Blog post: "Unified GPU/SMP benchmarking scheme ..."

[Nathan_P]

Just a thought, Linux-Gpu core is under development. If its ever launched and you have one of the 4p mobo's with some PCIe slots - you have a win-win 48 cores for BA/SMP and some gpu loving on top. 4p rig has been my goal for a while but i've gone a different route. Asus Z9PE and dual E5's, can do BA, SMP and throw in several gpu's if i want to

[Nathan_P]

I have looked again at building a dedicated / watt efficient platform for a GPU dedicated system. It is easy to spend $700 without the card...so $1000 for the entire system.

$244 I7 CPU
$100 Z77 MB
$100 Gold or PLAT PSU
$100 OS
$80 MEMORY
$50 BOX
SSD DRIVE, etc. etc.

$250 $300 GPU and you are around a grand.

Now, you are going to be at 200/300 watts and a $1000 for the expected 100K to 150K ppd.

This compares with what 600 Watts and $2000-$3000 that people making 4P systems were spending to generate 200K to 300K PPD.(OC systems were going as high as 400K if I'm not mistaken) So yeah, the GPU will win in comparison, but it is not anything like a 10x loss, or even a 5x loss compared to the BIG-ADV. points that people were generating.


Obviously, YMMV, people will build SLI systems to reduce overhead costs etc. etc. etc.

I'm just pointing out that using the entire system for GPU has costs beyond the GPU itself...and if GROMACS uses Heterogeneous compute, the entire system might be used even more than in the past.

Ultimately, the Science should and will determine the best system/ points. I'm just not sure it will be immediately a lopsided loss for the 4P systems in comparison.

For a dedicated GPU box you can forget the $244 cpu - you just need a cheap dual core, and AMD might be the way to go if you can find a cheap mobo with 3-4 PCIe slots

[JimF]

Do you buy a new car every year ... or every few years ... or every 10 years? Determining when a computer is "worthless" is based partly on the numbers but also partly on emotion. You're investment is not going to be a choice that lasts forever because at some point you'll decide that the cost of power per point will pass some threshold compared to the cost of an upgrade or the cost of a new system -- or if you're less analytical, you'll feel the competitive pressure from those who are earning more PPD and you'll upgrade anyway ... or you'll find a super-neat game that will play better on a new system ... or.... Nobody can make that decision for you.

It is also useful to have more than one PC folding with equipment of different vintages, so that something is coming due for an upgrade each year (motherboard, video card, memory or converting a hard drive to an SSD, etc.). All can keep your interest up in the subject. More interestingly now, we have the shift to Gromacs 4.6, which may reshuffle the deck with respect to Nvidia and AMD, so the old assumptions may no longer hold true.

That gives an example of why it is not a good idea to put all your eggs in one basket at one time though, but diversify a little and be ready to change something out when its time has come. Flexibility has an advantage when technology is changing rapidly, and it usually changes rapidly in this area.

[rhavern]

For a dedicated GPU box you can forget the $244 cpu - you just need a cheap dual core, and AMD might be the way to go if you can find a cheap mobo with 3-4 PCIe slots

+1 to this, I have several 3x pcie dual core systems. You *will* need a dual core to keep three gpus fed. If you start looking at dual GPU cards, you may need more cpu but won't need to be fast. High quality psu is the number one requirement then add gpus as funds allow. Shift a lot of air to keep it cool.

Edit: That being said, any modern board with 3x pcie should be upgradeable to a more substantial CPU if a new hybrid core would benefit from it.

[mdk777]

For a dedicated GPU box you can forget the $244 cpu - you just need a cheap dual core, and AMD might be the way to go if you can find a cheap mobo with 3-4 PCIe slots

Yup, that was the case for the old GPU projects. I followed / commented on the ATLASFOLDER design as it developed.

However, I am "anticipating" that the new Gromacs will use more CPU...

That will have some interesting and useful implications for F@h particularly as we look at more and more cores on donor CPU’s (and things like GPU integration). That’s all for now; we’ll keep you posted on progress.

That "anticipation" is what makes the OP have a hard time.

It may, or may not be required. We won't know until the WU are released.

PS..I would love to skip the $100 OS too if that is possible. Today I'd pay the penalty rather than have the aggravation. There are announcements regarding improved Linux support. When I see that happen, It will also change my build requirements.

[mihapiha]

Hmm. With the current P8102 WUs I'm getting, my folding farm actually manages 700k PPD. I'll start saving and then add another computer on GPU bases. Hopefully this will create the balance needed to maximize the folding effort from my part.

[patonb]

^Well, please share the 8102 wealth then

[Nathan_P]

8102 is hit and miss, i get half a dozen then back to the 8101's. Whichever way the cards fall now though you'll be set - that SR2 can take 4 gpu's and 7 if you water cool them

[patonb]

^but my power bill can't

I still have trouble wrapping my head around how the gpu can do the same work as a 4p cpu...... but hey, its not my job.

[Nathan_P]

^but my power bill can't

I still have trouble wrapping my head around how the gpu can do the same work as a 4p cpu...... but hey, its not my job.

It has me puzzled too, hopefully someone from PG will be able to explain in in simple terms.

[7im]

^but my power bill can't

I still have trouble wrapping my head around how the gpu can do the same work as a 4p cpu...... but hey, its not my job.

It has me puzzled too, hopefully someone from PG will be able to explain in in simple terms.

PG already did, back in 2006. Read the GPU FAQs...

The next major step forward: Streaming Processor cores (September 2006)
Much like the Gromacs core greatly enhanced Folding@home by a 20x to 30x speed increase via a new utilization of hardware (SSE) in PCs, in 2006, we developed a new streaming processor core to utilize another new generation of hardware: GPUs with programmable floating-point capability. By writing highly optimized, hand-tuned code to run on ATI X1900 class GPUs, the science of Folding@home will see another 20x to 30x speed increase over its previous software (Gromacs) for certain applications. This great speed increase is achieved by running essentially the complete molecular dynamics calculation on the GPU; while this is a challenging software development task, it appears to be the way to achieve the highest speed improvement on GPUs.

GPUs were already known to be much faster than PCs a long time ago. They just couldn't run the same solvent models as the PC, so while much faster, the science was much less helpful.

Now that GPUs CAN run the same solvent models as PCs, that muliple X speed up is being recognized.

[mdk777]

Previous post in a similar thread:

viewtopic.php?f=16&t=22756&start=60#p227514

They doubled the FLOPS from Jaguar to Titian with 90% of that increase coming not from doubling the CPU(which they did) but from updating/increasing the GPU.

GPU are math processing monsters.

[bruce]

I still have trouble wrapping my head around how the gpu can do the same work as a 4p cpu...... but hey, its not my job.

It has me puzzled too, hopefully someone from PG will be able to explain in in simple terms.

Simple terms? OK. Here's a shot at it (even though it's oversimplified and not perfectly accurate).

Let's suppose Your 4P has 48 cores and with SSE optimizations it can manage an average of maybe 160 simultaneous FP operations @ 2600 MHz. Let's suppose your GPU has 1500 shaders which can average 1200 simultaneous FP operations at 800 MHz. Your FPU is faster, even when it has to contend with the slowness of the PCIe connection as long as I/O can be overlapped with computing.

[alancabler]

Some things never change...
I read arguments about this issue (not so much in this thread) that go something like this: "My expensive rig has been blowing everybody away points wise, but now somebody can make more points than me for less money, so I'm gonna..."

While it's true that most of us must consider how to maximize our contribution with a minimum of total costs, such talk (as above,) can often ignore the whole body of donors and the points avg. donors receive vis a vis the lavish donor.

Let's touch on a bit of reality with an example:
Suppose a donor has (lavishly) spent $3400 on a 4p Opteron 6274 rig and makes 500,000ppd.
500,000ppd/$3400=150ppd/$
Then, let's consider an average donor who spent $370, 2 1/2 yrs. ago on an AMD XII that makes ~1850ppd.
1850ppd/$370=5ppd/$
Very often, the "small" donor is making a much greater proportional sacrifice in terms of costs in order to donate to the project than the lavish donor.

Listening to: Chris Ledoux- Five Dollar Fine (For Whining)

[mdk777]

No doubt.

Of course you leave out electricity cost which predominates in the long term;
Which makes the small donors participation even worse from a point/watt and point per total $ calculation.

However, that is one critical function of the point system; to give feed back on the efficiency of a donors participation to the project.

If you are not concerned about it, you simply ignore the points.
"FOLD with what you got"...and there is nothing wrong with that.

If you want to donate, "the best bang for the buck", this thread discussion is helpful.