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

[Rattledagger]

You may live it a cold place and have only 1 GPU. what is being referred to is multi GPU rigs which will normally require max fan speeds, the Atlas folder rigs were 4 double GPUs just millimeters apart. The complexity of running multi GPU systems 24/7 should not be overlooked, it may sound simple just populating vacant pcie slots with new cards but the heat, power supply and airflow factors are not to be underrated.

Well, mdk777 did also say "average GPU rig", so was assuming fan-speed was also relevant for the much more common single-GPU-rigs.

Going by Grandpa_01 first post, if these PPD is anything to go by, a single GPU-rig can get around 200k PPD and while this won't beat a 48-way CPU-system it should easily beat the systems to probably 99% of the active folders. Even with a fairly low-end card like GT 640 the majority of users should AFAIK get a nice boost in their PPD, atleast if theoretical FLOPS of these cards is anything to go by.

Conversely, SERVER MB and CPU are designed for long life and relatively high continuous use. Everything is relative...just throwing out some other variables other than just price and points.

Well, my dual-P2-333 back in the days was basically "eating" CPU-fans, but granted it wasn't any server MB.

As for fan life, you get what you pay for. That $10 rebate on a lowest end card will cost you eventually...

Rebates? Must be something they're using on the other side of the pond...

[Ben_Lamb]

Once explicit/cut-off simulations hit GPU street we can have that conversation. Right now there are none.

EDIT: plus, the only explicit project that was run on GPU has been withdrawn due to instability;
that tells us GPUs will continue to run implicit simulations _exclusively_ for quiiiite some time, heh, heh

I got the impression it was all sorted and 8057 was humming away folding an explicit simulation just like a cpu.

If the gpus are still limited by what they can do, makes me wonder what all the fuss is about.

Just the fact they cant even get a proper client sorted for the GPUs shows what a pig they are to program for.

Think I will have to put the gpus I bought on ebay to buy my bigadv stuff back lol.

[tear]

8057 is as implicit as it can be.

The only explicit GPU projects (if memory serves me well) were P7640-P7644 series (now defunct).

[k1wi]

The other elephant in the room is that it is not going to be cheap to buy motherboards/CPUs with sufficient pcie lanes for multi-GPU folding.

Most 1155 socket motherboards are limited to 20 lanes, unless they are using a lane multiplier.

[mdk777]

AMD FX boards can be had for $100 that have 2x 16 and 2x 8.

But again, we are optimizing hardware configurations for WU that might launch when??

[kromberg]

Going by Grandpa_01 first post, if these PPD is anything to go by, a single GPU-rig can get around 200k PPD and while this won't beat a 48-way CPU-system it should easily beat the systems to probably 99% of the active folders.

That right there is the whole point of my complaint. A pair of $200 GPUs can out point produce a 48-core SMP right. Stanford could have had the decency to accounce the major shift in point production. Let's face it, this was not an over the night decision. Within the last month I dropped about $3000 in a 3rd SMP box. If I would have known about the shift, obvisouly $3000 can buy a hell of a lot of GPUs.

[kiore]

Going by Grandpa_01 first post, if these PPD is anything to go by, a single GPU-rig can get around 200k PPD and while this won't beat a 48-way CPU-system it should easily beat the systems to probably 99% of the active folders.

That right there is the whole point of my complaint. A pair of $200 GPUs can out point produce a 48-core SMP right. Stanford could have had the decency to accounce the major shift in point production. Let's face it, this was not an over the night decision. Within the last month I dropped about $3000 in a 3rd SMP box. If I would have known about the shift, obvisouly $3000 can buy a hell of a lot of GPUs.

Well there was an announcement, and the test unit was released a week later, and as you will have seen from the discussion 200kppd is not actually easily achievable as things stand without manipulating cores etc. Also it has been pointed out that $3000 worth of GPUs do not work without some other hardware, whether the yet to be general released QRB units will be able to be run with minimal cpu resources also remains to be seen. What has been the case in the past may no longer apply in the future, even in the near future.

[Grandpa_01]

K1wi you can buy all the needed hardware from ebay relatively cheap
K9A2 Platinum 4 - PCIe slots $100
Any AMD Duel core $35
4 - GTX 580 $200 to $225 ea $900
1200 Watt Power Supply $150
memory $20

There is a 1,000,000 PPD rig for $1205.00 I just bought a GTX 580 for $215 shipped I have seen them go for $185 so the number can actually be less, You can build 4 of those for less than the cost of 1 - 4P it would take 10 average 4P's to make the PPD 4 of those rigs will make. The numbers are what they are. If things stay the way they are GPU folding will be the only way to go, most people will not fold smp or bigadv they will use the same reasoning that the GPU folders currently do (it is not worth it, there were allot of GPU's sitting idle before the recent QRB thing) which in reality it is not going to be worth it.

Kiore I am not manipulating cores, I changed it 1 time about 2 weeks ago it will only change to 2.25 when you get a 76xx WU then it can easily be changed back after that 1 finshes.

[kiore]

K1wi you can buy all the needed hardware from ebay relatively cheap
K9A2 Platinum 4 - PCIe slots $100
Any AMD Duel core $35
4 - GTX 580 $200 to $225 ea $900
1200 Watt Power Supply $150
memory $20

There is a 1,000,000 PPD rig for $1205.00 I just bought a GTX 580 for $215 shipped I have seen them go for $185 so the number can actually be less, You can build 4 of those for less than the cost of 1 - 4P it would take 10 average 4P's to make the PPD 4 of those rigs will make. The numbers are what they are. If things stay the way they are GPU folding will be the only way to go, most people will not fold smp or bigadv they will use the same reasoning that the GPU folders currently do (it is not worth it, there were allot of GPU's sitting idle before the recent QRB thing) which in reality it is not going to be worth it.

Kiore I am not manipulating cores, I changed it 1 time about 2 weeks ago it will only change to 2.25 when you get a 76xx WU then it can easily be changed back after that 1 finshes.

Sorry, changing cores then. The rig your describe above lacks an OS and the RAM seems very cheap as you will need quite a bit as these cards will have quite a bit of RAM themselves and more GPU RAM that mobo RAM is difficult, I am actually running a K9A2 plat at present and to squeeze in 4 Fermis would need a very cool environment indeed and narrow cards, GTX 580s are quite big, I was going to put my Asus GTX 570 on it but it covers 2 double spaced pcie slots and in the heat I have to deal with running 2 cards on it is already too hot in the daytime under aircon and in the open. Those living in very cold places will have an advantage here, why on my new rig I am going SMP as a i7 easier to cool and just 2 gpus. I used to have 2 x 4 card rigs and a 3 card one back when I had more space and less heat..

[JimF]

Those living in very cold places will have an advantage here, why on my new rig I am going SMP as a i7 easier to cool and just 2 gpus. I used to have 2 x 4 card rigs and a 3 card one back when I had more space and less heat..

They need to do some projects related to global warming. Then, the hotter it gets, the more incentive people have to continue with the project.

[Grandpa_01]

"kiore"
Sorry, changing cores then. The rig your describe above lacks an OS and the RAM seems very cheap as you will need quite a bit as these cards will have quite a bit of RAM themselves and more GPU RAM that mobo RAM is difficult, I am actually running a K9A2 plat at present and to squeeze in 4 Fermis would need a very cool environment indeed and narrow cards, GTX 580s are quite big, I was going to put my Asus GTX 570 on it but it covers 2 double spaced pcie slots and in the heat I have to deal with running 2 cards on it is already too hot in the daytime under aircon and in the open. Those living in very cold places will have an advantage here, why on my new rig I am going SMP as a i7 easier to cool and just 2 gpus. I used to have 2 x 4 card rigs and a 3 card one back when I had more space and less heat.

you can get by with 1Gb of ram, Linux with wine, 1 - $10 riser card, I think you get the point it is very cheap to build and run these rigs and they give great rewards, I have down clocked my 2700K to only drawing 20 W of electricity there is no need to run it any faster or use more energy than that smp is not worth it on a GPU folding rig.

If things stay as they are, my PPD will go up by 1,300,000 and I will have a very small investment in doing it my power consumption will drop about 10% to 15% so all in all this is a very good thing for me. Talk about heat all you want if you want to see some real heat and noise crank up 4 - 4P's at 3Ghz then talk about the heat my office stays 90+F in the summer with 2 of them running in it I cannot run all 4 in there they overheat themselves due to the fact they put out so much heat, and I live in a cool climate.

What is good for me may not necessarily be good for folding though, I have been wondering about it since I ran the numbers. If they can port everything over then all is well if not there are going to be some problems with other projects. What is the answer ? I do not know, I am a little concerned but I will follow PG's lead (put that carrot out there and they will come) anyway I hope things take shape pretty soon I am in the process of building the GPU farm and no I am not worried about the investment I already have everything on hand except 8 more GTX 580's and they are easy to sell if things change.

[JimF]

What is good for me may not necessarily be good for folding though, I have been wondering about it since I ran the numbers. If they can port everything over then all is well if not there are going to be some problems with other projects. What is the answer ? I do not know, I am a little concerned but I will follow PG's lead (put that carrot out there and they will come) anyway I hope things take shape pretty soon I am in the process of building the GPU farm and no I am not worried about the investment I already have everything on hand except 8 more GTX 580's and they are easy to sell if things change.

On the WCG/HCC project, a lot of people that don't have usable GPUs still want to crunch on their CPUs. In part, it is due to personal commitment to that particular project (related to cancer research), and part of it is badge-hunting, since they award badges on the basis of time spent crunching, not work done (they give points for the latter). And probably another part is inertia, if a PC is set up that way. So I don't think you will see a mass desertion of older projects that run on the older equipment anytime soon. More likely, the increased points will guide people when they upgrade. And that I think is what they are supposed to do.

[Ben_Lamb]

I have a feeling they wont be able to port everything over to gpu as has been said already there is not one gpu work unit available that can do the explicit simulation that a cpu can. This is not the greatest sign as GPUs realy have been slow at delivering in gpu compute due to being a pig to program for. I can see a gpu/cpu combo running everything more efficiently but not just a gpu. It will be easy when the time comes to reconfigure your pc to combined gpu/cpu work but you will still need multicore cpus. I recon at least a 12core mc to one gpu when this comes out eventually but gpu only wont take over and we will probably go a full circle and end up back on cpus again as we have seen all this happen before. Grandpa has the right idea just have boat loads of nice cpus and gpus in stock and configure as need be. Its all about resources realy gramps turns on 20 gpus, I turn on 2 its all relative.

[codysluder]

If things stay as they are, my PPD will go up by 1,300,000 and I will have a very small investment in doing it my power consumption will drop about 10% to 15% so all in all this is a very good thing for me. Talk about heat all you want if you want to see some real heat and noise crank up 4 - 4P's at 3Ghz then talk about the heat my office stays 90+F in the summer with 2 of them running in it I cannot run all 4 in there they overheat themselves due to the fact they put out so much heat, and I live in a cool climate.

You can probably reduce the power consumption and heat of your 4P by underclocking rather than overclocking but then you'd have more trouble arguing for your original premise. What do you really hope to accomplish by posting so many times on the same subject. It's all just spam anyway but it's FAH related so the Mods tend to keep it even though it's just a waste of forum space. (It is not just you, that applies to most of the posts in this topic from other people.)

[Rattledagger]

Well there was an announcement, and the test unit was released a week later, and as you will have seen from the discussion 200kppd is not actually easily achievable as things stand without manipulating cores etc.

Ok, let's say 100k ppd with a single GPU running v2.25 seems to be fairly easily accomplishable (despite many users fails to report which version they're actually testing with and if GPU is overclocked or not). While 100k PPD won't beat a 48-core bigadv-system, for probably over 95% of active users this will still be a huge increase in PPD for the cost of only a GPU and for some a PSU (afterall, they've already got a computer).