R9 290X Variance

[SimianR]

This is what SKYMTL of Hardwarecanucks had to say about the 290X variance issue:



Looks like he'll be getting some more retail cards to test. 4 cards with a 15% performance delta, that is utterly stupid.

It's nice that he's doing the right thing and getting more samples though. Not testing 1 review sample and 1 retail card and then making the claims. It does look like there's variance between the cards. If in fact it has something to do with the bios - what exactly is the difference between two review bios and retail bios? Is that all thats at play here?

 

[Silverforce11]

Well I changed the TIM on one of my cards, and compared to the default one, it throttles much less mining LTC, everything left on auto settings.

So because the cooler is at its limits, even minor variations in TIM application will yield different performance.

AMD made a massive mess by cheaping out on the cooler.

 

[krumme]

We know from marketing studies that consumers prefer products with less variance in quality, when they are compared to similar product with an on average above quality but have more variance.

Amd marketing competence is as always next to no existing. We knew before we would get the usual crappy coolers but this time it was worse because the cards as we have seen from the start is very sensible to trotling.

I think as consumers we should get used to more variance as we use the cards in different setups. Its natural. It gives us better product. But clearly the sensitivity of 290 and variance it can lead to is over the top and unacceptable.

But i think its save to asume its the last time we get those crappy coolers after a long consumer killstreak from amd. Lol.

 

[selni]

Isn't this actually hinting that AMD didn't cherrypick review cards? It all comes down to what that bios update changed, which they unfortunately didn't try to measure.

We know the reason for the performance drop is thermal throttling. We know the bios update removes the performance difference. Doesn't it stand to reason then that the bios update plays with the fan and/or voltages? AMD have already said there were issues with fan RPM in the retail units.

Presumably this can be addressed via driver changes with the new version of powertune, but it's still quite the mess.

 

[Silverforce11]

I can confirm the baseclock on the R290s, it does not drop lower than 662mhz while mining LTC, it just forces override on the fan cap of 47% and goes as high as it needs to maintain 95C @ 662mhz. It actually needs around 50% fanspeed to maintain that low throttled clocks mining. Which is insane how hard LTC thrashes these GPUs.

 

[JDG1980]

Well I changed the TIM on one of my cards, and compared to the default one, it throttles much less mining LTC, everything left on auto settings.

So because the cooler is at its limits, even minor variations in TIM application will yield different performance.

AMD made a massive mess by cheaping out on the cooler.

This article from Tom's Hardware appears to back up your hypothesis. The reviewer got substantial improvements in performance after replacing the TIM, which had been applied too thick. Oddly, there were also scratches on the bottom of the cooler plate. The article indicates that this was a retail card, not a review sample.

AIBs often tinker with the BIOS as well as the board and cooler. There's no reason to believe that these variations will persist once we get some decent designs. As I've said before, AMD should have had AIB boards available from day 1, and should have reserved the reference boards for water cooling only. Why they chose to unnecessarily sabotage the launch of their own flagship product in such a manner is beyond me.

 

[BallaTheFeared]

I can confirm the baseclock on the R290s, it does not drop lower than 662mhz while mining LTC, it just forces override on the fan cap of 47% and goes as high as it needs to maintain 95C @ 662mhz. It actually needs around 50% fanspeed to maintain that low throttled clocks mining. Which is insane how hard LTC thrashes these GPUs.

I think I saw a base clock for the 290x, I believe it was low 700s... I thought maybe Ryan listed it, but I can't remember where exactly I saw it.

 

[krumme]

Isn't this actually hinting that AMD didn't cherrypick review cards? It all comes down to what that bios update changed, which they unfortunately didn't try to measure.

We know the reason for the performance drop is thermal throttling. We know the bios update removes the performance difference. Doesn't it stand to reason then that the bios update plays with the fan and/or voltages? AMD have already said there were issues with fan RPM in the retail units.

Presumably this can be addressed via driver changes with the new version of powertune, but it's still quite the mess.

If this isnt already solved with a driver update it easily is. As a consumer you can also slide the fan speed up. Its very easy and its not like anyone is not capable of it. So its a bit of overblown imho. But its at the expense of added noise. But does that comes at a surprise? Lol.

Its old news and its beaten to death. Anyone caring for noise should wait for better coolers or go water. If anyone didnt knew that by now they will not have the iq to unbox the card so its a non existing problem. We have people in this forum who have written the excact same point in about 178 post resulting in permanent brain damage to some of us.

 

[Gloomy]

Mine doesn't throttle, I have it set to 55% fanspeed and it refuses to go past 46% :/

I have it up to 1050MHz so far, will finish overclocking this weekend maybe. But I am also looking into water cooling so not too keen on wasting time overclocking on air atm.

 

[selni]

And that is actually good safety and will hopefully come in all cards in the future from all vendors.

Complaining about it is preventing good tech. And that functionality is actually really good as it prevents crashing or damage.

Lets separate things here.

In the future? Throttling to avoid exceeding thermal limits isn't new, all current gen CPUs and GPUs do that. The only thing that is new with the 290 series is that the target clockspeeds are often not obtainable with the stock cooler.

 

[SiliconWars]

In the future? Throttling to avoid exceeding thermal limits isn't new, all current gen CPUs and GPUs do that. The only thing that is new with the 290 series is that the target clockspeeds are often not obtainable with the stock cooler.

Depending on

1) Case airflow
2) Load (certain games do better than others)
3) What mode you run the card in

Is AMD also supposed to provide you with a PC and choice of games to play?

The fact is, in uber mode the 290X consistently performs near it's maximum frequency. Nowhere has AMD said it would do this, especially not in quiet mode. In fact the official release clearly states "up to 1 GHz engine clock" - http://www.amd.com/us/press-releases/Pages/amd-radeon-r9-290x-2013oct24.aspx

 

[Skurge]

Uh yeah. All that is interesting.

Anyone have any idea what the BIOS did to affect performance? For those guys that have multiple cards. Can you compare BIOS version and voltages?

Also, MSI didn't have any reference cards to allocate so I will only be getting cards in Jan. Hopefully custom ones.

 

[SiliconWars]

Likely the press bios is just finer-tuned so that the fan revs up more aggressively. It could also be something as simple as 1 decibel's worth of fan RPM that makes the difference, while being inaudible.

 

[Gikaseixas]

I tested an R9 290 (flashed to an R9 290X) recently on a friends PC and it is a great card for the price. It was faster in every game tested compared to my GTX780 MSI TF Gaming. My GTX780 at ~1200/6700 is matching his R9 290X at stock Uber mode. Once he overclocked to 1180/1550 on his R9 290X it was all bets off

Edit: 2560x1440. Games tested were TR, HMA, MLL and Arma 3.

It was loud though

I believe it. This is why i'm holding until non reference stuff hits.
Again, AMD is only hurting themselves with this delay

 

[blackened23]

I tested an R9 290 (flashed to an R9 290X) recently on a friends PC and it is a great card for the price. It was faster in every game tested compared to my GTX780 MSI TF Gaming. My GTX780 at ~1200/6700 is matching his R9 290X at stock Uber mode. Once he overclocked to 1180/1550 on his R9 290X it was all bets off

Yeah, once the variability issues are fixed, the 290/290X will be great cards. I think there are multiple approaches to fixing variability - the one obvious fix which doesn't really need to be discussed, adjusting the powertune algorithm, and having better oversight of AIB makers as to ensure that this stuff never happens (ie ensuring they're using the proper release BIOS). By all accounts that is what nvidia's green light does and seems to have been pretty successful in terms of ensuring quality.

I think the issues can be fixed. That said, this still leaves a very sour taste though - I mean, despite what a few will state here, not everyone scans tech news daily for the latest tips and tricks on which BIOS to use on their GPU; most folks don't mess with flashing their BIOS. With that being the case I really hope AMD can step up their oversight of AIBs and quality control to ensure that these variability issues don't ever happen - it isn't hard to imagine there being a lot of folks out there who don't necessarily read forums, who bought a product that performs less than expected.

Basically, it can just be summed up in that AMD needs their QA to step it up a bit.

 

[monstercameron]

Yeah, once the variability issues are fixed, the 290/290X will be great cards. I think there are multiple approaches to fixing variability - the one obvious fix which doesn't really need to be discussed, adjusting the powertune algorithm, and having better oversight of AIB makers as to ensure that this stuff never happens (ie ensuring they're using the proper release BIOS). By all accounts that is what nvidia's green light does and seems to have been pretty successful in terms of ensuring quality.

I think the issues can be fixed. That said, this still leaves a very sour taste though - I mean, despite what a few will state here, not everyone scans tech news daily for the latest tips and tricks on which BIOS to use on their GPU; most folks don't mess with flashing their BIOS. With that being the case I really hope AMD can step up their oversight of AIBs and quality control to ensure that these variability issues don't ever happen - it isn't hard to imagine there being a lot of folks out there who don't necessarily read forums, who bought a product that performs less than expected.

Basically, it can just be summed up in that AMD needs their QA to step it up a bit.

I don't get it. What sour taste? you get what you pay for...

 

[stahlhart]

Last warning -- review the first post of this thread, understand clearly what we are discussing here, stay on topic, stop derailing it, or out comes the ban hammer.
-- stahlhart

 

[Abwx]

Reading LR article i ve got the impression that their tests were unscientificaly performed since it was not done in a controlled environment with the reviewer even stating that due to hot weather they had some tests made with case temp that was 40-45°C rather than the 33°C posted in the graphs.

For whom has a very basic understanding of thermal resistances
and the induced temperatures deltas this sound as if a baker was recycled
as a tech reviewer...

 

[3DVagabond]

Reading LR article i ve got the impression that their tests were unscientificaly performed since it was not done in a controlled environment with the reviewer even stating that due to hot weather they had some tests made with case temp that was 40-45°C rather than the 33°C posted in the graphs.

For whom has a very basic understanding of thermal resistances
and the induced temperatures deltas this sound as if a baker was recycled
as a tech reviewer...

So, let's see, they increase the ambient temps, set the card on quiet mode, and it throttles. Not all throttle identically, so there's another, "AMD is cheating" conspiracy?

 

[Abwx]

So, let's see, they increase the ambient temps, set the card on quiet mode, and it throttles. Not all throttle identically, so there's another, "AMD is cheating" conspiracy?

Worse , they had 40-45°C case temp in some tests but still they
wrote 33°C case temp in their graphs even for thoses tests....

 

[Silverforce11]

Worse , they had 40-45°C case temp in some tests but still they
wrote 33°C case temp in their graphs even for thoses tests....

Thats down-right dirty. No legit *pun intended* reviewer would do such a thing.

 

[BallaTheFeared]

That is not how it went down.... You're reading the redo part.

AMD asked us for the retail BIOS that we were using and informed us today that there is no difference between the press and retail BIOS versions and asked us to triple check our numbers. When triple checking our numbers we saw the same results and then sometimes the results were very different. The temperature in St. Louis was 45F so our our ‘hot’ case temperature started out higher than 33C and was more like 40-45C. We were shocked to see that the first run was slow and the second run was faster. This test was done on the retail card with the retail BIOS.



A closer look at the data logs and you can see that AMD PowerTune was doing some funky stuff. In the chart above you can see the clock speed in the first run dropped down to 727MHz during and the fan speed was bouncing around 43-44% in the first 90 second FRAPS run. This is pretty much what we’ve seen throughout our testing and is no big deal. We then quickly reloaded and ran the benchmark again, but this time around the fan speed ramped up to 51% or 2700RPM and the clock speeds were faster. So, the GPU temperature remained at 94C in both runs, but for some reason PowerTune bumped up the fan speed in the middle of this run with the card in quiet mode. The higher fan speed allowed the clocks to be set higher and that is why the benchmark numbers improved the second time around. So, it appears that when the ambient case temperature is around 40-45C that the AMD Radeon R9 290X starts to do some odd stuff. We saw the average clock speed drop to 727MHz for extended periods of time and then the fan profile ignore the maximum default setting. The air temperature of the room or case continues to play a huge roll in how the AMD Radeon R9 290X performs. If you test the card in quiet mode in a room at 18C and then again at 40C, you’ll have two different user experiences.

lmao 'nothing to see here' says AMD...

 

[Silverforce11]

Thats pretty funky, card detects high ambient case temps, overrides quiet mode and allow it to "breath better" albeit more noisy. An ingenious solution to clunky old cases with poor airflow or something more sinister? lol

Actually, i've just confirmed this. On auto settings, if i close my case, ambient MB temps get to 50C (its hot here nearing summer) and the thing ignores the 47% fan cap and goes nuts while mining LTC, activating jet engine mode BUT it sustains higher clock speeds. If i open the case, ambient drops to ~35C and it maxes at 47% fan speed but down clocks to 662mhz on the top card and ~770mhz on the bottom card mining LTC.

Very interesting indeed!!

 

[blackened23]

So that makes four websites that found differences between press and retail cards? So if you don't use the AC 24/7 and have high ambient temps, you're basically screwed if you own a 290 or 290X.

lmao 'nothing to see here' says AMD...

I'm frankly astonished that this kind of crap got past AMD's QA. What a botched launch, I can't believe they didn't do extensive testing of their product in a variety of ambient conditions. 50C is probably the norm for a good portion of the hot and humid southeastern US, unless they run the AC 24/7.

 

[Silverforce11]

Not screwed, just go deaf due to the crazy fan speed, it automatically overrides the fan setting cap and goes as high as it needs. It got up to 75% fan on the top card mining!! Hilariously loud, if I weren't working on the rig.. which I am.. so its not funny at all actually.

I have some water loop stuff coming in next week, should be all sweet then.

Edit: This launch is a wet-dream for companies making water cooling components (EK, Aquacomputer R290X blocks sold out everywhere here) as well as Artic Cooling and GELID.