The NVIDIA GeForce GTX 980 Ti is launching in two days and we have already seen how the reference cards look and perform. The main feat of the GeForce GTX 980 Ti is going to be the non-reference lineup which is going to be launched by NVIDIA’s AIB partners and will include several high-end coolers featuring triple-fan, hybrid and liquid coolers.


http://wccftech.com/nvidia-aib-partners-prepare-huge-geforce-gtx-980-ti-nonreference-lineup-hybrid-triple-fan-cards-unveiled/

https://twitter.com/TEAM_CSF/status/605247234679128064
____________
[CSF] Thomas H.V. Dupont
Founder of the team CRUNCHERS SANS FRONTIERES 2.0
www.crunchersansfrontieres

Ladies and Gentlemen, Boys and Girls,
May I present the fastest GERARD_FXCXCL12 ever processed to you:

e1s5_2-GERARD_FXCXCL12_LIG_13185441-0-1-RND4038_1
It took only 19517 seconds (that is 5h 25m 17s) to process on a Gigabyte GV-N98TG1 GAMING-6GD. Too bad that I'm only testing this card :)

e2s7_e1s6f207-GERARD_FXCXCL12_LIG_11680461-0-1-RND0759_0
19496 seconds (5h 24m 29s)

The specs:
GPU clock: 1447MHz (+80MHz set by MSI Afterburner)
GPU voltage: 1.230V (+38mV set by MSI Afterburner)
GPU Temp: 62°C (max) ambient temp: 20°C (the PC is in the open air)
GPU fan speed: 70%, 2936rpm (Windforce 3x)
GPU memory clock: 3505MHz (set by NVidia inspector)
GPU usage: 96%
GPU FB usage: 28% (@3505MHz)
GPU power: 85% (267W)
Total system power draw (measured at the 230V 50Hz wall outlet): 1 GPU task running: 350W idle: 83W
Power Supply: Corsair AX1200 (80 Plus Gold)
CPU: Core i7-4790k @4.00GHz
PCIe: 3.0@16x
MB: Gigabyte GA-Z87X-OC
no other GPUs were present.
no other tasks were running.
OS: Windows XP x64
NVidia Driver: 353.30
Comparing to a GTX 980 (1415Mhz@1.218V & 3505Mhz):
40% more CUDA cores (2816 vs 2048)
~32.3% better performance (~19500s vs ~25800s)
~50% more costs.
+34.6% total system power draw (260W vs 350W)
+46.7% GPU power draw (182W vs 267W) (total system power draw with GTX980 idle: 78W)

This 1447MHz@1.230V seems to be the maximum setting, as I had a task which errored out after 543 seconds.

The scaling of these cards are a bit lower than a direct ratio, so there must be some bottleneck present in the system (apart from the WDDM, as I'm using Windows XP x64). It could be the application (as it's CUDA 6.5), the GPU, or the PCIe bus, or a combination of them.

Awesome Retvari!
Thanks for the heads-up :)
____________
[CSF] Thomas H.V. Dupont
Founder of the team CRUNCHERS SANS FRONTIERES 2.0
www.crunchersansfrontieres

40% more CUDA cores (2816 vs 2048)

This 1447MHz@1.230V seems to be the maximum setting, as I had a task which errored out after 543 seconds.

The scaling of these cards are a bit lower than a direct ratio, so there must be some bottleneck present in the system (apart from the WDDM, as I'm using Windows XP x64). It could be the application (as it's CUDA 6.5), the GPU, or the PCIe bus, or a combination of them.

Nice, complete report, almost 1 Mppd wow!

Thanks Retvari for sharing these results !

Your report is very detailed, but I could not find if you were running one task or two (or three) simultaneously ?

Based on previous reports :
- 5pot's 980ti yield 25-27k sec / GERARD, on Win 8.1 and i7-4770k - 81% GPU usage
- localizer-s 980ti yields 25-27k sec / GERARD, Win 7 and i7-4770k - 79-80% GPU usage
- petebe's 980tis yield 22-26k sec / GERARD, Win XP and i7-4770k, unknown GPU usage
... but yours yields 96% GPU usage ?

...I could not find if you were running one task or two (or three) simultaneously ?

- petebe's 980tis yield 22-26k sec / GERARD, Win XP and i7-4770k, unknown GPU usage

... but yours yields 96% GPU usage ?

From the huge difference of his and my runtimes I came to the conclusion that one should not reduce any bandwidth to achieve full utilization of the GTX980Ti. This GPU is simply too big and too fast.

My 980Ti achieved over 1 million points yesterday. I've not been able to do that since I was running with two 780Ti and two 680s. It's a lot more power-efficient as well with current tasks only taking me up to about 80% power with a small overclock.

Last week, one of my GTX 690 cards failed in my windows 7 computer (the other one is still working), so I bought a GTX 980 Ti, and experimented. Here are my findings. The motherboard on my computer has the following slots: 1 x PCI Express x16 slot (PCI_E2) with x16 operation and 1 x PCI Express x16 slot (PCI_E3) with x4 operation. When I installed the 980 Ti into the x16 slot, a GERARD_FXCXCL12_LIG_ units would finish in a little over 9 hours with 70% usage, but in the x4 slot it would finish in just under 16 hours with 55% usage. This compares with 690 card which would finish 2 GERARDS units (one in each GPU) in over 16 hours in either slot with usage in the mid 80’s %. The temperature for the 980 Ti is in the mid 50’s C, with fan setting at 70%, while the 690 is in the low to mid 70’s C, with fan setting at 90%. The 980 Ti has a better cooling system. The 980 Ti is very bandwidth sensitive.

Is there any way around this or would next GPUGRID application version improve performance?

For the record, my computer has 1 GTX 980 Ti and 1 GTX 690 card, not 3 GTX 980Ti. See link below.

https://www.gpugrid.net/show_host_detail.php?hostid=127986

Otherwise, everything is working fine.

The only problem with that solution is, it would also be running 2 WUs concurrently on each of GTX 690 GPUs, which I don't want to do because they are already running at near capacity.

There wouldn't be much gain on the GTX690, but it wouldn't hurt either. I've run 2 concurrent WUs on my older GTX660Ti, and it was OK (although of no benefit) for the larger Noelias, which yield a higher GPU utilization than the Gerards. There was a clear benefit for short runs, though. Each GPU on your GTX690 has 8/7 the shaders and 4/3 the memory bandwidth of my old GPU, so it should profit a bit more from 2 concurrent WUs.

And which "capacity" do you mean that they're running at? Heat, noise and power consumption? In that case: simply lower the power target at bit. By default the GPU runs at full throttle, at maximum turbo boost clock with voltages of up to 1.175 V. Scaling it back so that it runs at ~1.10 V yields a ~14% improvement in power efficiency from the voltage alone. At the same time the clock speed would drop by ~7%. If you'd increase the GPU utilization from 85% to 92%, you make up for that performance drop.

Net gain: 14% lower power consumption & heat produced. This way you'd run the hardware more efficiently physically and use more of its capabilities via software.

MrS

On another topic related to the 980 Ti, how does one get to run it on a Windows XP machine?

Thanks Zoltan for the help. I knew it was going to just like last time, but I didn't know which lines to copy and how to edit them, with the exception of the substitution of "NVIDIA GeForce GTX 980 Ti" for "NVIDIA GeForce GTX 960". It's better to ask, then to screw it up.

The results for the XP computer were very good, especially since it is an old model, almost 9 year old.

14543988 11192988 16 Sep 2015 | 5:35:24 UTC 16 Sep 2015 | 13:02:44 UTC Completed and validated 24,567.65 22,693.45 255,000.00 Long runs (8-12 hours on fastest card) v8.47 (cuda65)

https://www.gpugrid.net/result.php?resultid=14543988


Which is a lot better than my Windows 7 model.

14543025 11192376 15 Sep 2015 | 20:26:30 UTC 16 Sep 2015 | 6:10:17 UTC Completed and validated 32,033.23 30,958.19 255,000.00 Long runs (8-12 hours on fastest card) v8.47 (cuda65)

https://www.gpugrid.net/result.php?resultid=14543025


The maximum GPU usage is 93% for XP, and 70% for 7, per MSI afterburner.

And all I did was install the card and driver, and cleaned up the register and file clutter. Clean and easy!

The interesting thing is, when both XP and 7 computers were using the GTX 690 cards, the finish times were about the same.

I think the issue may be in the GPUGRID application itself, 900 series cards on windows 7, (I am not sure about windows 8 or 10).

Did anybody say in one of the posts to "I think you need to ditch XP for 7 or 8..."? I don't remember!

I will update when the next application fixes this slow performance issue. Maybe!

Thanks Zoltan for the help.

The interesting thing is, when both XP and 7 computers were using the GTX 690 cards, the finish times were about the same.

I think the issue may be in the GPUGRID application itself, 900 series cards on windows 7, (I am not sure about windows 8 or 10).

Did anybody say in one of the posts to "I think you need to ditch XP for 7 or 8..."? I don't remember!

I will update when the next application fixes this slow performance issue. Maybe!

How exactly can one run two work units on the same GTX 980? I used to do this on Einstein, but I do not where to find the BOINC config program for GPUGRD which allows one to set usage like on Einstein. I may be able to get slightly higher performance. Would this dual unit configuration also work on a GTX 780? That one only has 3GB RAM. GPU utilization is 80% on both cards. I have thermal headroom on both of my CPU's. Would overclocking them 300 MHz help enough to offset the diminished CPU lifespan which occurs at 10C hotter temps? My i7 3370K runs @ 75C @ 4.2 GHz @ 65% load.
____________

How exactly can one run two work units on the same GTX 980?

Works great! Now if I could get two more work units for the second GPU? Apparently there are none available to send? Thanks for the help!

Works great! Now if I could get two more work units for the second GPU? Apparently there are none available to send? Thanks for the help!
____________

Just got my 980Ti under full cover water. It is now 75% utilized and 37C when running single task at 1496Mhz. Am I maximizing the GPU power properly?

Just got my 980Ti under full cover water. It is now 75% utilized and 37C when running single task at 1496Mhz. Am I maximizing the GPU power properly?