Stren's R9-290/290x Water Block Testing

[stren]

Finally some data - after 7 attempts to run the XSPC block I finally have good data and can move on to another block! Tested at +100mV, +100mV aux, PT1T no throttle bios, 1150MHz core.

Bear in mind this data is on the replacement GPU and can not be directly compared to the previous data. As always look at the axis - I keep the scale (1C/division) the same, but the X axis does not cross at zero and the crossing point moves in order to show the data better. Oh and the Y axis text needs fixing to match the chart title.



As expected the backplate makes little to no difference to core cooling. VRM cooling is a different story:



VRM1 is the inbuilt sensor near the main VRM section near the power connectors. This can be measured with GPU-Z or hwinfo. VRM2 on the other hand is near the monitor connector end of the board near the memory. It is cooler as it runs less power:



Now just to make sure there wasn't anything funny going on with VRM measurements as AC's results from before looked suspiciously good with a backplate I added a probe near the VRMs manually (see post a few up). This as expected tracks VRM1 but a few degrees off:



Conclusion - the XSPC backplate seems to make VRM cooling worse! This is not a surprise when you realize that XSPC do not use any thermal pads on the backplate unlike AquaComputer or EK for example. Despite this the VRM's are still cooled acceptably so it's not the end of the world.

So now to move on to the watercool block and then to retesting the other blocks.

[stren]

Finally something to compare the XSPC too - I got the HeatKiller block tested. WC didn't include the heatkiller backplate so I dont' have results for that, but I'm expecting that like XSPC it will make VRM temps a bit worse due to not using thermal pads (I believe). However the temps will most likely still be acceptable given the performance of the block:



Core performance is a good bit better (2C at high flow, 1C at low flow). This is very different to the Titan results where the XSPC block had the best core temps. I do want to re run this block after seeing it stutter and lag a lot which may have resulted in lower actual power usage during the test. The other test runs didn't show this anomaly. HWInfo's power usage measurement makes the test data seem real, but I'd like to be sure.

VRM 1 also is cooled well:



VRM 2 which is cooler anyway is cooled very well indeed:



The manual VRM probe also confirms there is nothing funky going on:



I think I'll run the AquaComputer block next so that I can give a bridge between earlier results on the earlier card and the new card. The new probe should also tell me what's up with AC's backplate's results.

I also have to run the EK Supremacy-VGA block, the Raystorm-VGA block and the Watercool VGA block with DIY VRM cooling kit. I'm thinking of leaving those to the end though unless there is significant demand.

[tiborrr]

Good work Simon, keep it up!

[stren]

Thanks Niko!

More results - got the AC block and backplate retested. Almost identical thermal performance to the Heatkiller but with higher flow - I'd call that a win (although we're not done yet)



VRM1 - the AC block is not so great on it's own, but the AC backplate does an exceptional job whether it's the active or passive backplate:



As the backplate does not have any thermal pads on VRM2 then we saw no real change there:



Although the heatkiller wins here all the deltas are so small that it's not a big deal.

There was some debate over whether the AC backplate was really cooling the VRMs or just a seperate temperature sensor. So as you may know I added a seperate temp sensor probe located by the main VRMs and it backs up the internal results:



Summary

If money is no object then the Aquacomputer block and active backplate provide the best performance *so far*. In reality the active part of the backplate is a marginal gain at best.

I need to retest the EK and Koolance blocks with this card and Swiftech are sending a block now that they are finally ready.

[stren]

Now with Koolance - legend order tracks the order of the lines just in case it's confusing:

[stren]

Koolance done, EK on the rig. Swiftech and Bitspower have also agreed to participate!

[fathom12]

Hi Simon,

if you don't mind, could you please give me the alternative link for the charts as the site is blocked by my ISP..
Thank you in advance.

[iggy2k]

would the heatkiller work with the AC backplate?

[Vetalar]

would the heatkiller work with the AC backplate?

well _passive_ will. but i'm not sure about mounting hardware compatibility... active obviously not

[stren]

Back from fire evacuations, work deadlines, vacations and sickness. So yeah a bit of a break sadly. BP and Swiftech blocks did arrive while I was out. During evac the motherboard got damaged and a PSU died, so I'm in the process of RMAing and buying new ones. Hoping to restart the testing soon!

Hi Simon,

if you don't mind, could you please give me the alternative link for the charts as the site is blocked by my ISP..
Thank you in advance.

PM me your email and I'll send you the image

would the heatkiller work with the AC backplate?

well _passive_ will. but i'm not sure about mounting hardware compatibility... active obviously not

Yes passive only at best, I haven't tried it, it would all be about screw compatibility (head height for the screws that go under the backplate, screw length, diameter and thread type for the ones that attach the backplate to the block) between the two. When i'm done I want to create a matrix of the blocks and screws used so people can see what should work by default. Either way you should be able to make it work by buying your own screws it might just not work out of the box! You could also try just adding thermal pads to the heatkiller backplate in the same pattern that AC uses.

[Vetalar]

what type of screws included with aquacomputer bkock? i want to play with custom backplate and i guess will need to find longer one

[LucikMucik]

When I buy AC fullblock and backplate, can I instantly mount them? Or I need longer screws and other things? Thank you for answer.

[stren]

what type of screws included with aquacomputer bkock? i want to play with custom backplate and i guess will need to find longer one

Not sure off hand - but yes you may need longer.

When I buy AC fullblock and backplate, can I instantly mount them? Or I need longer screws and other things? Thank you for answer.

Yes they will come with the screws, you only need different screws if you are trying to mix manufacturer's i.e. ek block with ac backplate.


So after buying another impact board I'm back up and running - there was one thing in the last set of tests that bugged me and I finally got to the bottom of it. Sometimes furmark would lag and fps would drop - I finally figured out it was because hwinfo was logging too much stuff. I turned off the logging of everything and limited it to only GPU logging and now it's not laggy at all. Lag wouldn't be a problem if it was repeatable, but it would come and go, and sometimes I'd have to discard data because of it and retest again and again until I got lagfree runs. Hopefully this means that data should now be more repeatable and more trustworthy!

This of course means that I need to restart testing for the third time! Yay for gpu block reviews! Seeing as I have to do this anyway I decided to upgrade my very very old drivers to the latest 14.4. The drivers increased FPS on furmark by 33%, while core temps are bizarrely down but VRM temps are now up. If some explains that magic to me then I'll be impressed.

Eitherway retesting has begun, and the first block this time is EK because it's already on the card, then I'll move on to the newest blocks that I have but haven't run before e.g. Swiftech and Bitspower. That way we should get some coverage between all the results ASAP to see any significant differences in performance. So far though I think everything is going to be very close yet again on core temps.

That's it for now!

[Church]

FPS up but temps down? My bet is of yet another case when vendor puts in some benchmark or game specific "optimisations" by taking shortcuts or worsening rendering quality in such cases. IIRC i even read some article/test of this issue, where both AMD nor Nvidia were catched toying with "optimisations" for extra bench results, each in different set of games. From reading about alike cases sometimes they can be workarounded as simple as renaming filename of benchmark's or game's executable .. and getting heck of a different results.

[stren]

FPS up but temps down? My bet is of yet another case when vendor puts in some benchmark or game specific "optimisations" by taking shortcuts or worsening rendering quality in such cases. IIRC i even read some article/test of this issue, where both AMD nor Nvidia were catched toying with "optimisations" for extra bench results, each in different set of games. From reading about alike cases sometimes they can be workarounded as simple as renaming filename of benchmark's or game's executable .. and getting heck of a different results.

Yeah I can understand the optimizing part, the early drivers were terribad, but how can core temps go down and vrm's go up? The only thing I can think of is if they changed VRM switching frequency to provide better power delivery at the expense of heat? Well we'll see if this trend continues. Here's to round three of the 290 block testing - fingers crossed for finishing this one, and then it'll be onto the universal blocks!

[Church]

With universal blocks & testing top-end gpus using lot of power it's probably worth to double think out vrm cooling, so as to not follow skinneelab fate of burning gtx 480 during testing

[Vetalar]

at the days of hd4870 and 5870 i saw recomendations to _rename_ furmark executable file - this allows much more powerdraw and temp rise
for higher VRM temps: we have very little info about internal clocking, power limits, temp. limits, sensor placement, sensor's logic etc. some forumites notice that significant decrease temps on VRM - chillers e.g. - caused significant decrease of powerdraw via gpu-z sensors with same load.
"you know nothing, snow"

[stren]

With universal blocks & testing top-end gpus using lot of power it's probably worth to double think out vrm cooling, so as to not follow skinneelab fate of burning gtx 480 during testing

Yeah I'll have to figure something out for that - VRMs on this run are running 50C over coolant with a waterblock on them lol.

at the days of hd4870 and 5870 i saw recomendations to _rename_ furmark executable file - this allows much more powerdraw and temp rise
for higher VRM temps: we have very little info about internal clocking, power limits, temp. limits, sensor placement, sensor's logic etc. some forumites notice that significant decrease temps on VRM - chillers e.g. - caused significant decrease of powerdraw via gpu-z sensors with same load.
"you know nothing, snow"

Yeah with +100mV/+100mV 1150MHz core I'm getting 18-25C deltas on the core (depending on flow) for the EK block and an estimated power on the card of 330W. Based on that I don't think I want to push any higher on power so I won't try the renaming the furmark file. From what I've seen of actual temps of end users I suspect I'm running higher power than any "normal" non-benching watercooler would. The extra power is good in terms of differentiating block performance, but I don't want to risk the card by pushing it too hard. I don't want to restart this test for the 4th time lol.

Yeah chilling the VRMs will decrease their resistance and hence self-heating which is wasted power therefore total power for the card should go down even with the same load.

[stren]

For those who were interested - the AquaComputer backplate in passive mode (using the screws for the AC backplate) is compatible with the EK block, testing that right now because I'm a sucker for punishment lol.

[Grinder]

Good stuff. Too bad about having to redo tests!

I'm actually curious about the universal blocks (and tasteful vrm cooling), as well as full cover EK vs XSPC. I have a single 290 that I'm about to put under a full EK, and will be adding a second. In another rig I have dual 290's as well, so I'm looking for cost effective options there (no side window on that one, yet...). I actually like the look of the EK bridge edition (not so much the thermosphere due to higher cost + adapter needed) but I can't seem to find a matched pair anywhere due to their being EOL.

The raystorm do not officially declare support for 290's. Has anyone tried them yet?

[stren]

Hey Grinder - I have a raystorm gpu block so I'll be trying it (eventually).

EK results in:



Things to note - EK backplate reduces main VRMs by ~5C, while the AC backplate reduces them by ~20C. This is not a surprise based on the previous data but still good job to AC on that.

Also there is a slight difference on the core temps. The AC backplate insulated the PCB near the GPU core and hurts temps slightly, while the EK backplate uses a thermal pad and reduces temps slightly. So props to EK on that. Future improvements on GPU cooling may be to get extra heat out via an even more active backplate.

[stren]

Bitspower results - I didn't ask for the backplate, so I only have the block to test:



Core results surprised me, while VRMs are better - VRMs are not surprising because this block uses fujipoly 0.5mm pads and also cools another chip in the VRM area. Photos of that in the final writeup.

[stren]

Swiftech results - this block comes with the backplate - so this is compared only to results with backplates:



There were a lot of things I noted that Gabe & Co had changed on this block and as far as I could tell all were for the better. The backplate attempted to cool the VRM area, the LEDs and window are nice and the overall quality felt higher, cleaner and classier. There were also only three types of screw now rather than the five on the titan. Progress! Sadly there is no change in thermal performance from the titan roundup where Swiftech did well on the core but very poorly on VRM temps. Bear in mind those VRM temps are deltas. If your coolant is 35C as mine was then your VRMs would be at 110C!

[PuffinMyLye]

This thread is awesome and so helpful thanks a ton!

You say the passive AC backplate works with the EK block? Did you the test Rev. 1 or Rev. 2 of the EK block? Reason I'm asking is my card (MSI Twin Frozr) only works with Rev. 2 of the EK block.

Also if the passive backplate works shouldn't the active one work as well since the screws holes are the same?

[Jakusonfire]

More great detailed info. Any thoughts on the possibility of closing the gap between the EK and AC backplates by just replicating the extra thermal padding of the AC plate on the EK plate?
The EK plate seems to make only a marginal improvement.