use, it will be fine.

[Whaler_99]

The work units folded by gpu's are coded specifically for gpu's. Therefore a cpu couldn't do them if they tried, they're just there to manage the data flying through the PCI bus.

A cpu and a gpu folds 2 separate work units.

Disclaimer: I didn't research that at all so it likely isn't 100% true, though this is my understanding as to how gpu and cpu WU's are separated ;) @Whaler_99 please correct me if I'm wrong.

So to answer your question. No, an Athlon wouldn't bottleneck a titan when only folding GPU WU's

 

You is correct sir.

 

If you look up some of the project lists - typical GPU projects include Core16, Core17, Core15, etc. - projects 78xx and such

CPU typically are Corea4, Corea5, etc and projects typically look like P8xxx

 

They are specifically coded basic on the instruction set - GPU versus CPU.

 

There is even some further definition when we look at CPU WU's. WU's developed for bigadv projects are huge - a LOT larger then normal ones with return times of a day or two versus weeks for the normal CPU WU's - this is because bigadv ones are only available to Linux clients running 16+ cores, so they know these bad boys can crunch them and fast - and are rewarded accordingly...

 

I wouldn't be surprised to see in the future some WU's for specific GPU's, if it isn't to complicated. Something like a Titan or 780 can run circles around a 600 or 500 series for instances. So why not put them to work doing some major ones....

[Mentalguy]

I wouldn't be surprised to see in the future some WU's for specific GPU's, if it isn't to complicated. Something like a Titan or 780 can run circles around a 600 or 500 series for instances. So why not put them to work doing some major ones....

If they do that, i see gk110s in your future