I found a couple EVGA Geforce GTX 560TI's at my local Fry's for $240 each.

Is this a good card?

That seems inexpensive compared to how much the new 580's cost.. but then the newest stuff is always the most expensive..

Per this post: http://www.gpugrid.net/forum_thread.php?id=2373&nowrap=true#20183

it seems to imply that the GTX480 would be better than the GTX560.. Is this still true? I don't really understand why..

I always assumed the higher the number the better but is that not the case?

Or is it the x80 cards are better than the x60 cards? And whats the TI stand for?

The GTX560 Ti and GTX550 Ti are fairly average cards for crunching at GPUGrid.
The GTX580 and GTX570 are the best. Tomorrow you will be able to add the GTX590 to the top of the list. The GTX465, GTX470 and GTX480 cards are all better for crunching here than the GTX560 Ti, GTX550 Ti, GTX450 and GTS450.

The difference is in the architectures:

GTX465, GTX470 and GTX480 are Compute Capable (CC) 2.0 cards (GF100)
GTX460 is CC 2.1 (GF104)
GTS450 is CC 2.1 (GF106)
GT440, GT430 and GT420 are CC 2.1 (GF108)

GTX580 and GTX570 are CC2.0 (GF110)
GTX560 is CC 2.1 (GF114)
GTX550 is CC 2.1 (GF116)

GTX 590 is presumably CC2.0 (GF110)

All the CC2.0 cards have 32 cuda cores (shaders) per SM (core) and all the CC2.1 cards have 48shaders per SM. Unfortunately the CC2.1 cards under perform relative to the shader per SM as if they only have 32shaders per SM, making them about 33% slower per SM.

Bottom line - Get a CC2.0 card and not a CC2.1 card.

All the CC2.0 cards have 32 cuda cores (shaders) per SM (core) and all the CC2.1 cards have 48shaders per SM. Unfortunately the CC2.1 cards under perform relative to the shader per SM as if they only have 32shaders per SM, making them about 33% slower per SM.

Wow good info!

Do you think the performance difference is a software issue or a hardware one?
Meaning do you think they could fix it with some patch to the drivers or is this something that at a hardware level they just didn't do a good job?

It's fundamentally a hardware issue and so far there has been no software work around. I don't expect this to change.

A GTX470 costs the same as a GTX560Ti but will do much more work at GPUGrid.

It's more of a "hardware design choice, which doesn't work very well with our code" than a "hardware issue". Those 48 Shader blocks are not much more expensive than the 32 shader blocks and they do work with some code. It's just that the current GPU-Grid app does not beling into this group. And I wouldn't hold my breath for any changes here - the team is pretty busy doing other stuff already.

MrS

---

Scanning for our furry friends since Jan 2002

All of you are right, but imho if you plan to crunch for Gpugrid 24/7, then GTX 470/480/570 aren't a wise choice: heat/consumption are 2 great issues you should consider very well.
Considering the results, I think my GTX 560 Ti is performing very well.

---

Your GTX 560 Ti is a good GPU, but it is not especially good when it comes to GPUGrid crunching; a GTX 470 can do twice as much work. Yes it uses more electric (TDP is 215W) but a GTX 560 Ti has a TDP of 170W. So all things considered a GTX470 does more work per Watt.

The best cards in order of performance are,
GTX 590, 580, 570, 480, 470, 465.

Agreed - high power consumption doesn't automatically mean inefficient. The opposite is true here, as SK points out. Of course you have to handle (and pay for) the power consumption. But taking your argument to the extreme one could say that your GTX560Ti is not very good either, as heat/power consumption are too high. A smaller 40 W card with 1/10th the performance of yours can be handled much easier ;)

MrS

---

Scanning for our furry friends since Jan 2002

My will was to buy a pretty fast gpu card to crunch for this project, BUT with a "decent" price/heat/consumption. I wasn't looking for the better point per watt ratio. I think 560 Ti is a good compromise. At least, I hope so :)

---

It's certainly not a bad card. SKs point was just that for "only" 45 W more (26%) and similar price you could have gotten the GTX470, which would do ~100% more work at GPU-Grid.
But we're not here to talk you into hating your card.. so enjoy it anyway and wait for software, which can make better us of it :)

MrS

---

Scanning for our furry friends since Jan 2002

Your GTX 560 Ti is a good GPU, but it is not especially good when it comes to GPUGrid crunching; a GTX 470 can do twice as much work. Yes it uses more electric (TDP is 215W) but a GTX 560 Ti has a TDP of 170W. So all things considered a GTX470 does more work per Watt.

I doubt that. When an application (like GPUGrid) utilizes only 2/3 of the shaders, the GTX560Ti won't consume it's peak TDP (however it will be slightly over the 2/3 of the max TDP). I think the GTX560Ti does more work per Watt than a GTX470, while it won't do more work than a GTX470.

In the pioneer times (did we leave them yet?) ne GPU drew anywhere near it's TDP at GPU-Grid. Better hardware utilization can have changed this in recent times, however. And the relation "typical power draw" / "GPU-Grid power draw" should be similar for different cards using the same architecture.

It might be in interesting to see this measured and compared for CC 2.0 and CC 2.1 cards. The unused shaders in CC 2.1 chips are not power gated, after all, so they consume at least sub threshold leakage current. Which is less than if they were constantly switching, but is significant at TSMCs 40 nm process (that's why it's so difficult to make an idle GPU draw no power, even using power gating).

MrS

---

Scanning for our furry friends since Jan 2002

The fact that a GTX560 Ti (and other CC2.1 cards) can’t use all their shaders here is the negative aspect of the card. While there may be a tiny power saving from not using 1/3 of the shaders, this does not do much to offset the loss of performance.

The main power advantage actually comes from the architectural changes/advancements from 400 to 500series, rather than from power savings using 2/3 of the shaders and these are already accounted for in the 170W TDP.

Relative GTX560 advantages include:
The fact that the GTX470 has 2 disabled cores (still uses some power)
The GTX560 has smaller GF114 dimensions (1.95B transistors=improved heat dissipation)
Leakage reductions in the 500series over the 400series cards allow for higher shader frequencies and/or lower voltages
Circuit board improvements and relatively better heatsinks

The negative aspects of the GTX560 are:
Loss of 33% shaders but still having to power all the cores and unused shaders (to some extent)
Reduced shader density of 384 (really 256) compared to the 448 in the GTX470.
Higher clocks increases leakage

The 1/3 shaders loss is still a massive loss and not well offset by all of the energy savings when considering a performance per Watt comparison.

While the performance per Watt debate is usually important, in this case it’s somewhat moot anyway; the purchase costs for both cards are about equal and quite high. So it would take a long time to offset the purchase cost of two GTX560’s via power savings; even if the GTX560 had a better performance per Watt (which it doesn't) it would need to be considerably better. For example, the running costs of a GTX470 going flat out 24/7 (200W) for a year would be about £250. Even two GTX560’s at a mere 100W each would only match the running costs and performance (well almost). They would need to be running at 75W each for 3years to make up the initial purchase cost. In reality the running power is probably over 100W, so they would be less efficient, and you would be running at a relative loss.

It's fundamentally a hardware issue and so far there has been no software work around. I don't expect this to change.

It's a software issue. The GPUGrid app does not properly utilize the architecture of the CC 2.1 cards. Wouldn't think that it would be difficult to fix...

For me, the GTX 560 Ti is so close the the power limit my computer room can stand that the GTX 470 is out of the question, even if it does more work for each watt.