Re: Low price GPUs

[PSI_Performance]

figured the CPU was a little on the low end LOL

It wasnt anything spectacular when it was new!

Just curious, would the CPU still be an issue if i threw a decent GPU in there and tried to just use the GPU? Or is the CPU still going to cause me grief?

[mmonnin]

That'll probably give you more PPD that way. An AMD GPU needs 1 CPU core to run while Nvidia will need a couple %. I'd guess an nvidia GPU would run full throttle while the AMD may be CPU starved to a point.

[PSI_Performance]

at this point, having it do anything at all, is a big advantage over whats going on right now LOL

[bruce]

at this point, having it do anything at all, is a big advantage over whats going on right now LOL

Agreed.

I'm searching for people who can enable more CPU projects for this class of machine but can't promise when that might happen.

The addition of a supported GPU would be a very significant improvement but the real question is what type of slots your motherboard supports. Most modern GPUs use PCIe slots and that generation of motherboard probably doesn't have any.

[Napoleon]

Depends on your PSU & motherboard, I guess. I feed an old NVidia 9400 GT with my Sempron 3400+, the mobo just happens to have one x16 PCIE 1.1 slot. I'd venture Sempron 2800+ is capable of feeding a low/mid range NVidia Fermi or Kepler card just fine. Single core of an Atom330 CPU is no match for the Sempron (yes, Atoms are that weak), yet it can feed a GT430 without too much trouble. A high end GPU might not be used to its full potential. If for no other reason, due to PCIE bandwidth. I very much doubt any Sempron board would have PCIE 2.0, let alone PCIE 3.0. PCIE 2.0 or 3.0 cards are compatible with 1.1, but the PCIE bus bandwidth is limited accordingly.

If your PSU has a 6 (or 6+2) pin PCIE external power connector, GTX 650 (or even GTX 650Ti) will probably do well. If not, GT 640 is pretty sure fire. It's powered directly from the PCIE slot (card power draw is max. 65W at stock clocks, PCIE slot can supply max. 75W). You can even go for fanless versions of GT 640, such as Zotac ZONE. Provided that your case has sufficient ventilation on its own, of course.

[art_l_j_PlanetAMD64]

PCIE 2.0 or 3.0 cards are compatible with 1.1, but the PCIE bus bandwidth is limited accordingly.

This is true about 99.9% of the time, but I had some ASUS A8V-E SE motherboards that would not work with a Gigabyte 9800GT. The motherboards would work with 9500GTs, 9600GTs, 9600GSOs, GTS250s, GTX260s, GTX275s, and GTX285s, but for some strange reason, the Gigabyte 9800GT was a no-go.

[PSI_Performance]

Its been so long since ive looked at the thing, but i do believe its only got an AGP 4/8x slot, no PCIe. Back in the day i used to have a decent card in there, but gave it to a buddy when his burnt up (and at the time wasnt using this computer at all), thus the super sweet mach64 card thats in it now haha

[GreyWhiskers]

at this point, having it do anything at all, is a big advantage over whats going on right now LOL

Sorry I didn't catch your comments when you posted them.

Anyway, if your computer has an AGP bus video card, or even just plain PCI bus (no PCI-Express), then you might consider the solution I implemented last May when they depreciated the older ATI video cards available on my old HP computer's AGP bus. This is giving ~6,000-ish ppd - astounding considering what the computer used to do. The best I could get out of the Pent 4/HT 3.2 GHz single-core CPU running a UNIPROCESSOR client was about 200 ppd.

I bought a Zotac ZT-40605-10L nVidia GeForce GT430 512MB DDR3 VGA/DVI/HDMI PCI Video Card. This plugs into one of the PCI (not PCI-Express) slots in my old HP Win XP AGP bus computer. The card is fanless, is a 96-CUDA core Fermi card with a base 700 MHz core clock. It takes its power from the PCI bus, so doesn't need any additional power supply connections. Because it is fanless, and I'm running it heavily overclocked 24/7, I bought a Thermaltake Mobile Fan 12 External USB Cooling Fan 12CM AF0007 that I have lying on the bottom of my case right under the card blowing on the huge heat sink fins that cover the entire surface of the card. This is all it needs to keep the temps cool (about 60 deg C) and the card stable. Being a USB fan, it just plugs into one of the spare ports on the back of the computer. I long ago removed the side panels from my two desktops, so they get direct ambient air.

I popped it in to my system replacing the old ATI card, installed the appropriate Nvidia drivers. I'm still running the 301.42 drivers - which work fine. I've seen no need to update them.

FAH V7.2.9 immediately recognized the card, and started folding with no issues. Since then, one of my HFM.net databases shows 124 successfully completed work units. There are more - I'm taking the stats from my laptop which I often take on the road, where it can't see the completions.

Anyway, this card processes all of the same Fermi GPU work units that I've gotten on my other GPUs (GTX660Ti, GTX560Ti, GTX560M laptop), only slower.

The base 700 MHz clock is wonderfully overclockable - I'm running it at an absolutely stable 900 MHz, with the memory clock also overclocked. With both overclocked and stable, I'm getting ~6,000-ish ppd - not too shabby considering the best I was able to get with my old AGP-bus ATI4xxx card was about 1,500 ppd.

8073 p8073 HP a475c Slot 01 GPU OPENMMGPU 2.25 (TPF) 00:09:39 0 (points) 3,874.0 (ppd) 5,780.9 1/24/2013 3:26 PM 1/25/2013 7:34 AM 100 FINISHED_UNIT 917 0 1899 37

7660 p7660 HP a475c Slot 01 GPU OPENMMGPU 2.25 (TPF) 00:12:57 0 (points) 4,431.0 (ppd) 4,927.1 1/18/2013 7:34 AM 1/19/2013 5:12 AM 100 FINISHED_UNIT 1179 658 0 1

7623 p7623 HP a475c Slot 01 GPU OPENMMGPU 2.25 (TPF) 00:30:41 0 (points) 14,093.0 (ppd) 6,614.0 1/2/2013 3:38 PM 1/4/2013 6:50 PM 100 FINISHED_UNIT 1980 659 0 85

Looking at the current market, there are still these three PCI (non-express) cards that have been made. The 430, 520, and 610. Interestingly, the GT430 I have is a 96-CUDA core card, while the others are only 48-CUDA core cards. Both the 430 and 520 have been discontinued, but there may be some residual stock lying around. The 610 is a current production card.

[PSI_Performance]

Wow, thats insane for a 'crappy' little PCI card!

If a ~$60 card (just checked a GT610 and it was $59) can pump out ~6000ppd, thats definitely the way to go. especially since the CPU itself could never compare to that!

But, now that worries me even more...

My current pair of 8800GTX's in my other computer only throws down 5000-6000ppd(not counting CPU contributions) with a mild overclock... WTF? LOL

Not that this computer is new by any means, but these 8800GTX's sure as hell werent a $60 card when i bought them, its sickening to think that a new $60 card can be comparable to a pair of ~$400 cards(when they were brand new!)

Or, do i have some sort of issue... Ive never folded with these 8800's so i dont know if this is just the norm, or...

[art_l_j_PlanetAMD64]

If a ~$60 card (just checked a GT610 and it was $59) can pump out ~6000ppd, that's definitely the way to go.

My current pair of 8800GTX's in my other computer only throws down 5000-6000ppd(not counting CPU contributions) with a mild overclock... WTF? LOL

Not that this computer is new by any means, but these 8800GTX's sure as hell werent a $60 card when i bought them, its sickening to think that a new $60 card can be comparable to a pair of ~$400 cards(when they were brand new!)

That's just the way things go, with GPUs advancing so rapidly. I have two GTX 275s ($400 new, 7000 PPD each) and two GTX 285s ($500 new, 8000 PPD each), and they're just about obsolete. I just bought a pair of GTX 660 Ti's link (on sale $220 new, 39000 PPD each). Two cards for the price of one of the older cards, and a big boost in performance. It's just amazing.

[GreyWhiskers]

I just wanted to reiterate what I said about the number of CUDA cores and the implication on processing performance.

From theWikipedia Comparison of Nvidia graphics processing units:

The GT430 has 96 Cuda cores and a rating of 268.8 GFlops
The GT520 and the GT610 both have 48 CUDA cores and a rating of 155.5 GFlops

My experience is with the GT430, and as far as I've found, it is the highest performing PCI-non Express card I could find. Go Figure. It's the one that overclocks so well, and produces the ppd ratings I listed. I would expect the 520 and 610 would have proportionally less ppd considering the fewer cores, even if they have a bit higher factory clock rate. The 520 and 610 probably could be overclocked too, but not to make up for half the CUDA cores.

[bruce]

Wow, thats insane for a 'crappy' little PCI card!

If a ~$60 card (just checked a GT610 and it was $59) can pump out ~6000ppd, thats definitely the way to go. especially since the CPU itself could never compare to that!

But, now that worries me even more...

My current pair of 8800GTX's in my other computer only throws down 5000-6000ppd(not counting CPU contributions) with a mild overclock... WTF? LOL

Not that this computer is new by any means, but these 8800GTX's sure as hell werent a $60 card when i bought them, its sickening to think that a new $60 card can be comparable to a pair of ~$400 cards(when they were brand new!)

Or, do i have some sort of issue... Ive never folded with these 8800's so i dont know if this is just the norm, or...

Moore's law in action. Remember that the 8800 GTX was introduced in 2006. (That's really old in "dog years") The GT 610 was introduced in 2001 when it was called a GT 520. Besides the initial price difference, the power consumption goes down from 155W to 29W so it saves you on your electric bill and pays for the upgrade in a very short time.
... but we're way off topic.

[PSI_Performance]

It still sucks that a $60 card can blow the doors off of something 10x the price LOL

But, this has also got me thinking...

Is it possible to run multiple PCI GPUs and fold with them all?

For a $60 card, if it can lay down 5k ppd, why not throw 4 of them in for a $240 machine that can throw down 20k ppd with just the GPU's?

Aside from saving the wear and tear on my 'decent' machine, the old girl would have substantially less power consumption

[Nathan_P]

AFAIK its never been tried, i would have 3 concerns.

1. The cpu might not be able to feed 4 gpu's
2. The mobo might not be able to prove enough power to 4 gpu's in PCI slots - remeber PCI was never designed with multiple graphics cards in mind.
3. The PCI bus might not have enough bandwidth.

If you do want to go this road - and it does sound like an interesting idea, i would start with 2 cards and go from there

[TonyStewart14]

I have a 430 running with a Pentium D and the 430 gets as high as 10x the points (6000 vs 600). I run SMP to make the main system fan go faster and cool the GPU about 3 degrees C, which is big when you go down from 99-100 to 96-97.