AMD Responds to R9 290 Series Performance Variance

Sorry, I haven't read every post if anyone else has posted this.

The way AMD's new turbo works is it lets the GPU get to it's temp target (~94°C), then increases fan to maintain that temp. If the fan can't maintain that temp then the card will throttle to do it.

1, Not all cards operate at the same voltage. Therefore cards that operate at lower voltages won't have to throttle as much as those that operate at higher voltage. Some chips are leakier than others and require higher voltage to be stable. This is exaggerated by the use of quiet mode.

2, These are mass produced cards. Not every heat sink is going to be attached exactly the same. Again, any differences are going to be exaggerated by quiet mode.

3, Remember in quiet mode the fan has been turned down lower than what is optimal for the cooling of this chip. It is optimized for low noise (relatively speaking) instead. In uber mode the fan speed is set to be able to maintain optimal performance. Some sites have used the excuse that the card comes out of the box set to quiet mode to randomly determine that is the default setting. AMD has stated that both settings are fully supported and are considered default settings.

4, Have we ever seen such scrutiny between the performance of review cards and retail cards. For example, are any of the nVidia cards used in the comparisons been tested against retail cards to see if their performance is typical of what is sold over the counter? Ask yourself why. ;)
 
Hmm, I wonder if this explains the behavior I've been seeing with my pair of 290s.

My cards start at 800 khash/sec and the system is pulling about 700W at the wall. Then it hits 95C, the fans spin up, power draw drops 50-100W, and performance drops to 600 khash/sec.

Increasing the fan speed on each card brought performance back up. Infuriatingly, neither the CCC nor the Gigabyte OC software would let me adjust the fan speed on my second card. Fortunately MSI afterburner wasn't picky about running on Gigabyte cards and I was able to throttle up both cards.

I find my experience curious when the [H] review specifically states that upping the fan speed had no performance impact. Difference between compute and gaming workloads, perhaps?

Anyway, these beasts are loud. Can't wait to put them under water.

Can't
 
i have two cards that run at different voltages.
Under full load, identical 800Mhz clock, 60% fan, 90degrees: the 71.1% ASIC card runs at 1.1x volts, the 78.4% ASIC card runs at 1.0x volts
[I compared them at 80% speed, because I did not want to invoke auto-throttle at 94degrees.]
the chip's yield quality appears to be a factor in the applied voltage - thermal/power regulation. but would need a larger sampling pool to determine if it's true.
 
Last edited:
Relayer said:
Sorry, I haven't read every post if anyone else has posted this.

The way AMD's new turbo works is it lets the GPU get to it's temp target (~94°C), then increases fan to maintain that temp. If the fan can't maintain that temp then the card will throttle to do it.

1, Not all cards operate at the same voltage. Therefore cards that operate at lower voltages won't have to throttle as much as those that operate at higher voltage. Some chips are leakier than others and require higher voltage to be stable. This is exaggerated by the use of quiet mode.

2, These are mass produced cards. Not every heat sink is going to be attached exactly the same. Again, any differences are going to be exaggerated by quiet mode.

3, Remember in quiet mode the fan has been turned down lower than what is optimal for the cooling of this chip. It is optimized for low noise (relatively speaking) instead. In uber mode the fan speed is set to be able to maintain optimal performance. Some sites have used the excuse that the card comes out of the box set to quiet mode to randomly determine that is the default setting. AMD has stated that both settings are fully supported and are considered default settings.

4, Have we ever seen such scrutiny between the performance of review cards and retail cards. For example, are any of the nVidia cards used in the comparisons been tested against retail cards to see if their performance is typical of what is sold over the counter? Ask yourself why. ;)
Click to expand...

The systerm isnt quite smart enough to use fan speed rather than choosing to throttle.

I've set many targets in hopes of letting CCC take care of fan speed, but it was just crap. 100% as max fan speed and no OC and it couldnt prevent throttling with an 85C target. Its incredibly stupid. When it finally puts the fan speed high enough to prevent throttling it lowers it shortly after, and the cycle repeats. It never dampens and becomes stable, at least in my case.
 
LstOfTheBrunnenG said:
Hmm, I wonder if this explains the behavior I've been seeing with my pair of 290s.

My cards start at 800 khash/sec and the system is pulling about 700W at the wall. Then it hits 95C, the fans spin up, power draw drops 50-100W, and performance drops to 600 khash/sec.

Increasing the fan speed on each card brought performance back up. Infuriatingly, neither the CCC nor the Gigabyte OC software would let me adjust the fan speed on my second card. Fortunately MSI afterburner wasn't picky about running on Gigabyte cards and I was able to throttle up both cards.

I find my experience curious when the [H] review specifically states that upping the fan speed had no performance impact. Difference between compute and gaming workloads, perhaps?

Anyway, these beasts are loud. Can't wait to put them under water.

Can't
Click to expand...

after days of mining they are steady at 82c, 70% fan, OCd to 1100/1500.

heatsink mod, look the difference in temps
unbn.jpg

sbe4.jpg
 
45454asasa said:
The systerm isnt quite smart enough to use fan speed rather than choosing to throttle.

I've set many targets in hopes of letting CCC take care of fan speed, but it was just crap. 100% as max fan speed and no OC and it couldnt prevent throttling with an 85C target. Its incredibly stupid. When it finally puts the fan speed high enough to prevent throttling it lowers it shortly after, and the cycle repeats. It never dampens and becomes stable, at least in my case.
Click to expand...

Thats not how powerplay 2.0 works.

setting an 85c target will force it to throttle.

95c is its normal operation temp, in order to lower it, it needs to lower the power consumption through throttling..

The only thing you can do it leave it at 95, and raise the fan limit.

Or get an aftermarket cooler, or put it under water.

ill say it again: LOWERING THE TARGET TEMP WILL FORCE IT TO THROTTLE 100% OF THE TIME.
 
Last edited:
wantapple said:
after days of mining they are steady at 82c, 70% fan, OCd to 1100/1500.

heatsink mod, look the difference in temps
unbn.jpg

sbe4.jpg
Click to expand...

what mod did you do? and it does look like gpu 0 is the one without the bracket. did you do something else?
 
i just remove the bracket from gpu's, about the cgminer i dont know why is always showing gpu 0 on gpu 2 temps.

anyway mild oc to 1050/1500, room temperature between 78 to 80.
they have been running for days.

2zpd.jpg

1te3.jpg

07lx.jpg


if i open my door they drop to 74's LOL
 
was it hard to take off the bracket? i may as well do that when i change the TIM tomorrow I dont think Xfx voids warrenty in North america
 
If that's actually true, and really does work, you ought to be posting that info in more highly viewed places.
Although it still looks like GPU2 is the one on the far-right since it's sucking in cold air.
 
i did the same this morning, it made a huge difference. Nothing like what his thing is claiming. Im sitting @ 91 degree's with a slight overclock and my fans are @ 50%. before i would have to put my fans @ 65% inorder for my GPU to not throttle @ stock clocks.
 
TaintedSquirrel said:
If that's actually true, and really does work, you ought to be posting that info in more highly viewed places.
Although it still looks like GPU2 is the one on the far-right since it's sucking in cold air.
Click to expand...
Bracket mod has been known about a week after release.
 
You must log in or register to reply here.
Back
Top