POV GTX680-SLI - 91c Temp GPU#1

[OptimaLnrg]

Looks pretty much to be what you'd expect with a non-ref cooler at the top. It's a combination of having to suck air through that tiny gap between the cards coupled with the way in which the cooler works, it just tufts the hot air all over the place and relies on case airflow to expel it.

 

[mmj_uk]

I don't really know what to suggest apart from trying to improve case cooling, this is presumably the reason why AMD/NVidia still use loud blower type coolers on their reference cards as it ejects all of the hot air outside of the case instead of letting it build up inside.

 

[Kaapstad]

However, the issue remains that on testing them again (installing POV GTX680 on it's own) there is still a discrepancy in the clocks from one another as posted before. I ran Heaven twice and then Furmark to monitor the vitals as follows again (spent a few hours testing these GPUs ):


[Test 1] Single-GPU#1 installed only in first PCI-E 2.0 x16 slot (plenty of room for air)
GPU#1 - 57c max, Core: 1189MHz, Shader: 2378MHz, Fan: 83% (auto), Voltage: 1.175

[Test 2] Single-GPU#2 installed only in first PCI-E 2.0 x16 slot (plenty of room for air)
GPU#1 - 58c max, Core: 1215MHz, Shader: 2430MHz, Fan: 83% (auto), Voltage: 1.175
No thermal throttling so why the lower clocks on the first GPU? Must be due to BIOS being incorrect and not being tested to verify, POV-TGT fault there maybe. They should be identical because I bought them together. And is it right that the BIOS is different too from one-another?

Its the way it works no 2 gpus are the same. The boost works in multipules of 13mhz. Thats why one of your cards is 26mhz above the other 2 x 13mhz. Providing you are getting more boost than POV claim for the cards it just means one GPU is better than the other.

I get the same thing on my GTX690s in SLI.

I used the following settings the other day overclocking.

GPU 1023mhz Mem 1751mhz Boost 1127mhz

This is what I got for boost

Card 1 GPU 1 Boost 1176mhz GPU 2 Boost 1163mhz

Card 2 GPU 1 Boost 1163mhz GPU 2 Boost 1176mhz

What was interesting was the boost was different for each GPU on the same card.

The other interesting thing was Card 1 GPU 1 which had the highest boost also had the highest temp @91c the other temps were in the low - mid 80s

I also get a difference in boost running the cards at stock speeds.

The only thing I find strange with your cards is the difference in boost is as high as 26mhz rather than the 13mhz I get with mine.

The bottom line is no 2 cards overclock or boost the same

 

[FMTopfan]

Thanks Kaapstad for your feedback as it was interesting to note. Things must have always been this way for Nvidia, way back my old reference 4870s were the exact same clocks as one another. Normally one could assume that one would overclock better than another but to have different boosts at defaults is odd.

This bit of the fudzilla review explains how it should result in nice auto-overclocking and the low-leakage/low-temps was a selling point over the others (see link for more):

Spoiler

TGT handpicked only the best chips with low leakage, which should result in top auto-overclocking results. Thermals are also important and GTX 680 Ultra Charged’s cooler keeps GPU temperatures below 70°C – as much as 10°C lower than on the reference GTX 680.

GTX 680 Ultra Charged’s Base clock is 1110MHz, Boost is 1176 MHz. As you can see, clocks are up to 1228MHz. It appears as if TGT managed to nail GPU Boost to the highest possible auto-OC clock.

[ link ]

Could you clarify if the shader is normamly boost too as there is also a +52MHz difference on my GPU#2 (2378 vs 2430)?

I will clock the 680 to same default clocks as SC Signature with no change in volts and see what that throws up, hopefully no crashes!

 

[Kaapstad]

Could you clarify if the shader is normamly boost too as there is also a +52MHz difference on my GPU#2 (2378 vs 2430)?

If I remember right shader is twice GPU speed so if GPU is boosting to 1215mhz then shader = 2430

 

[Aurhinius]

Both my EVGA's boost differently as I said.

One will voltage throttle before the other usually dropping down from 1.175 to 1.162 while the other carries on at 1.175.

I will test tonight and post the difference in boost clocks.

Nvidia have said the boost clock is based on several parameters some of which they have not told us about to date including type of workload!

It wouldn't surprise me to find that SLI does produce a slightly different workload on one of the two cards causing this difference.

It can also just be the difference of better silicon though.

 

[ghutchison14078]

Those type of coolers dump all the hot ain inside the case....hot air rises...top card gets much warmer!!!

A side case fan to extract the hot air around the GPUs would defo help!

I noticed a bit of an increase in temp of other components when I added my KFA2 GTX 480 Anarchy (same type of cooler as your 680's)

 

[Kaapstad]

Nvidia have said the boost clock is based on several parameters some of which they have not told us about to date including type of workload!

I think one type of workload is some of the benchmark tests which involve running hugh amounts of frames a second. The 3D graphics part of Performance Test 7 is quite interesting. It will not run SLI so it only uses one GPU. The first test is a load of balls bouncing around. When the test starts and for the first 10 seconds or so it runs at about 2000 fps then about halfway through the test you can see the GPU throttle back to about 1000 to 1200fps. In the later tests which need more GPU horsepower per frame and therefore run at lower frame rates there is no throttling back.

 

[Aurhinius]

Here is some data from EVGA precision running a stress test on both GPU's

GPU 1 GPU 2
1175.806 1215

Both run at 1.162v
Both at 70% usage
GPU 2 uses 10% more power due to higher clock.
Both GPU's at 70°c


GPU 2 is just better silicon no matter what I do the result is always the same.
GPU 1 will not boost to GPU 2 speeds

 

[Aurhinius]

de-synced the GPU's in precision and found I can force GPU 1 to match GPU 2 speeds by forcing this

GPU 1 boost 153 mhz
GPU 2 boost 115 mhz

This causes both GPU's to boost themselves in sync in terms of GPU speed.

 

[FMTopfan]

A little update on hours of tests to find the ideal location to place fans for SLI whilst maintaining the best temps. I have temps for GPU#1 down to a maximum of 71c (idle=31c) and GPU#2 at 60c (idle=28) at room temp maintained constantly between 22-23c.

This is with a slight overclock of 3% of the memory as I wanted to maintain some performance but not at the cost of too much heat.

Placing the high-rpm fans near the GPUs made sense and results from running Heaven 3 on maximum settings for 30 minutes (see screenshot below) resulted in what I would consider acceptable temperatures due to the improved case cooling. The voltage of GPU#1 did drop to 1.162v once it hit 70c.

The top three 5.25" drive covers are exposed and I believe allows either cool air to enter or raising hot air to escape, placing my hand a few inches inside the case has a nice cool touch to it. Changing the exhaust fan rear of the case to push cool air in probably explains why the interior is much cooler, exhausting the heat from the GPU via the quickest route which is the custom side-panel. Avoids recycling too much of the hot air around the case at the height of this issue. Has dropped motherboard temps from 40c to 30c again.

Previous GPU#1+GPU#2 temps: 91/60c
Current GPU#1+GPU#2 temps: 71/60c - a 22% drop!

Aurhininius; if I synced the identical GPUs speeds exactly what benefits will that bring? Only example I can think of this late hour is screen-tearing as the slower card tries produce the same framerates in SLI.

Talking of performance, here is my official Heaven run pushing these cards as near to the max as I had time for and surprisingly the temps were quite good bar the fans being 100%!!

Spoiler

GPU#1 - POV-TGT GTX680 Ultra Charged (Core: 1215MHz, Boost: 2430MHz, Mem: 1702MHz, Power: 112%, Voltage: 1.175, Temp: 70c@100% fan)
GPU#1 - POV-TGT GTX680 Ultra Charged (Core: 1233MHz, Boost: 2471MHz, Mem: 1702MHz, Power: 112%, Voltage: 1.175, Temp: 58c@100% fan))

CPU - 4.8GHz, 1.42vcore - validation

GPU-Z validation

Run #1 GTX680-SLI (all as above)
Score: 3908 - 1680x1050

Link to full-size

Run #2 GTX680-SLI (all as above)
Score: 3585 - 1920x1080

Link to full-size

Run #3 - Single GTX680 defaults clocks (Core: 1189/Boost: 2378/Mem: 1502, PCI-E 2.0 x16)
Score: 2203 - 1680x1050

Link to full-size