Folding Forum

In a couple of months AMD 6XXX series cards will launch. There are rumors of improved shader design and 256 bit controller bus on lower end cards.
Combined with a new GPU 3 client there might be hope.

I'm yawning wide

I was folding with my HD2400, and I heard it.
I was folding with my HD3700, and I heard it.
I was folding with my HD4700, and I heard it.
I was folding with my HD5700, and I heard it.
I was folding with my GTS250, and I was shocked how fast it is...

PG can cope with the customer with the ATI graphics card, it's ATI that cannot.

The same way like A2 core for windows and a3 for linux ? Microsoft cannot and Linux distro maker cannot ?

I'm happy folding, it build up my patience, things should take weeks and are years long...

It is like Buddha monk's meditation, at the end of the day gives you peace and harmony...

I'm yawning wide

I said hope and not certainty.
Hope springs eternal.

Such negativity

I'm also not going to hold my breath, but I also will not just accept Amd will not catch Nvidia, ever. I think they have not focused on gpgpu at all the previous iterations of their architecture, now they have a solid foundation for their gaming cards, maybe they will redirect some of their resources into gpgpu. FireStream could make them allot of money if they are able to beat Nvidia's architecture ( looking at HPC clusters, those are big sales ).

MtM wrote:I think they have not focused on gpgpu at all the previous iterations of their architecture, now they have a solid foundation for their gaming cards, maybe they will redirect some of their resources into gpgpu. FireStream could make them allot of money if they are able to beat Nvidia's architecture

Here are some projects where ATI is more efficient than NV: MilkyWay, Collatz Conjecture, distributed.net.

Now, the most important question : do you know why those projects are doing so well on ATI ?

This is just turning into another ATI vs Nvidia thread. I've written several times about why ATI/AMD is inherently a different (harder-to-program) architecture than Nvidia, and why just using the peak FLOP counts to compare cards is unrealistic. Here are a couple things to consider:

viewtopic.php?p=117995#p117995
viewtopic.php?p=118275#p118275

toTOW wrote:Now, the most important question : do you know why those projects are doing so well on ATI ?

Because their GPU client plays nicer with Brook than F@H GPU client does?

toTOW wrote:Now, the most important question : do you know why those projects are doing so well on ATI ?

I do not know, but I hope that, however, someone explain this to me. In addition, I am trying to tell you that there are people who themselves are trying to solve this problem by creating a wrapper. From the author of wrapper I got an email, which I am not going to publicize that the wrapper is already being tested.

orion wrote:

toTOW wrote:Now, the most important question : do you know why those projects are doing so well on ATI ?

Because their GPU client plays nicer with Brook than F@H GPU client does?

Because 2+2 is simpler to calculate than A^2+B^2=C^2, and fah uses complex equations for it's simulations? If I had to guess, I'd say the math in FAH's simulations are a lot more complex.

Let's remember fah has been around a lot longer than many of these other projects. Their simulation methods are very advanced and have had time to mature more than other projects.

filu wrote:

toTOW wrote:Now, the most important question : do you know why those projects are doing so well on ATI ?

I do not know, but I hope that, however, someone explain this to me. In addition, I am trying to tell you that there are people who themselves are trying to solve this problem by creating a wrapper. From the author of wrapper I got an email, which I am not going to publicize that the wrapper is already being tested.

Putting a wrapper around Folding@home to make it run under BOINC will not make it magically perform faster on any GPU, ATI or otherwise. The network and client layer (i.e., the stuff that communicates with the FAH servers) is independent from the actual simulation core, which is what runs on the GPU (or, for that matter, on the CPU).

If you want to learn why it is nearly impossible to actually achieve peak FLOPS on ATI cards (actually, any GPU - but ATI is even harder), please read the posts I linked above.

I propose to close this thread. If I know the official start of the wrapper, then definitely I will inform you.

7im wrote:

orion wrote:

toTOW wrote:Now, the most important question : do you know why those projects are doing so well on ATI ?

Because their GPU client plays nicer with Brook than F@H GPU client does?

Because 2+2 is simpler to calculate than A^2+B^2=C^2, and fah uses complex equations for it's simulations? If I had to guess, I'd say the math in FAH's simulations are a lot more complex.

Let's remember fah has been around a lot longer than many of these other projects. Their simulation methods are very advanced and have had time to mature more than other projects.

Do we know for a fact that they use 2+2?

By comparison, IMO, yes.

Proteins fold in solution, so all of the molecules you see in those Youtube sims not only have the forces of each other to calculate, but also the forces between each moleculed and the solution.

MW@home *may* work on simulations with as many nodes(stars, planets = molecules in protein) but in large, space is an empty vacuum.

Also consider that fah protein molecules are very close together, so each molecule affects many others directly. In space, the nodes are very far apart. It would seem to me the force calculations in MW are much simpler than in FAH.

But then I'm not a molecular biologist or astrophysicist, so this is just an educated guess.


Either that, or the data from Milkyway is much easier to compile in to VLIW instructions than the F@h data.

bla, bla, bla
When Ferme become to be GTX480 client was ready in few weeks...
For ati nothing is changed for years...
Answer...

I don't know...

http://en.wikipedia.org/wiki/Jen-Hsun_Huang