Overclocking Results, Voltage Limits
Despite what I would describe as complete overkill cooling for a power-efficient GPU like the GM204 used in the GTX 980, I was a bit let down by the total maximum overclock I was able to reach.
Click to Enlarge
Using a reference card as our starting point, and knowing that this card was able to reach a 225 MHz offset using the stock air cooling, reaching a maximum offset of 280 MHz is lower than expected. This was accomplished with the maximum allowed software voltage offset of +87 mV as well – there was no room for us to go any higher.
I asked around at both EVGA and at NVIDIA, looking for ways to increase the voltage over this amount; anything, something that would give us the ability to really push this GPU further. Unfortunately, at least for now, the only way to get higher voltages on the GTX 980 is to hardware modify the board – something I just wasn't willing to try yet.
Click to Enlarge
There is some good news though; even though the offset target was only 55 MHz higher with the EVGA Hydro Copper than with the reference air cooler, we were able to achieve more consistent and faster clock speeds. The green line up at the top is the GTX 980 with the custom water block – it is clearly the fastest card of the bunch and never wavers from that top clock speed.
Click to Enlarge
Our average clock speed over that time period was 1582 MHz, exactly 100 MHz faster than the same card using the reference cooler. That is a 6.7% increase in clock speed over our previous top overclock and 30% faster than the card tested without any offset changes. That's a decent result but not nearly as high as I was expecting to go. Nor is it as high as I think the GPU can realistically be pushed.
Click to Enlarge
These temperature results indicate why I think this is the case. With maximum allowable voltage we were only hitting 42C with this GPU under a full load. The EVGA Hydro Copper water block is truly a beast and combined with our 480mm total of radiators, the GM204 is running at about half of the temperature that it is targeted to run at. That is in no way a small feat!
But if you combine the information provided here, 40C temperatures, clock speeds nearly flat at 1583 MHz, a software-limited voltage regulation, you'll easily see the limiting factor. Voltage. WE NEED MORE! For now, it looks like NVIDIA is locking down the voltage adjustment on the GTX 980 like it did with the GTX 680 at its launch. At that time it took pressure from board vendors and from online media to force NVIDIA to be more open with voltage adjustment. Looks like we may have to do the same this time around if cards like the EVGA Classified and KingPin editions are to see store shelves.
Performance Result
Since our clock speeds are only 100 MHz higher than the overclocked results in our initial air-cooled overclock, we have a good idea of what to expect.
Average frame rates in Metro: Last Light at 2560×1440 and the highest quality presets jump from about 58 FPS to 68 FPS, an increase of 17%.
Pricing and Availability
Pricing and availability haven't really been set by EVGA quite yet, but based on feedback from the company we expect the price of the Hydro Copper water block on its own to be around the $150 mark. Shipping dates are ranging from mid- to late-October at this point. We know that EK is going to be selling a water block that is basically identical to this minus the EVGA branding (and maybe the LEDs, I'm not sure) and pre-orders for that are in the $140 range.
You can also expect to see an EVGA GTX 980 card sold with the water block pre-installed within a similar time frame. I am curious to see how EVGA chooses to address the issue of voltages though if they sell a complete card – we have seen that the current software solutions for voltage adjustment just don't provide enough flexibility to really push the GM204 at these super-low temperatures. Will EVGA wait for NVIDIA to give the go ahead on higher voltages before release the Hydro Copper GTX 980 graphics card? Hard to say; but I would think they would HAVE to in order to capitalize on the potential.
Early Thoughts
First, our thanks goes to EVGA for letting us steal this early prototype of the GTX 980 Hydro Copper water block. It was a fun couple of days getting it setup and running a couple of our GTX 980 cards through the paces under water cooling. Obviously I am a bit disappointed in the end result we saw – only a 100 MHz clock speed advantage over air cooling – but it is not really the fault of the EK water block.
The cooling performance of the block coupled with our 360mm+ radiator setup produced incredibly low temperatures on the GPU – 42C under a sustained gaming load. Though I can't prove it quite yet, I think it's more than fair to say there is a TON of headroom left in GM204 and we should be seeing much higher clock speeds with some additional voltage flexibility. (And if KingPin's results are in any way an indication, we should see a LOT of clock headroom.)
What effect do these
What effect do these adjustments have on power consumption ?
Ah, I did mean to put that
Ah, I did mean to put that in. We are only looking at 50-60 watts more power.
You definitely need to test
You definitely need to test power draw again. Cooler transistors run more efficiently, and 70C+ to 40C is a big drop. I had a first gen Radeon 7970 drop 30 watts when I put a water block on it.
Where are you getting this
Where are you getting this info from? The difference between 70C and 40C is nothing when comparing efficienty. Dift velocity doesnt alter much from a 30C change. Putting a water block also increases consumption as it can stably boost to higher clock rates with out the need of user overclocking.
Leakage current increases
Leakage current increases with increasing temperature – this is a well known phenomenon and the decrease of power consumption as temperature is reduced has consistency been observed with liquid cooled GPUs.
For a systematic analysis of this, I refer you to this presentation from GTC earlier this year:
http://on-demand.gputechconf.com/gtc/2014/presentations/S4484-how-temperature-affects-gpu-performance.pdf
what an absolute fantasic
what an absolute fantasic card, nVidia really outdone themselves with this, evga has pulled it out the bag with there cooler also. Really making me want to upgrade my system reading the review and watching the video, keep up the good work guys at pcper
Just an FYI they purposely
Just an FYI they purposely cheery pic very best cards to send to reviewers.
Can’t say that isn’t true for
Can't say that isn't true for sure, but I did test two different cards with same results.
i didn’t realize you
i didn’t realize you installed the block thought they sent you a hydro 980
Nope, the Hydro 980 is not
Nope, the Hydro 980 is not even a thing yet, considering we were testing the only block in existance and had to install it on a 'regular' EVGA 980 to test it.
Also for those interested in
Also for those interested in how kingpin achieved these results you can read his “uncorking the gtx 980” guide here : http://kingpincooling.com/forum/showthread.php?t=2886
What about memory over
What about memory over clocks?
I could not care about
I could not care about pushing the overclockers even higher i have a small room that gets very hot and i care much more about the temperatures so 42c is very helpful
If you overclock it as Ryan
If you overclock it as Ryan did, it is still going to put out 225 W worth of heat into your room, even though the gpu stays at 42 degrees.
“individually sealed coolers
“individually sealed coolers for processors and even graphics cards that offered some of the benefits of standard water cooling (noise, performance)” That’s not really true, closed loop watercoolers still haven’t overtaken good air coolers. When you take noise into account, flagship closed loops offer equal or somewhat worse performance compared to flagship air coolers while costing more, and on lower budgets, there’s absolutely no contest. It’s almost impressive that Asetek has managed to make watercooling that’s no better than air.
Even if the maximum clock speed only increased by 100mhz, I wonder if performance would see a slightly larger increase due to clock speeds being more stable at their maximum. 87mv is pretty disappointing, my AMD 7950 gave 207mv of room out of the box. Still, 100mhz from watercooling isn’t a bad jump. I wonder how much voltage you can safely give the GPU, assuming temperatures remain low and power delivery is up to the task. There isn’t much information out there about voltage-related degradation, but it is often the limiting factor when using custom watercooling.
“That’s not really true,
“That’s not really true, closed loop watercoolers still haven’t overtaken good air coolers. When you take noise into account, flagship closed loops offer equal or somewhat worse performance compared to flagship air coolers while costing more, and on lower budgets, there’s absolutely no contest.”
For cpus, there isn’t too much of a reason to water cool since you usually have plenty of space and heat pipe based coolers compete quite well with water coolers. This wasn’t the case before heat pipes started to get use in cpu coolers; I think almost all high-end air coolers currently use heat pipes. At this point, it may be a lot easier to install a tiny water block than a giant air cooler though.
For gpus air cooling does not do as well; the space is obviously constrained. You do not have 140 mm of space above the gpu for a giant air cooler. You have maybe 20 mm of space per slot. It is amazing that they have been able to cool more than 250 W with such low-profile coolers. Water cooling gpus makes a lot of sense, although it will still be expensive. You also have the case of multiple gpu configurations where you may have two 250 W or more cards a few millimeters apart. This doesn’t happen with cpus.
I agree with your statement about performance, but only for cpus, not gpus. As far a price is concerned, if you are going to spend $600 on a high-end video card, then a little extra for water cooing isn’t out of the question.
According to kingpin, “safe”
According to kingpin, “safe” voltages on LN2 can go up to 1.6v so there should be plenty of margin from 1.25v even though the max voltage you would apply with watercooling could be around 1.4v
I can’t believe that EVGA
I can’t believe that EVGA left you guys the prototype cooler for testing! Way to go boys!!
Does this waterblock support
Does this waterblock support the EK-FC Terminal? It looks like it does… can you confirm?
How about Ares III review
How about Ares III review sometime soon?
The Metro Last Light
The Metro Last Light Frequency Comparison chart (https://pcper.com/files/imagecache/article_max_width/review/2014-10-01/clockspeedevgawater-avg.jpg) seems to have gotten mislabelled.
Blue is labelled GTX980 + 125%PT instead of 980GTX stock
Yellow is labelled GTX980 + 125%PT + 225MHz, instead of just GTX980 + 125%PT.
power won’t be a problem on
power won’t be a problem on my 980 KPE’s….. hurry up with the blocks for this EKWB. :drooling:
SS