The Cause Of And Fix For Radeon R9 290X And 290 Inconsistency

[EarthDog]

Unigine heaven and valley barely respond to cpu speed inceases. I know heaven for a fact barely budges...sadly, what you are seeing is likely more gpu than you even thought.

 

[kn00tcn]

if the max fan setting was raised to 100%, all the cards will perform the same wont they? they would just end up at different fan speeds if they really are generating different amounts of heat

 

[EarthDog]

Assuming 100% fan prevents thermal throttling, yep!

 

[qubit]

Same exact system, same architecture, and graphic driver, do you see the discrepancy in performance, the spread is close to 15%, that definitely does not fit under normal manufacturing irregularities in my book.

Look, I talk from personal experience, check the unigine thread in our very own forums, I tried to max out my card, and my system was beat by Ivy Bridge and Sandy Bridge systems with even less OC than my Haswell CPU, the difference was so big I just quit trying to get more performance from this card.

It's so bad that my original intention was to put this card under water, but why invest all that money for a lemon? I finally just decided to get an Arctic Hybrid for it and call it a day.

This is me, a 290X owner speaking from personal experience, there are way too high discrepancies between 290X cards in the wild, and this article opened my eyes to the issue, the solution, like many others said before is to either get a third party cooler or don't get a reference card at all, sad but true, I learned my lesson the hard way unfortunately.

That must have been really frustrating.

I'm curious, did the Arctic Hybrid stop it throttling or reduce the amount of throttling at least? Did the card then get near the performance scores that you were expecting?

It really looks to me like the only really bad thing about the 290X is the cooler. With a proper one like your aftermarket one , it should give NVIDIA a run for its money on noise and framerate performance.

 

[radrok]

Unigine heaven and valley barely respond to cpu speed inceases. I know heaven for a fact barely budges...sadly, what you are seeing is likely more gpu than you even thought.

0,5 FPS between 4,4 GHz and 5,2 GHz lol, (3930k)

 

[15th Warlock]

Unigine heaven and valley barely respond to cpu speed inceases. I know heaven for a fact barely budges...sadly, what you are seeing is likely more gpu than you even thought.

Yes, I fully agree, it's mostly GPU dependent, but processor OC should at least make it so my card scores a bit higher than another system with an older processor, and the same card on lower clocks, however that's not the case unfortunately:

http://www.techpowerup.com/forums/showpost.php?p=3004723&postcount=367
Core i7 4770K @ 4.7GHz

http://www.techpowerup.com/forums/showpost.php?p=3006016&postcount=388
Core i7 3770K @ 4.4GHz

As you can see my score is lower, even though my OC for the GPU and CPU are a little higher, I even set the fan at 100% and it didn't make any difference

That must have been really frustrating.

I'm curious, did the Arctic Hybrid stop it throttling or reduce the amount of throttling at least? Did the card then get near the performance scores that you were expecting?

It really looks to me like the only really bad thing about the 290X is the cooler. With a proper one like your aftermarket one , it should give NVIDIA a run for its money on noise and framerate performance.

It's, don't get me wrong, this is a monster of a card, and I run BF4 maxed out on it with absolutely zero issues, the performance you get for the price is, at the moment 2nd to none, but the card obviously has trouble maintaining high clocks with the reference cooler.

I haven't received my Hybrid yet, I'm really looking forward to putting it on my card and finally seeing it realize its full potential will definitely let you know how it works once I do

 

[qubit]

It's, don't get me wrong, this is a monster of a card, and I run BF4 maxed out on it with absolutely zero issues, the performance you get for the price is, at the moment 2nd to none, but the card obviously has trouble maintaining high clocks with the reference cooler.

I haven't received my Hybrid yet, I'm really looking forward to putting it on my card and finally seeing it realize its full potential

Epic, I wanna see those benchies man!

I also want it to eat NVIDIA's best cards for lunch directly competing with them head to head.

 

[EarthDog]

Yes, I fully agree, it's mostly GPU dependent, but processor OC should at least make it so my card scores a bit higher than another system with an older processor, and the same card on lower clocks, however that's not the case unfortunately:

http://www.techpowerup.com/forums/showpost.php?p=3004723&postcount=367
Core i7 4770K @ 4.7GHz

http://www.techpowerup.com/forums/showpost.php?p=3005985&postcount=387
Core i7 3770K @ 4.3GHz

As you can see my score is lower, even though my OC for both the Process and the GPU are a little higher, I even set the fan at 100%

That's what I am saying. These benchmarks are literally 99% gpu. Within reason, which a 3770k and 4770k are even with your slightly different clocks, won't be a difference maker in these benchmarks. Your gpu, for whatever reason, is seemingly throttling more than his it appears. The only way to confirm would be a gpuz fps log of both your cards at the same settings and analyze those results.

 

[TheoneandonlyMrK]

I do wonder what could be made if the collective brainpower of both companies engineers were put together with both companies intellectual property. Saddly this is called monopoly.

An emense argument followed by pigs smashing the windows in and eating all the cake.

 

[15th Warlock]

0,5 FPS between 4,4 GHz and 5,2 GHz lol, (3930k)

I noticed a sizable difference going from SB-E to Haswell using the exact same cards on this particular benchmark:

3930K at 5GHz 5863

4770K at 4.7GHz 6154

Both the X79 system and the Z87 system were running at PCIe 3.0. The GPUs had been OC'd on the second bench, but I don't think the difference in GPU clock would account for 300 more points, what do you think?

Your gpu, for whatever reason, is seemingly throttling more than his it appears.

Precisely, and that, my dear friend, is the conundrum

 

[EarthDog]

141Mhz on the GPU? Uhhh, yeah!

 

[natr0n]

I'm not a rocket scientist, but I'm pretty sure changing out the thermal paste to something better would make a world of difference.

I would do that 1st then if it doesn't help get an aftermarket cooler and be done.

 

[AsRock]

http://img.techpowerup.org/131112/geforcegtx_titan_3qtr2a.jpg

http://img.techpowerup.org/131112/AMD-Radeon-R9-290X-Cooler-635x426.jpg

I'm not expert but something seems different.

Looking at that it's like comparing a 120X25mm fan to a 120X35mm fan lol. Clearly just looking at the fan on the titan cooler is way deeper.

Cooler looks the same deal to as it looks taller on the titan.

 

[HTC]

http://img.techpowerup.org/131112/geforcegtx_titan_3qtr2a.jpg

http://img.techpowerup.org/131112/AMD-Radeon-R9-290X-Cooler-635x426.jpg

I'm not expert but something seems different.

The nVidia's cooler has less fins but seems sturdier (?) and the fan is higher.

Having never owned either card, i can't really tell but that's what it seems to me, anyway.


I've said it a few times already but i'll say it again: AMD shot themselves in the foot with a cannon ball by using that poor-excuse-of-a-cooler as a reference cooler.

For comparison purposes, i just would like to know what the performance would be on these cards if they were benched @ a fan speed (or water cooled) that made sure it did not throttle (not thinking of overclock), and compare the findings with the reference cards: that way, we could estimate the size of the cannon ball AMD used to shoot themselves in the foot.

 

[EarthDog]

I'm not a rocket scientist, but I'm pretty sure changing out the thermal paste to something better would make a world of difference.

I would do that 1st then if it doesn't help get an aftermarket cooler and be done.

That depends on the factory application really. I have seen a few C drop with a swap to none. It is worth a try as in this case, every little bit helps.