I've got budget to replace one of my gpugrid cards. Are the "lower end" Titans a viable card for gpugrid tasks?

Currently the 780ti is an excellent choice for this project. These cards crunching GPUGRID tasks, given all things being equal and the same (performance tasks at other projects may differ), will out perform a TITAN, however the TITAN Black editions will outperform the 780ti. Given the cost and performance difference between the 780ti and the TITAN Black, the 780ti would be the better choice.

Agreed. The cost-benefit ratio for the 780Ti is much better than for either Titan.

No - get 780Tis instead.

Matt

my box has a single gpu slot.

Should I get one slightly used 690 or one new 780ti?

my box has a single gpu slot.

Should I get one slightly used 690 or one new 780ti?

According to this:
http://www.gpugrid.net/forum_thread.php?id=2507
... a GTX 690 is recommended over a GTX 780 Ti.

According to these:
http://en.wikipedia.org/wiki/GeForce_600_Series
http://en.wikipedia.org/wiki/GeForce_700_Series
... a GTX 690 has 2x1536 (3072) cores operating at 1019-1058 Boost Mhz at 300 Watts TDP, whereas a GTX 780 Ti has 2880 cores operating at ~928 Boost Mhz at 250 Watts TDP.

For GPUGrid purposes, if you're looking for performance, the GTX 690 is the better choice I believe, even if slightly used.

Edit: There may be other factors involved, instead of just shader core count and operating Mhz, but if those factors would have made a difference, the FAQ would have indicated it. I just tend to focus on shader core count and operating Mhz.

my box has a single gpu slot.

Should I get one slightly used 690 or one new 780ti?

I would rather choose the GTX780Ti than the GTX690, as the cooling of the GTX690 emits half of the heat inside the PC's case.
While the GTX690's PPD (aka RAC) is slightly higher than the 780Ti's, the GTX690 processes two workunits simultaneously, so the GTX780Ti finishes a single wu nearly half time as the GTX690 does, which makes the GTX780Ti more futureproof.

interesting, Retvari,

How is the 780Ti more "future proof"?

Shouldn't the 690 process approximately twice as many WU's per unit of time than a single 780Ti?

Teach me, please!

As stated... a GTX 690 has 2x1536 (3072) cores operating at 1019-1058 Boost Mhz at 300 Watts TDP, whereas a GTX 780 Ti has 2880 cores operating at ~928 Boost Mhz at 250 Watts TDP.

So... from what I gather:
The GTX 690 is basically like 2 GPUs, each with 1536 cores.
The GTX 780 Ti is basically like 1 GPU, with 2880 cores.

So...

The GTX 690 will do 2 work units at the same time, and get them both done in like 12 hours. So, each work unit takes 12 hours, but because it got 2 done in that time, it processes at an overall rate of 6 hours per work unit.

The GTX 780 Ti will only do 1 work unit at a time, and will get one done in something like 7 hours; it processes at a rate of 7 hours per work unit.

Note: These numbers are made-up; they're just demonstrating how the GTX 690 does work in parallel (getting a higher overall throughput), whereas the GTX 780 Ti does work in serial (getting a single work unit done faster, but at a slower overall throughput).

The GTX 690's throughput is not twice the GTX 780 Ti; it's likely only a bit better.

Want more made-up numbers?
GTX 690: 3072cores * 1058Mhz = 3,250,176
GTX 780Ti: 2880cores * 928Mhz = 2,672,640
... so, the GTX 690 has ((3,250,176 / 2,672,640) - 1) = 21.6% more throughput, theoretically.

one should always run 2 WUs per card to take advantage of 10% idle time while CPU is doing something.
Also you want to downclock memory as much as possible for GPUGRID since it's not memory intensive, so the card can boost it's base clock without exceeding power limit.
Example: 1 WU on my older titan uses 95% base power at 1058Mhz core, 6000Mhz memory.
2 WUs hit power limit and use about 99% at 1032Mhz core (but still 2 WUs at the same time finish faster than if they are processed serially).
2 WUs with memory downclocked to 5000Mhz use 97% power at 1058core - and finish even faster than in 1032/6000 case.

I do have some EVGA titan with 868 base core that goes up to 1058 core if within power/temp specs.

Why not just raise the Power Target %, and then force Max Boost (using nVidiaInspector; I posted a thread about it)? That's what I do on my Keplers, to keep them running full steam.

When I last checked actual performances the GTX690 does about 3.5% more work than a GTX780Ti. However, that might have changed slightly.

What to get really depends on the purchase costs. You might be able to get a good deal for a second hand GTX690, while a 780Ti is likely to be more expensive.

However, in theory I would narrowly agree with Zoltan on this one for a few reasons:
The GTX780Ti is new revisions of the GK110 while the GTX690 has two original-ish GK104 based GPU cores. The GK110's are somewhat more efficient.
The cooling, as already mentioned is a bit better, and you can buy bespoke GTX780Ti's.
The GTX780Ti boosts higher, is designed for single precision (unlike the black ed), has 3GB DDR5 on it's one GPU core, is higher clocked and due to all of the above uses less electric than the combined GTX690 GPUs (about 8% more efficient).
One GTX780Ti being a single GPU is also likely to have a lower failure rate and because it's a newer card is likely to last longer.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Ken,

I have two 780Ti cards running in Win7. One maxes out at 1124MHz and the other at 1137MHz.

Here are some recent run-times.

I also typically run 5 World Community Grid tasks at a time while doing GPUGrid, so that slows my GPU times just a little bit (reduces my GPU utilization by about 3% - 5%).

Thanks to all!

Recently I have made new PC for gpugrid and I have tried 4x 750ti:

- with longruns only credit is about 900k daily
- power consumption with full load is 300w

Your systems average credit takes time to build up. It's potential is around 1.1M.
Good going for a 300W system!
For comparison a system with 3 GTX660Ti's would use about 150W more and do a bit less work. A dual GTX 780 Ti system would also cost more to build.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

A dual GTX 780 Ti system would also cost more to build.

And a lot more energy :)

---

Greetings from TJ

Recently I have made new PC for gpugrid and I have tried 4x 750ti:

- with longruns only credit is about 900k daily
- power consumption with full load is 300w

This is very good performance for that power consumption.
However, I think these cards don't have their own PCIe power connector, so all of the current needed by them is going through the motherboard's PCIe connectors. Therefore it is highly recommended to use such motherboard which has extra power connectors for the PCIe cards, and connect those to the PSU. If you don't use such motherboard, then the two 12V pins on the 24-pin ATX power connectors feed the 20A (240W) needed by these four cards (plus the motherboard's components and the RAM). This is too much for those two pins, and they gonna burn (not in flames, hopefully). I had such experiences with one of my cruncher PC equipped with a GTX690 and a GTX680.

However, I think these cards don't have their own PCIe power connector, so all of the current needed by them is going through the motherboard's PCIe connectors.

They have their own PCI 6 pin connector. I think it's imposible to have 4 pci card with 60W power in one mobo without own power. Even when board has special 12V+ connector for PCI (my Gigabyte Z97X-SOC has one).

Many GTX750Ti's have a 6-pin power connector,



A long time ago I had four GT240's in the one system (69W TDP each). The GPU's didn't have their own Power connector. The motherboard had an additional power connector (might have been Molex, but some have a 6-pin PCIE connector and or an extra 8-pin power connector),


Look just above the top PCIE slot, and just above the CPU.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help