bit-tech.net

AMD denies cherry-picking R9 290, 290X review cards

AMD denies cherry-picking R9 290, 290X review cards

AMD has denied accusations that it has cherry-picked 'golden sample' R9 290 and R9 290X cards for review, but admits that performance is variable.

AMD has denied accusations that Radeon R9 290 and 290X graphics boards supplied to publications for review are hand-picked 'golden sample' boards that perform considerably better than the same boards purchased at retail.

AMD's latest R9 family of graphics processing units (GPUs) have caused something of a stir, offering considerable performance at a reasonable price-point with the flip-side of extreme heat exacerbated by a less-than-optimal stock cooler. One site, however, claims that the high performance seen on the AMD-provided review sample isn't reflected when running the same benchmarks on independently-purchased retail hardware - accusing the company of cherry-picking so-called 'golden sample' hardware for review.

In its analysis of the theoretically lower-end Radeon R9 290 board, Tom's Hardware noticed a discrepancy: the AMD-provided R9 290 review sample was often outperforming a more expensive and supposedly more powerful R9 290X board purchased from retailer Newegg.

The site places the blame on the dynamic clocking mechanism, which aims to keep the core temperature of the board below a certain level. This mechanism was seen to drop the retail R9 290X from its theoretical 1,000MHz core clock maximum to just 727MHz - crippling performance. 'For R9 290X, we can go out and buy boards to compare [performance],' the site explained. 'But there’s no way to know if the R9 290s you buy will operate at the top of their range (947 MHz) or the bottom (662 MHz).'

Responding to press enquiries, an AMD spokesperson denied that the company has provided cherry-picked hardware to reviewers. 'A media outlet has uniquely reported instances of AMD Radeon R9 290X boards purchased in retail that have exhibited an uncharacteristic level of performance variance as compared to press samples issued by AMD,' the company's statement read. 'We’re working to secure the board(s) in question for further analysis. Boards purchased by other media outlets have not exhibited similar characteristics that we’re aware of.'

Although it denies any wrongdoing, AMD has admitted that performance variance is an issue with the R9 series. 'We’ve identified areas where [performance] variability can be minimised,' the spokesperson added, 'and are working on a driver update which will minimize this variance.' No release date for the update has been provided.

UPDATE 20131108:
True to its word, AMD has released a beta driver which it claims resolves the issue. The problem, the company has claimed, is in an unexpected variability in fan speeds which impacts the performance of the stock cooler - resulting in certain board batches reaching their target temperatures earlier than others, and performing accordingly poorer.

'We’ve identified that there’s variability in fan speeds across AMD R9 290 series boards,' an AMD spokesperson explained via email. 'This variability in fan speed translates into variability of the cooling capacity of the fan-sink. The flexibility of AMD PowerTune technology enables us to correct this variability in a driver update. This update will normalize the fan RPMs to the correct values. The correct target RPM values are 2200RPM for the AMD Radeon R9 290X ‘Quiet mode’, and 2650RPM for the R9 290.'

The updated driver, currently only a beta, uses the company's PowerTune platform to adjust the fan speeds to match between batches and should, in theory, mean all 290X and 290 boards have the same performance.

47 Comments

Discuss in the forums Reply
Harlequin 6th November 2013, 10:50 Quote
so nothing like titan`s slowing down when they get hot then.....
Gareth Halfacree 6th November 2013, 10:58 Quote
Quote:
Originally Posted by Harlequin
so nothing like titan`s slowing down when they get hot then.....
Depends: do two supposedly identical Titans with identical coolers in identical systems show up to a 30 per cent disparity in performance? If so, then yes; it's exactly like that.
rollo 6th November 2013, 11:17 Quote
Toms hardware ran the real world tests most of us have asked for since launch. The downclock is significant it would seem on retail boards due to the core temp been so high. A custom cooler would fix this but this puts AMDs stock cooler in even more bad light.

Still don't understand why AMD did not give us a real cooler.
DbD 6th November 2013, 11:24 Quote
They are running them right on the limit. The chip is locked to it's 95 degree temp at all times and performance is completely dependent on cooling. You run at 40% fan speed and it's much slower then 55%.

Not surprising then that the hotter running chips are substantially slower for a fan speed then the cooler running ones. Equally quality of cooling in your case would have a similar difference. As would several months of dust. As would playing in the summer vs the winter.

Not surprising that reviewers got cherry picked samples - I bet they always do for nvidia and AMD - it's just in this case it makes a huge difference because AMD messed up and didn't put a cooler on the card that can really cool it at anything less then jet engine noise levels.
GeorgeK 6th November 2013, 11:42 Quote
Quote:
Originally Posted by Gareth Halfacree
Depends: do two supposedly identical Titans with identical coolers in identical systems show up to a 30 per cent disparity in performance? If so, then yes; it's exactly like that.

I know there's always been a variance in the overclocking ability of parts (the silicon lottery I believe it's called) but to have two 'identical' items that at stock have such a different performance is totally unacceptable.

Imagine that you and a friend bought a car advertised to do 50mpg - his did 50mpg and yours only did 35mpg - you'd send yours back...
rollo 6th November 2013, 11:44 Quote
I'd send mine back if it was down in the 700mhz - 800mhz range.
Spreadie 6th November 2013, 12:44 Quote
Quote:
Originally Posted by GeorgeK
I know there's always been a variance in the overclocking ability of parts (the silicon lottery I believe it's called) but to have two 'identical' items that at stock have such a different performance is totally unacceptable.
That's right - you can only really compare competing products at their guaranteed clocks. While it's true that one type of GPU may generally overclock better than another, you cannot guarantee that speed.

If AMD cannot produce cards that will run at stated clocks, their value and appeal is drastically reduced. I've been very critical of AMD's new cards' power consumption and operating temperature, but I didn't want this to become an embarrassment for them; and it has the potential to become just that.

What I'm struggling with is, if they were fully aware of this issue, why not bite the bullet and delay while developing a new cooler, or target them against the 770s and cripple the GPU enough the manage the heat with the current cooler and then launch a flagship product with a redesigned cooler to steal the performance crown?
teppic 6th November 2013, 13:00 Quote
It's based on a single press sample versus two retail samples. No other review site has found anything similar. This isn't news, it's a way to generate hits.
rollo 6th November 2013, 13:06 Quote
Don't know of anyone on bit tech looking at either card( 290x 290) similar story is repeated across a lot of forums. Not seen a user review of one yet on any of the forums I regularly vist.

Cherry picking review samples would not be the first time it's happened in the pc sector. Most likely you will see other review sites test this themselves. Assuming they can find a card to buy.
Spreadie 6th November 2013, 13:12 Quote
Quote:
Originally Posted by teppic
It's based on a single press sample versus two retail samples. No other review site has found anything similar. This isn't news, it's a way to generate hits.
As far as I can tell, no other review site has bothered to compare their review samples with retail cards. Although, Bit-tech's follow up article on the 290X did pick up on the thermal throttling and resultant performance drop-off.

Tom's Hardware incorporated their additional testing in their review - and added a potential caveat to the title, based on testing results. This will naturally attract more interest, but that doesn't make this a manufactured click-bait article.
teppic 6th November 2013, 13:20 Quote
A quick glance and things like this come up:
Quote:
Originally Posted by Gibbo
Absolute rubbish about press cards being golden samples.

We had a press sample card, it did OK, we then got an Asus card and a Sapphire card form our warehouse stock, they both beat the press card in all our benchmarks as they hit higher overclocks. None of the cards experienced any throttling issues.

So without a doubt, complete BS. :)
Quote:
Originally Posted by w3bbo
I have reviewed GPU's for the past 7 years now and not once can I categorically state that a card direct from AMD or NVIDIA was any faster than one from retail or one of the main re-sellers. Some cards perform better than others but then the same can be said about retail cards. Every card will not clock exactly the same, I have had good and not so good clockers both from AMD/NVIDIA. Like retail, it is luck of the draw or perhaps, just maybe, user error?

Never heard so much tosh in my life.

Toms have based their conclusions on a couple of retail samples vs one press sample, which says it all. They haven't bothered to allow AMD to test the cards to verify anything, but just publish what amounts to 'our press sample works, our two retail samples have problems as far as we can see, so all retail cards have problems'.
Corky42 6th November 2013, 13:24 Quote
Quote:
Originally Posted by Spreadie
What I'm struggling with is, if they were fully aware of this issue, why not bite the bullet and delay while developing a new cooler, or target them against the 770s and cripple the GPU enough the manage the heat with the current cooler and then launch a flagship product with a redesigned cooler to steal the performance crown?

Rumor's are saying it was originally meant to target the 770, but with Nvidia's recent price cuts they may have pushed it to go against the 780 and Titan. It wouldn't have looked good if the 290 matched the 770 now its some £50 cheaper.
GeorgeK 6th November 2013, 13:32 Quote
Quote:
Originally Posted by teppic
It's based on a single press sample versus two retail samples. No other review site has found anything similar. This isn't news, it's a way to generate hits.

That may be true but in that case they would have to either:

(a) been sent a stonking press sample (possible but theirs seems to be in line with others' press samples)

or

(b) they've bought / received two duff retail samples (unlikely but possible)

It'll be interesting to see real people's findings but I wouldn't want to splash out on one thinking that it may perform worse than I've seen in reviews...
teppic 6th November 2013, 13:32 Quote
Quote:
Originally Posted by Corky42
Rumor's are saying it was originally meant to target the 770, but with Nvidia's recent price cuts they may have pushed it to go against the 780 and Titan. It wouldn't have looked good if the 290 matched the 770 now its some £50 cheaper.

The 280X already matches (and often beats) the 770, plus has better memory specs.
Spreadie 6th November 2013, 13:33 Quote
Quote:
Originally Posted by teppic
A quick glance and things like this come up:





Toms have based their conclusions on a couple of retail samples vs one press sample, which says it all. They haven't bothered to allow AMD to test the cards to verify anything, but just publish what amounts to 'our press sample works, our two retail samples have problems as far as we can see, so all retail cards have problems'.

OK, who are Webbo and Gibbo, and do they provide proof comparative testing? They may have a point, but all I see so far is a copypasta forum post so I couldn't possibly accept it at face value.
teppic 6th November 2013, 13:35 Quote
Quote:
Originally Posted by GeorgeK
That may be true but in that case they would have to either:

(a) been sent a stonking press sample (possible but theirs seems to be in line with others' press samples)

or

(b) they've bought / received two duff retail samples (unlikely but possible)

It'll be interesting to see real people's findings but I wouldn't want to splash out on one thinking that it may perform worse than I've seen in reviews...

All the things I've seen from 290x owners on ocuk forums suggest this problem doesn't exist. Throttling can happen on quiet mode, but not to a large extent - but that's reported in the reviews too. The problem is toms haven't done a real investigation, it's just FUD.
Quote:
Originally Posted by Spreadie
OK, who are Webbo and Gibbo, and do they provide proof comparative testing? They may have a point, but all I see so far is a copypasta forum post so I couldn't possibly accept it at face value.

Gibbo's the boss at ocuk, and w3bbo does reviews for techspot.
Goty 6th November 2013, 13:37 Quote
Quote:
Originally Posted by Spreadie
As far as I can tell, no other review site has bothered to compare their review samples with retail cards.

http://www.sweclockers.com/artikel/17845-snabbtest-radeon-r9-290x-i-blasvader-for-golden-samples-sweclockers-undersoker

There you go.
Panos 6th November 2013, 13:38 Quote
Tom's is full of BS.

Can we have names of the "retail" cards tested at Tom's?

Because people who got the 290X and put MK26 on them, they are flying, while running cool and quiet.

Too many sour grapes from Nvidia side it seems.
Panos 6th November 2013, 13:39 Quote
Quote:
Originally Posted by teppic
Quote:
Originally Posted by Corky42
Rumor's are saying it was originally meant to target the 770, but with Nvidia's recent price cuts they may have pushed it to go against the 780 and Titan. It wouldn't have looked good if the 290 matched the 770 now its some £50 cheaper.

The 280X already matches (and often beats) the 770, plus has better memory specs.

ANY 280X beats a GTX770, as the 7970Ghz did. And the 280X Toxic beats a GTX780.
teppic 6th November 2013, 13:41 Quote
Quote:
Originally Posted by Panos
ANY 280X beats a GTX770, as the 7970Ghz did. And the 280X Toxic beats a GTX780.

Overall the 280X beats a 770, but in nvidia sponsored games (e.g. Metro 2033) it won't. Overclock them both fully and the 280X is definitely ahead. The point really is that AMD wouldn't target the 770 with the 290 as it's miles ahead, and the 280X does the job.
Spreadie 6th November 2013, 13:46 Quote
Quote:
Thanks
Quote:
Originally Posted by Panos
Tom's is full of BS.

Can we have names of the "retail" cards tested at Tom's?

Because people who got the 290X and put MK26 on them, they are flying, while running cool and quiet.

Too many sour grapes from Nvidia side it seems.
Wait, someone spent £50 on a 3rd party cooler and their card ran cooler and quieter?

:|
Corky42 6th November 2013, 13:49 Quote
http://arstechnica.com/gadgets/2013/11/amd-stomps-nvidia-with-r9-290-at-least-in-reviews/
Quote:
There is a downside to this performance, however: noise. AMD originally planned for the 290 to go up against Nvidia's $399 GTX 770. The 290 handily beats the 770—but thanks to Nvidia's price cuts, the 770 is no longer a $399 card. To keep the 290's performance looking good, AMD has had to make its fans spin faster, giving it more thermal headroom and allowing its GPU to run faster.

And don't shoot the messenger although feel free to shoot the actual message :D
teppic 6th November 2013, 13:50 Quote
Well, lots of owners are reporting over 1GHz stable on the 290x with reference cooler with no real throttling problems - up to ~1200MHz. Gibbo's post says:
Quote:
Sapphire (Asus BIOS) R290X did 1200 core easy and 6400MHz RAM.
Asus R290X did 1200-1220 core and 6600-6800MHz RAM.

The reference cooler would obviously be noisy at these clocks, but they're just regular retail models.
RedFlames 6th November 2013, 14:49 Quote
This kinda thing gets filed under 'no **** sherlock...'

Of course they're going to deny it [likewise nvidia would deny it if it was aimed at them]...

Even *if* it's true AMD/nVidia are hardly gonna fess up and say 'Oh yeah, those cards that the reviewers say is so awesome? Yeah... those aren't the cards we're selling to you...' are they?
rollo 6th November 2013, 15:05 Quote
Quote:
Originally Posted by Panos
Tom's is full of BS.

Can we have names of the "retail" cards tested at Tom's?

Because people who got the 290X and put MK26 on them, they are flying, while running cool and quiet.

Too many sour grapes from Nvidia side it seems.

This sort of news is bad however way you look at it, Let me get this straight people who brought £50 aftermarket coolers are flying oh rly? lol

Its the shitty stock cooler people are moaning about not what some aftermarket mod does.

There is no aftermarket card for the 290x in retail yet so what you get is the stock model only. 99% of people who buy such a card would have no clue how to mod its cooler.

Cant read that language whatever it is so translation would be awesome.

Personally would like to see Bit tech or Anandtech do this test as both are pretty balanced review sites, Every other site is so pro Nvidia or pro AMD so its irelivent.
GuilleAcoustic 6th November 2013, 15:14 Quote
not to mention that installing an aftermarket cooler will void the warranty ...
schmidtbag 6th November 2013, 15:55 Quote
I don't see how hand-picking a GPU would make that much of a difference anyway. The reviewers all prove the clock settings, driver versions, OS, etc. When you've got over 6 billion transistors, you could probably cut out a few thousand and not have much of a performance impact.

Honestly, I actually expected the 290 to have the same performance as the 290X the week before the 290 was even released. I thought the 290X produced maximum heat way too easily, to the point that it would down-throttle itself to below the performance of the 290.

I sincerely think the test results shown for the 290X and 290 from any website are relatively pointless. Their results are so skewed by those ineffective heatsinks. Do some benchmarks with water blocks and then we've got some solid numbers to compare.


@rollo
guru3d.com is probably the most un-biased website I'm aware of, and also of the most thorough and informative.
Harlequin 6th November 2013, 16:31 Quote
Quote:
Originally Posted by RedFlames
This kinda thing gets filed under 'no **** sherlock...'

Of course they're going to deny it [likewise nvidia would deny it if it was aimed at them]...

Even *if* it's true AMD/nVidia are hardly gonna fess up and say 'Oh yeah, those cards that the reviewers say is so awesome? Yeah... those aren't the cards we're selling to you...' are they?

except that multiple reviewers have purchased cards `blind` and got similar if not better results....
teppic 6th November 2013, 16:58 Quote
AMD rep commented with:
Quote:
Originally Posted by Thracks
I'll chip in with official word on this.

Two outlets in hundreds sampled have uniquely reported instances of AMD Radeon R9 290(X) boards purchased in retail that have exhibited an uncharacteristic level of performance variance as compared to press samples issued by AMD. As retail products purchased by almost every other outlet (e.g. Sweclockers) do not demonstrate this phenomenon, we’re working to secure the aberrant board(s) in question for further analysis. In the meantime, we’ve identified areas where variability can be minimized and are working on a driver update which will minimize this variance.

We will be releasing that driver in the next 24 hours to fully correct this behavior by normalizing fan RPMs vs. PWM control. The 290X in Quiet mode should be at 2200RPM, and the 290 should be at 2650RPM. If anyone is seeing fan speeds below that, they are affected by this issue, and the driver will resolve the issue in 24 hours.

//EDIT: All boards should perform similarly to the AMD-issued samples seen in reviews. Period.
Corky42 6th November 2013, 17:07 Quote
Quote:
Originally Posted by schmidtbag
I don't see how hand-picking a GPU would make that much of a difference anyway. The reviewers all prove the clock settings, driver versions, OS, etc. When you've got over 6 billion transistors, you could probably cut out a few thousand and not have much of a performance impact.
<snip>
I sincerely think the test results shown for the 290X and 290 from any website are relatively pointless. Their results are so skewed by those ineffective heatsinks. Do some benchmarks with water blocks and then we've got some solid numbers to compare.

Because hand-picking anything for review does not show a typical example of your product, and leads to your customers losing faith in the reliability of any future favorable reviews.

AMD isn't selling cards with water blocks on so reviews are done on what is being sold, not what maybe possible on water or LN2.
somidiot 6th November 2013, 20:31 Quote
I thought there was something fishy when bit-tech and toms were so vastly different in their results.
The_Crapman 6th November 2013, 20:52 Quote
I for one can't wait for 3rd party cards to come out and we can forget about all this heat malarkey. Get them all overclocked to the 9's and see who's left standing. God only knows why AMD are making us wait so long. It's like a game of poker where both player have called, but are showing their cards one at a time. ;)
GuilleAcoustic 6th November 2013, 21:42 Quote
I want Kaveri :'(
Gareth Halfacree 8th November 2013, 08:21 Quote
Quick update: there's a beta driver out now which is claimed to fix the problem: seems that some cards were, for whatever reason, running their fans slower and reaching their target temperature faster. The update tweaks PowerTune to ensure all 290/290X boards run their fans at the same speed.
rollo 8th November 2013, 10:55 Quote
That beta driver breaks BF4 as it gives memory leaks, Do not upgrade if running bf4. Its all on the EA forums.
Assassin8or 8th November 2013, 18:37 Quote
Quote:
Originally Posted by Gareth Halfacree
Quick update: there's a beta driver out now which is claimed to fix the problem: seems that some cards were, for whatever reason, running their fans slower and reaching their target temperature faster. The update tweaks PowerTune to ensure all 290/290X boards run their fans at the same speed.

Have Bit-tech investigated whether this issue affects their card, and if so, will you be rerunning results and updating the article/issuing an updated article?
Gareth Halfacree 9th November 2013, 10:08 Quote
Quote:
Originally Posted by Assassin8or
Have Bit-tech investigated whether this issue affects their card, and if so, will you be rerunning results and updating the article/issuing an updated article?
No idea - I just do the news!
mi1ez 10th November 2013, 05:16 Quote
If anyone's interested, Tom's has a follow up where they look into the differences between the press card, and those bought on Newegg and it appears that the retail models (an Asus and a Sapphire) have lower fans speeds that are causing the GPU to slow:

http://www.tomshardware.co.uk/radeon-r9-290-driver-fix,review-32821.html
Harlequin 10th November 2013, 08:57 Quote
Quote:
Originally Posted by mi1ez
If anyone's interested, Tom's has a follow up where they look into the differences between the press card, and those bought on Newegg and it appears that the retail models (an Asus and a Sapphire) have lower fans speeds that are causing the GPU to slow:

http://www.tomshardware.co.uk/radeon-r9-290-driver-fix,review-32821.html

so toms after the link bait headlines are now trying to backtrack fast.....
Spreadie 10th November 2013, 11:21 Quote
Quote:
Originally Posted by Harlequin
so toms after the link bait headlines are now trying to backtrack fast.....
Actually no. They are describing exactly the glitch that AMD have now released a patch for. How about you remove those red-tinted spectacles for a moment?
Harlequin 10th November 2013, 11:27 Quote
Quote:
Originally Posted by Spreadie
Actually no. They are describing exactly the glitch that AMD have now released a patch for. How about you remove those red-tinted spectacles for a moment?

and I`ll place them ontop of my GTX 670 then....


toms were caught flaming the crap out of AMD , then got caught hard and got slammed for being an NVidia shill site
Corky42 10th November 2013, 11:37 Quote
So they didn't identify a problem that AMD had no knowledge of, and didn't test to see if the fix AMD implemented solved the problem they first reported ?
AlienwareAndy 10th November 2013, 11:48 Quote
Quote:
Originally Posted by Corky42
Because hand-picking anything for review does not show a typical example of your product, and leads to your customers losing faith in the reliability of any future favorable reviews.

AMD isn't selling cards with water blocks on so reviews are done on what is being sold, not what maybe possible on water or LN2.

Even if it were cherry picking it would be no different to Haswell would it?
Harlequin 10th November 2013, 11:58 Quote
Quote:
Originally Posted by Corky42
So they didn't identify a problem that AMD had no knowledge of, and didn't test to see if the fix AMD implemented solved the problem they first reported ?

actually toms said amd were cherry picking sampels - which is proven to be bollcoks - also , this `issue` has so dfar affected 4 boards out of all the thousands made .... so toms is once again NVidia link baiting.
Spreadie 10th November 2013, 12:02 Quote
Quote:
Originally Posted by Harlequin
and I`ll place them ontop of my GTX 670 then....
Your point being?
Quote:
Originally Posted by Harlequin
toms were caught flaming the crap out of AMD , then got caught hard and got slammed for being an NVidia shill site
I have no love for Tom's, but their review results were genuine - AMD have admitted that not all cards' cooling fans were running as intended, and had to release a fix - thankfully in the drivers, because a hardware recall would be a massive blow. If TH are a nvidia fan site, as you state, does that mean that every time they say anything negative about the opposition it should be automatically ignored?

They may have over-egged the pudding, with claims that the review samples were cherry picked but, regardless of their supposed bias, the concerns were valid.

The 290/290x looks like a stellar performer, but I wouldn't touch one with a reference cooler - a lot will depend on the premium added by third party coolers.
Corky42 10th November 2013, 12:45 Quote
Quote:
Originally Posted by AlienwareAndy
Even if it were cherry picking it would be no different to Haswell would it?

So comparing Haswell chips that didn't OC as much as engineering samples with a GPU that down-clocks due to over temp protection is the same how ? One runs fine at stock speeds and doesn't allow the same overclocking potential, the other refused to run at stock speeds when put under load.
Either way cherry picking is bad for both the company doing it, and the people that review any hardware. Just because someone does something wrong doesn't mean its OK for others to do it.
Quote:
Originally Posted by Harlequin
actually toms said amd were cherry picking sampels - which is proven to be bollcoks - also , this `issue` has so dfar affected 4 boards out of all the thousands made .... so toms is once again NVidia link baiting.

Spelling mistakes aside Toms didn't accuse AMD of cherry picking, that's what other sites reported. They actual raised a valid question of consistency, and they also shared this information with AMD prior to publication allowing AMD to identify a problem and issue a fix. Would you rather they stayed silent and allowed not the 4 cards as you claim but maybe hundreds of cards to remain with this problem ?
rollo 10th November 2013, 13:01 Quote
More AMD fan posts by none AMD owners. If everyone who defended AMD owned full AMD rigs I could understand the hate of nvidia. ( not to mension how rich AMD would be)

Toms did some basic checks notified AMD who then corrected the problem with a driver patch. Tom then retested again to see if it is fixed.

How is tom been biased again?

The fact that I still have not seen a user review suggests sales of the r290x are not as fast as all the AMD fans hope. Can't be all waiting for custom editions. Know a few with the 290 now all report similar story's fast card bad fan.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums