Defiance
Member
+438|6673

I just installed an Accelero S1 Rev. 2 on a 5870 and the installation went smoothly enough, and then I went to test it. Running Furmark and watching GPU-Z on another monitor, the card hit 90 C before I stopped it. Even worse, the GPU Temp 2 and 3 in GPU-Z ran up to 111 C. I added a fan to the cooler and removed the spacers in case they were limiting contact. Now, I'm here:

https://i.imgur.com/1mXoo.jpg

Even after that, GPU Temp 2 is at 110. I know most video cards can survive those temps, but doesn't that still seem high? Obviously, the S1 isn't the best at cooling this thing but the core temp should indicate that it's capable. I can't understand why Temp 2 and 3 are still so high.

My only next step is to reapply the TIM; the paste that's on there now has a good spread but it might be a bit too much.

TL;DR: Are those die-temps normal for a 5870?

(Edit: I realize it's basically a google-able question, but I tried that, and my fu failed. Best I could find was that, respectively, the temps are core/memory/shader. The memory controller is fucking hot. Funny enough, but hopefully not related, is that when I pulled off the stock cooler, none of the RAM chips had heatsinks at all. They do now, I used the ones included in the S1 with AS adhesive, but adding heatsinks to the RAM chips doesn't equate to fucking up the memory controller in the die imo.)

Last edited by Defiance (2010-07-31 02:19:12)

GC_PaNzerFIN
Work and study @ Technical Uni
+528|6417|Finland

No those are not normal die temperatures, they are dangerous.

I am guessing you forgot to put heatsink to certain chip near the memory chips, and also what did you do the the VRM? (what heatsinks did you use for those)

Not only that but Accelero is poor cooler for HD 5870. It was never designed for that kind of heatload.

E: I can't quite understand why anyone would play hard ball with a 350€ card, why didn't you just get a Zalman VF-3000 or something like that?

Last edited by GC_PaNzerFIN (2010-07-31 05:25:29)

3930K | H100i | RIVF | 16GB DDR3 | GTX 480 | AX750 | 800D | 512GB SSD | 3TB HDD | Xonar DX | W8
alexb
<3
+590|5942|Kentucky, USA

Temps 2 and 3 are most likely VRM and PCB temps. Are you sure you secured it properly? Also, less goop is better than more goop.

Last edited by alexb (2010-07-31 07:22:52)

Bevo
Nah
+718|6523|Austin, Texas
Why are you overclocking a 5870?
alexb
<3
+590|5942|Kentucky, USA

The previous heatsink probably had contact with the memory controller, and that's probably why your temps are that high; what did your temps look like on the stock cooler?
Freezer7Pro
I don't come here a lot anymore.
+1,447|6200|Winland

Temp 2 and 3 are probably the voltage regulator MOS-FETs. They're usually fine up to about 170°C. Running Furmark, mine usually get to about 130°, thanks to Nvidia being unable to stick to one VRM design.
The idea of any hi-fi system is to reproduce the source material as faithfully as possible, and to deliberately add distortion to everything you hear (due to amplifier deficiencies) because it sounds 'nice' is simply not high fidelity. If that is what you want to hear then there is no problem with that, but by adding so much additional material (by way of harmonics and intermodulation) you have a tailored sound system, not a hi-fi. - Rod Elliot, ESP
GC_PaNzerFIN
Work and study @ Technical Uni
+528|6417|Finland

Freezer7Pro wrote:

Temp 2 and 3 are probably the voltage regulator MOS-FETs. They're usually fine up to about 170°C. Running Furmark, mine usually get to about 130°, thanks to Nvidia being unable to stick to one VRM design.
The thing is the cards VRM throttles at much lower temperatures than the old NV cards.

And the HD 5870 has its own sensors for VRM, seperate from the GPU sensors.

https://i882.photobucket.com/albums/ac25/Asmola85/zalman50percent.jpg?t=1274816861

e: not my pic and with VF-3000A

Last edited by GC_PaNzerFIN (2010-07-31 09:41:00)

3930K | H100i | RIVF | 16GB DDR3 | GTX 480 | AX750 | 800D | 512GB SSD | 3TB HDD | Xonar DX | W8
Freezer7Pro
I don't come here a lot anymore.
+1,447|6200|Winland

GC_PaNzerFIN wrote:

Freezer7Pro wrote:

Temp 2 and 3 are probably the voltage regulator MOS-FETs. They're usually fine up to about 170°C. Running Furmark, mine usually get to about 130°, thanks to Nvidia being unable to stick to one VRM design.
The thing is the cards VRM throttles at much lower temperatures than the old NV cards.

And the HD 5870 has its own sensors for VRM, seperate from the GPU sensors.

http://i882.photobucket.com/albums/ac25 … 1274816861

e: not my pic and with VF-3000A
VRM throttling? What are you talking about?

And while what you're saying might be true, it's obvious that Defiance's card doesn't have the same amount of sensors as that which you posted. For all we know, it could be reporting the VRM temps as "GPU Temp" values. It doesn't look like they're values from the core, at least, seeing how there's an almost 40°C difference between the three.

Defiance, have you tried removing the cooler and replacing the thermal paste? The layer of MX-2 on the S1 is ridiculously thick. Have you also made sure that everything on the card that had a thermal pad on it, connecting it to the stock heatsink has an Accelero heatsink, too?

Last edited by Freezer7Pro (2010-07-31 09:58:16)

The idea of any hi-fi system is to reproduce the source material as faithfully as possible, and to deliberately add distortion to everything you hear (due to amplifier deficiencies) because it sounds 'nice' is simply not high fidelity. If that is what you want to hear then there is no problem with that, but by adding so much additional material (by way of harmonics and intermodulation) you have a tailored sound system, not a hi-fi. - Rod Elliot, ESP
GC_PaNzerFIN
Work and study @ Technical Uni
+528|6417|Finland

Freezer7Pro wrote:

VRM throttling? What are you talking about?

And while what you're saying might be true, it's obvious that Defiance's card doesn't have the same amount of sensors as that which you posted. For all we know, it could be reporting the VRM temps as "GPU Temp" values. It doesn't look like they're values from the core, at least, seeing how there's an almost 40°C difference between the three.

Defiance, have you tried removing the cooler and replacing the thermal paste? The layer of MX-2 on the S1 is ridiculously thick. Have you also made sure that everything on the card that had a thermal pad on it, connecting it to the stock heatsink has an Accelero heatsink, too?
The HD 5870 VRM has both overcurrent and overheat protections. Either will make sure the VRM doesn't get to dangerous temperatures, 130c is the absolute max rated temperature the VRM can take, throttling somewhere under that.

GPU-Z reports following:

GPU Temp. #2 shows the temperature of a specific section of the GPU-die

The exact location is unknown.

e: and http://forums.techpowerup.com/showpost. … ostcount=2

Last edited by GC_PaNzerFIN (2010-07-31 10:13:33)

3930K | H100i | RIVF | 16GB DDR3 | GTX 480 | AX750 | 800D | 512GB SSD | 3TB HDD | Xonar DX | W8
Freezer7Pro
I don't come here a lot anymore.
+1,447|6200|Winland

GC_PaNzerFIN wrote:

Freezer7Pro wrote:

VRM throttling? What are you talking about?

And while what you're saying might be true, it's obvious that Defiance's card doesn't have the same amount of sensors as that which you posted. For all we know, it could be reporting the VRM temps as "GPU Temp" values. It doesn't look like they're values from the core, at least, seeing how there's an almost 40°C difference between the three.

Defiance, have you tried removing the cooler and replacing the thermal paste? The layer of MX-2 on the S1 is ridiculously thick. Have you also made sure that everything on the card that had a thermal pad on it, connecting it to the stock heatsink has an Accelero heatsink, too?
The HD 5870 VRM has both overcurrent and overheat protections. Either will make sure the VRM doesn't get to dangerous temperatures, 130c is the absolute max rated temperature the VRM can take, throttling somewhere under that.

GPU-Z reports following:

GPU Temp. #2 shows the temperature of a specific section of the GPU-die

The exact location is unknown.

e: and http://forums.techpowerup.com/showpost. … ostcount=2
Have you got a source for the part about the VRM throttling? I find it quite odd, as it'd be much cheaper to just have it cut when it gets too hot.

And if the GPU Temp. #2 value actually is one of the die, I think it's quite definitive that Defiance has some issues with the thermal contact of the cooler.
The idea of any hi-fi system is to reproduce the source material as faithfully as possible, and to deliberately add distortion to everything you hear (due to amplifier deficiencies) because it sounds 'nice' is simply not high fidelity. If that is what you want to hear then there is no problem with that, but by adding so much additional material (by way of harmonics and intermodulation) you have a tailored sound system, not a hi-fi. - Rod Elliot, ESP
Defiance
Member
+438|6673

GC_PaNzerFIN wrote:

No those are not normal die temperatures, they are dangerous.

I am guessing you forgot to put heatsink to certain chip near the memory chips, and also what did you do the the VRM? (what heatsinks did you use for those)

Not only that but Accelero is poor cooler for HD 5870. It was never designed for that kind of heatload.

E: I can't quite understand why anyone would play hard ball with a 350€ card, why didn't you just get a Zalman VF-3000 or something like that?
When I was looking for a 5870 cooler, the VF3000 never came up, but the S1 came up multiple times in favor.
https://imgur.com/uOtDC.jpg

That's the stock, albeit non-reference, XFX cooler that came on the card. As said above, it doesn't even contact the memory at all, let alone any other small chips. The VRMs are handled by another heatsink that I left on the card. Here's my card:
https://imgur.com/9iIt8.jpg

Sorry about the glare. In any case, this XFX also differs in layout from the stock board. Here's the reference cooler:
https://www.techpowerup.com/reviews/ATI/Radeon_HD_5870/images/cooler3.jpg

There's that little strip off to the side of the memory, near the crossfire bridges. On the reference board, in the red box are what's being cooled by that strip:
https://imgur.com/fKGYU.jpg

So, in conclusion, the stock cooler wasn't cooling anything that I'm not cooling now and I've given it the benefit of cooling the memory. If anyone can identify what chips are being cooled in the reference design and not on this XFX card, I've got spare heatsinks that I could put on it. For now, I'm going to try some fresh TIM.

Bevo wrote:

overclocking something or other....
I'm not overclocking it, it's automatically underclocks when it idles.
GC_PaNzerFIN
Work and study @ Technical Uni
+528|6417|Finland

http://www.anandtech.com/print/2877

"In both cases, the key was the overall load on the GPU cores, and consequently the amount of power required to drive the GPUs. When a bank of VRMs reached roughly 120C (this being averaged among all the VRMs in that bank), overcurrent protection kicked in and throttling began."

http://www.anandtech.com/Show/Index/359 … amp;page=1

HD 5970, but same applies to HD 5870...

e: there is very small chip that needs cooling, near the memory chips on reference card

https://img31.imageshack.us/img31/2950/5870vrm.jpg

The strip on the base doesn't cool the cap, lol... it cools that tiny chip

Last edited by GC_PaNzerFIN (2010-07-31 10:39:15)

3930K | H100i | RIVF | 16GB DDR3 | GTX 480 | AX750 | 800D | 512GB SSD | 3TB HDD | Xonar DX | W8
Freezer7Pro
I don't come here a lot anymore.
+1,447|6200|Winland

GC_PaNzerFIN wrote:

http://www.anandtech.com/print/2877

"In both cases, the key was the overall load on the GPU cores, and consequently the amount of power required to drive the GPUs. When a bank of VRMs reached roughly 120C (this being averaged among all the VRMs in that bank), overcurrent protection kicked in and throttling began."

http://www.anandtech.com/Show/Index/359 … amp;page=1

HD 5970, but same applies to HD 5870...
Hm, that was indeed a quite odd approach to it by AMD. Kudos to them for not cheaping out.
The idea of any hi-fi system is to reproduce the source material as faithfully as possible, and to deliberately add distortion to everything you hear (due to amplifier deficiencies) because it sounds 'nice' is simply not high fidelity. If that is what you want to hear then there is no problem with that, but by adding so much additional material (by way of harmonics and intermodulation) you have a tailored sound system, not a hi-fi. - Rod Elliot, ESP
GC_PaNzerFIN
Work and study @ Technical Uni
+528|6417|Finland

Freezer7Pro wrote:

Hm, that was indeed a quite odd approach to it by AMD. Kudos to them for not cheaping out.
Yeah I was surprised about that too. I guess they thought its about time to do something with the VRM protection eventually...

Defiance wrote:

When I was looking for a 5870 cooler, the VF3000 never came up, but the S1 came up multiple times in favor.
Accelero S1 is about last coolers I would use on HD 5870, there are much much better ones like Zalman VF-3000A, Accelero Xtreme HD 5870, some revision of the Accelero Twin Turbo....

Zalman has seperate heatsinks for the ram so that would have been better for custom HD 5870 than the AC Xtreme tho...

Last edited by GC_PaNzerFIN (2010-07-31 10:55:46)

3930K | H100i | RIVF | 16GB DDR3 | GTX 480 | AX750 | 800D | 512GB SSD | 3TB HDD | Xonar DX | W8
Defiance
Member
+438|6673

Ah yes, I was fooled by the flip...

In any case, I took a look underneath the heatsink where the VRMs are on the reference board and there were only 4 short, wide chips that I can only assume are the VRMs because there is nothing else on the board like the VRMs on the reference. And, because it's a non-reference board, the VRMs don't have temp monitoring.

After re-gooping the HSF and seating it again, the temps dropped about 10C so the highest temp is Temp 2, at 99C. Temp 3 is 10 degrees lower, and Temp 1 is 30 lower. I'm going to see if I can get anywhere through XFX as far as finding out what each of these references. For now, though, I'll be keeping a close eye on temps during gaming... on a related note, ffuuuuuu.
Defiance
Member
+438|6673

(Bump for conclusion.)

I reseated the S1 and got the temps down to 80.5 C max in Furmark, and 58 C max during actual gaming.

I spoke with an XFX tech and he confirmed that those sensors are from the die itself, and that it seems normal to see a spread in temperatures. He tested a reference 5870 and, under Furmark, he got 91.5, 94.0, and 103.0 C. His spread was 11.5 C, and I find it odd that mine goes as high as 16 C, but since none of the temperatures are at dangerous levels any more, I'm content.

Board footer

Privacy Policy - © 2024 Jeff Minard