AMD Radeon R9 290X reviews

It's not all that simple, TechReport for example had 2 review samples, which of 1 acted pretty much identically with the retail samples. They use only the better review sample for the conclusions.
Also, when they tested the better review samples BIOS on retail card, they used the overall worse performing retail card for the tests.

It's clear that some review samples are better than some retail samples, but there's evidence of retails doing just as fine as review samples too (on sites which only had 1 review sample, too)

 

Did anybody else besides TR put out an article comparing retail cards bought by Nvidia to a press sample?

 

Not really. http://techreport.com/blog/25751/an-update-on-radeon-r9-290x-variance

 

PC Perspective did 1 retail vs 1 review sample as did SweClockers. Tom's had 2 retail samples I think but posted results only for 1, and not sure if they used the driver update to fix fanspeeds.

http://www.sweclockers.com/artikel/...der-for-golden-samples-sweclockers-undersoker
http://www.tomshardware.com/reviews/radeon-r9-290-review-benchmark,3659-2.html
http://www.pcper.com/reviews/Graphi...-290X-Frequency-Variance-Issues-Still-Present

 

Sapphire released today too a 3fans coolers version:

http://www.guru3d.com/news_story/sa...and_r9_290x_with_tri_x_triple_fan_cooler.html



I can bet, MSI, HIS and other ones will be annonced thoses next days.

 

And we'll probably see cards with bigger factory overclocks somewhere down the line. For now I think AMD's partners are mostly worried about getting some custom-cooled cards out as quickly as possible. In any case, this ought to shake things up a bit more than reference 290(X)s did.

 

Uh, yes really. Unless you put the card into Uber mode I guess, which is not the default configuration. Besides the cards are already loud in "Quiet" mode. The consensus is that the Hawaii retail cards are slower than the press samples in the default quite mode.

I don't think you guys realized how much variance there would be when you sent out the review samples. At least that's my optimistic take on it.

The custom cooled cards will be dynamite - especially the R9290. If I were in the market for a new GPU, I'd definitely wait for that. The NVIDIA stuff in that performance range it too expensive. Plus the Hawaii cards ship with more VRAM which is nice going forward.

 

Why isn't it default? Either mode is - one is called quiet the other is called uber, not "default" and Uber.

And that doesn't necessarily result in them being unreasonably so.

http://www.hardocp.com/article/2013/12/13/4_weeks_radeon_r9_290x_crossfire/#.UrOQcmB3vuo

In quiet mode the variability is certainly going to be higher because it is acoustically (thermally) limited and this also means that run-to-run variability will increase in uncontrolled test environments. The fact that different results have been obtained when a different the press BIOS was flashed actually points to run-to-run variability because the BIOS's had the same settings.

 

I really don't understand this "we must test everything in restricted, quiet mode because that's they way it comes out of the box". People who bought and installed a high end graphics card for hundreds of dollars are not going to have a problem pushing a little switch on the side of their cards.

Cars are not tested in one gear because "that's the way they come". TVs and monitors are not just tested at default settings. The 290s come with two standard configurations, and both should be used and reported on. Instead the usual suspects only test the lower performing mode and ignore the higher performing one.

 

Quiet mode is literally the default configuration. It comes that way out of the box. Am I misunderstanding this?

Initial reviews are misleading in that the performance we saw in Quiet mode is not the performance you are likely to get in Quiet mode if you go out and buy a card for yourself.

This is up for debate and since I haven't heard them myself I can only go by what I read. Anandtech seemed to think the card was just too damn load for recommendation even in Quiet mode. This was before the fan speed tweak, so perhaps his card was running at a higher RPM?

Still don't think I'd pay an extra $100 for the GTX780 over the 290, and again the 290 with aftermarket cooling is going to shake things up in a very good way.

 

It's not that simple, some review samples have higher performance than some retail samples, but it's not "universal rule". Same applies to all GeForces with Boost, but no-one raised any fuss about that.

 

I don't think it's nearly the same thing. They don't even set a base clock, only an advertised and wildly optimistic (in quiet mode) boost clock.

I like the way it works on my 7950 Boost and GTX670. 7950 Boost has a base clock of 850MHz with a (rather conservative) boost clock of 925MHz. It will actually overclock to 1GHz easily.

Works pretty much the same on my 670 except the 670 seems to constantly stay above the advertised boost clock. I've actually never seen it run at or below the boost clock.

 

That only matters psychologically. You didn't buy your GTX 670 because it was advertised at a certain clock speed, you bought it because reviews indicated a certain level of performance. The only question is whether your retail sample actually performs at that level.

 

And it does. If I bought a 290X expecting to run it in Quiet mode with the performance we saw in most of the reviews, that would not be the case.

Maybe what I'm trying to say is I dislike the new Powertune system. Just giving a guaranteed base clockspeed would be a nice improvement. We can only speculate as to why they didn't, but it seems the reason is because the actual base clockspeed is very low (in the 700-800MHz range). A better reference cooler would also be much appreciated.

 

I absolutely can't see AMD doing a new reference cooler at this point. Too many third party cards are arriving right now with better coolers, so AMD is not going to pull the rug out from under their partners' feet. I doubt that AMD have the will or had the time to do a redesign.

I hope to see something better for 20nm cards in a year's time, but who knows looking at AMD's past record for minimum effort on coolers?

 

Assuming you're referring to our initial reviews, 290X quiet mode was good enough. 290 doesn't have a quiet mode, and its one fan configuration was a quite a bit louder than 290X's (so that was the card we made a big deal of).

However now the fan speed changes have halved the gap between 290 and 290X (290X got louder).

 

Exactly why wouldn't it be the case? Because some specific reviews have noted it's not the case on their sample size of couple cards? But what about the sites that have noted that there's no difference between their sample and retail?

Just like you got GTX 670 that performs the same as in reviews, you can get a R9 290 or R9 290X that performs like in the reviews, in Quiet-mode too.

 

Right, sorry I was confused.

Eh, NVIDIA was confident enough that they drop shipped a couple of cards directly from Newegg to Scott at TR. Unless you believe there is some conspiracy going on there I'd say the level of variation between retail and review Hawaii chips is out of the norm.

 

I like the new PowerTune system but not because of what it's causing to cards thermally limited.

Let me explain! I often game with vSync or frame limiter ON purely because I can't see more than 60FPS on my 60Hz TFT. Gaming with vSync OFF gives me 80-200FPS in games which are simply turned into wasted heat as card is taxed 100% all the time. With vSync ON I limit that to maximum my screen can output and as added bonus limit power consumption by quite some margin. Simple example would be my system power while playing BF3 on single HD7970 1050MHz/1425:
- vSync ON = around 280-345W (60FPS)
- vSync OFF = around 380-420W (60FPS - 100FPS)
With old PowerTune in both cases GPU clocks were kept @1050MHz as both loads generated enough work to keep core clock in 3D mode.
New PowerTune is responding so quickly to changes in load that I can see GPU clocks to adjust depending on current frame demand and therefore oscillating usually at lower than maximum speeds. Another example, this time from two different games played on R9 290X @1050/1375:
- Sleeping Dogs vSync ON: GPU 720MHz-1050MHz (average around 800MHz)
- Crysis 3 vSync ON: GPU 920MHz-1050MHz (average around 1000MHz)
[No meaningful power data as I so far hadn't have chance to take it because my second R9 290 card constantly mines coins while I play skewing any readings]
What new PowerTune does is not only on-the-fly core clock adjustments but also vGPU adjustments tracking core speed changes. This means for vSync enabled players like me R9 290(X) quite often should be more energy efficient than older generation Tahiti based GPU's. It's something I would love to see being investigated by reviewers when doing power measurements, as for now they only concentrate on max. gaming power and Furemark numbers. I think Hawaii boards would fare significantly better in these tear free vSync Enabled scenarios as people tend to think now.

If I had time I would do straight 7970 vs R9 290X shoot-out!

 

And had TechReport only had their 2nd press sample, it would have performed the same as the 2 retail samples they had, by all indications there their first sample was just abnormally good, not the other way around