7950 VRM temps - what should they be?

[Angilion]

I've recently bought a 2nd hand 7950. Performance in benchmarks at stock is well over double the 4870 1GB I had before, but of course I wanted to see how it would clock. More is...well, more I've been using it at stock because that's plenty, but I did some benchmarking with various overclocks just to see.

It's 930/1275 stock, but went easily to 1000/1500 +20% power. GPU was 75C under constant 100% load. No worries - they're specced for 85C. I tried some different GPU and memory clocks to see what effect it had. Then I switched to a different hardware monitoring program and saw the VRM temps. 97C with the GPU at 950 and memory at 1500 (and GPU at 71C) with +20% power limit.

That seems far too high to me. I wasn't expecting them to be so much hotter than the GPU.

So I benched again at stock and the VRM temps were still 77C (GPU was 64C).

I'm using HWINFO64. Are the VRM temps accurate? What's normal for VRM temp on a 7950? Surely not 97C. Is it normal for them to run so much hotter than the GPU? I'm thinking that maybe the cooler on my card doesn't do much to cool the VRM (it's a VTX3D X-Edition).

I'm content running it at stock and will continue to do so, but 1000/1500 +20% power gives me a 45% increase in performance in Heaven, compared with stock, and it's only the VRM temps that are stopping me running it at that speed.

 

[Angilion]

I solved the problem, so I'll post in case it's relevant to anyone else.

I just used Afterburner to reduce voltages. GPU down to max 1.05V (from max 1.15V), VRAM down to 1.50V (from 1.60V).

Despite the ambient temperatue being 7C higher than when I got the above results, GPU temp was 12C lower and VRM temps were 19C lower. I was hoping for some reduction, but that's brilliant.

I also ran it at stock (930/1250) with +20% power limit and dropped another 8C on the GPU and another 5C on the VRMs with performance only ~8% lower than I was previously getting at 1000/1500 with +20% power limit. The reduced voltages are clearly reducing power limit throttling, as they should. Looking at load graphs shows some throttling even with +20% and stock speeds with increasing throttling with increasing overclocking, which explains the relatively small gains with overclocking. So stock speed it is - overclocking increases temps a lot more than it increases performance. For now, anyway. Maybe at some point I'll reduce GPU voltage some more and see what happens.

Clearly, my card was effectively over-volted out the box for some reason.

EDIT: I know that the VRMs are rated to at least 115C, but I think VRMs temps are still an issue well below that because they will be heating the PCB rather a lot.

 

[Angilion]

some good info there, and something that i will check if i oc

You might find it useful for increasing performance even if you don't overclock.

These cards are automatically throttled on the basis of power drawn but you won't see it even if you're monitoring the clock speeds. If you want to check, you need to monitor GPU load and run something that will put it under 100% load. If it's being throttled, you'll see the GPU load graph as a jagged line rather than the flat line it should be.

The change in performance can be quite large - in my testing, I could vary the average FPS in Heaven between ~40 and ~54 solely by varying the amount of power drawn and the throttling limit. Everything else remained the same - the benchmark settings, all clock speeds, everything. All I varied were the voltages (which reduces the power drawn) and the throttling limit (which has the biggest effect).

These figures are all at stock GPU and VRAM speeds (930/1250), all with the same benchmark settings and all with everything else (CPU, etc) the same):

Out the box (1.6V VRAM, 1.15V GPU): 39.8
+10% power limit, 1.6V VRAM, 1.15V GPU: 49.6
+20% power limit, 1.5V VRAM, 1.05V GPU: 53.2
+20% power limit, 1.5V VRAM, 1.0V GPU: 54.2

and it's still being throttled for power consumption, just much less.

Also useful is that despite the large increase in performance it's running cooler. I'm getting the same performance as I was with a sizable overclock but lower temps than I was getting at stock out of the box. The reduced voltage makes overclocking pretty much impossible (it'll fall over at 970 now, whereas it was fine at 1100 with stock volts), but this is a much more efficient way to increase performance.

Reducing throttling has a much bigger performance improvement than overclocking with a much smaller increase in temps even if you don't reduce the voltages. Earlier on, I tried overclocking to AMD limits - 1100 GPU, 1575 VRAM - with stock volts and no power limit increase. Same benchmark settings, same everything else. FPS was 45.1, i.e. that much of an overclock gave only half the performance increase of simply increasing the throttling power limit by 10% while remaining at stock speeds but raised temps far more.

The only issue I have with the card is that I don't like the tone of the cooler at 20% fan speed (which I use at the low power state, when it's running at 300MHz for 2D). I think it's vibrating - it's a long card with a plastic shroud and there's nothing stopping it vibrating. Maybe I'll rig up some kind of card support. Or maybe I'll just get used to it Oh, and for some reason it always starts at 55% and will stay there until I open the fan profile settings in Afterburner and click OK (without changing anything - I've set the fan profile as I want it).