Is there much life left for Uniprocessor WUs?

[Sumner_R]

Naturally I don't expect any concrete answers to this question, but I was kind of wondering what some of your opinions might be on the subject. I have a couple of old Pentium 4 machines that I fold on occasionally. What I've noticed is that (I'm guessing because of the new benchmark machine) these older computers of mine are coming dangerously close to the deadlines these days even when they are folding 24/7. I still like to pat my older computers on the head and smile thinking that they still have a use after all these years, but I'm afraid that one day very soon even folding wont be an option for them and it's off to the big recycling center in the sky. So what would you guess to be how much life they have left in them for folding@home? 1 year? 2? 6 months?

[P5-133XL]

a couple of years ago, I took both my Dell 8400 P4's and placed GTX 460's in them so I could GPU fold but you really need a P4 with PCI-e for that to work and many P4's never had that bus. Worked like a charm. They get the exact same PPD for the same GPU WU's as much more modern CPU's. I do not CPU fold on the same machine because that decreases my GPU folding PPD.

That being said, if you do not have a GPU to use it is hard to justify CPU folding on them because the PPD/W is so low. On the other side, every WU that a P4 CPU finishes is still a step in the right direction. So as long as the CPU's can finish on time then all is good.

Next point, there are still low-end dual core processor CPU's being produced and lots of people are buying them because they are cheap and good enough. With the default of the v7 client being at medium where it only runs a single core there are lots of individuals running uniprocessor WU's so they will continue.

[Napoleon]

The ancient core_78 for uniprocessors only is still alive and kicking, but looks like they get assigned only to really, really really slow CPUs like my Atom330. But like P5 pointed out, there has even been specific demand for dualcores or lower quite recently: viewtopic.php?f=66&t=23349&p=233160#p233160

Not every project is for big iron, so every little bit helps.

[7im]

If you have a PCIe slot, and $250, the new GTX 760 GPU is a good sweet spot for price versus performance. While your P4 will produce 100-200 PPD running a CPU client, it will produce around 60,000 PPD using that new GPU, and will be supported for many years. It's a really great way to extend the life of an older computer.

[folding_hoomer]

Or you buy an AMD-GPU like 7870 to fold with Core17 and save money while the CPU is only needed to initiate the WU and is than only used infrequent.
Nvidia-GPU´s like GTX760 utilize the whole CPU (P4 without HT) while folding with Core17 - and the P4 is a real power hog . . .

[Sumner_R]

Thanks for all of your replies! I just took a look and none of them have any PCIe slots. I remember seeing a GT430 once that was a regular PCI card. Do you think those would fold any better than just the CPUs themselves?

[Zagen30]

Much better. From a couple of reports they do around 5-6.5k PPD on core 15 WUs. They appear to fare worse on core 17 WUs, as I'm seeing a report of around 3k PPD, but that's still a lot better than a P4.

[GreyWhiskers]

I can attest to the ability of a GT430 to pull down significant performance.

I had an old HP desktop with a Pentium 4/HT 3.2 GHz processor (one core, 2 HT threads) with a AGP bus. When the ATI card I had in the AGP bus was depreciated by FAH so it could no longer get Core 11 work units, I bought a GT430 (Zotac) for the PCI (non express) bus.

As you can see from the GPU-z shots, the card wonderfully overclocks (core 910, memory 780) with complete stability (except for the P8900 zeta unit core unit I show below). The card has no fan, so I bought a USB fan that lays in the bottom of my case blowing directly onto the massive heatsink of the card.

I've included a selection of the projects, both Core 15 and Core 17, it works with. As you can see, it almost finished a p8900 - getting an error after 92% , but did complete a p7663. The ppd stats show that the card can do these larger runs, but its 96 CUDA core design just isn't very efficient, compared with the Kepler card I also use (e.g., GTX660Ti). I turned off the Advanced Methods flag, so am just getting the basic Core 15 WUs, of many different projects. The ppd runs from 7663 to 2697. Note that this lowest ppd came from a time before I started overclocking the GT430.

Code: Select all

Project  WU Name         Name       Core name   Core ver frame time kfact   credit      ppd       D/L (UTC)      Complete (UTC)   frames compl   atoms  run clone  gen
 7624     p7624    HP a475c Slot 01 OPENMMGPU     2.22    00:26:33    0    14,093.00 7,643.70 10/10/2012 07:00  10/12/2012 09:47       100       1980   352   0    87
 7625     p7625    HP a475c Slot 01 OPENMMGPU     2.25    00:32:28    0    14,093.00 6,250.70 12/22/2012 01:02  12/24/2012 11:51       100       1980   442   2     7
 8074     p8074    HP a475c Slot 01 OPENMMGPU     2.25    00:09:26    0    3,874.00  5,913.70  7/9/2013 16:03   7/10/2013 00:59        100        917   20    10   141
 7660     p7660    HP a475c Slot 01 OPENMMGPU     2.25    00:13:03    0    4,431.00  4,889.40  1/20/2013 09:03  1/21/2013 16:22        100       1179   198   0    12
 8010     p8010    HP a475c Slot 01 OPENMMGPU     2.25    00:13:24    0    2,510.00  2,697.30  6/22/2012 12:02  6/23/2012 10:27        100        917    0   1337  77
 7625     p7625    HP a475c Slot 01 OPENMMGPU     2.25    00:31:18    0    14,093.00 6,483.70  5/6/2013 01:40    5/8/2013 06:21        100       1980   109   0    169
 8900     p8900    HP a475c Slot 01    zeta               49:45:00   0.75            2,400.00                                          92        40000  473   2     6
 7663     p7663    HP a475c Slot 01    ZETA        0      00:18:06   0.75  2,898.50  2,306.00  5/20/2013 10:25  5/21/2013 16:49        100       22000   3    34   44

[Sumner_R]

That's amazing. I never would have thought a regular PCI card could get those numbers. Can you use more than one at a time? They have some on ebay for $45 so maybe I'll try one out.

[7im]

Yes, if your P4 has Hyperthreading, you could run 2 GPUs very well.

[Sumner_R]

Only one of them has hyperthreading. All of my older computers include:

2.4 GHZ P4
2.8 GHZ P4 (HT)
3.0 GHZ Celeron D
2.2 GHZ Athlon 3500+

I can't help it--it's the nostalgia, you know?

[7im]

1 is better than none. And 2 is a bonus.

[GreyWhiskers]

Actually, the core15 work units take practically no Cpu time for this card so you might be able to run more than one.

Of course, you need to run win Xp or higher -- no Linux -- for core 15. And DO watch your temps.
F
[EDIT] Do also look out for your power supply capacity. More than one of the 430s might overtax a small PS.

[EDIT2-] There are no guarantees in this business so there is always the possibility that Core 17 will become the wave of the future of GPU folding. I just hope that we get a way to have the assisnment server capable of assigning smaller work units to productive HW like these. But no guarantees.

[Napoleon]

Of course, you need to run win Xp or higher -- no Linux -- for core 15.

WINE works for the GPU v6 console client. Takes a bit of tweaking, but it's doable - viewtopic.php?f=54&t=23958.

[Sumner_R]

Hey guys, I was actually rummaging through my old pile and I found another old machine that does have a PCIe slot. It was mentioned earlier that a GTX 760 or a Radeon 7870 would be good choices to use to extend the life of old hardware. But it only has a 300W PSU in it so I was hesitant to run out and buy a new card for it without knowing if that was even enough juice. If not would you happen to know any other recommendations that would run on a 300W PSU?