videocardzAMD Radeon R9 290X confirmed to feature 64 ROPs

[blackened23]

I have a feeling Hawaii will only start pulling away from GK110 at 4K/Eyefinity resolutions, while going back and forth at lower resolutions.

Also, if AMD and Nvidia made absolutely equivalent hardware Nvidia would probably come out ahead due to their better software expertise. For the last few generations it seems that ATI/AMD has had the edge in hardware, while Nvidia has had the edge in software. Ironically, between Mantle and AMD's need to cut expenses I can see their positions switching around in the future.

The impression I get from ocUK and xtremesystems is that the 290X beats the GTX 780 consistently and trades with the Titan. It certainly won't "crush" Titan as that wording would indicate clear wins across the board - In terms of trading with the Titan, it seems the Titan may win more (let's say, 60-40 favoring Titan) but that is fine as long as the 290X is priced correctly. Then again, both of these sites focused mainly on synthetic benchmarks. They haven't hinted directly at any gaming benchmarks, but they ran the gamut of Unigine, firestrike, and stuff along those lines. Regarding price, I still think it should be 550-600$ and that would be perfectly fine given the performance.

It is also rumored to overclock well on air (the 290X, that is). So generally speaking a 290X overclocked would match any GTX 780 overclocked in terms of performance, but there are many other overclocking variables which we don't know.

Basically, the only real target for the 290X is the GTX 780 overclocked - as the 780 made the Titan irrelevant long ago. So AMD should situate their pricing as such to be very much appreciably lower than the GTX 780 OC editions. If AMD had software on par with NV, then pricing parity would be fine, but given things as they are (unless they change in the next week) I think 550-600$ is the best pricepoint for the 290X - considering it beats the 780 pretty much nearly 100% of the time in stock configurations and is rumored to OC pretty well on air. In my mind, overclocking is the most important wild card.....for the average lay person who doesn't overclock, maybe not.

 

[toyota]

I am still shocked they got all of that crammed into a gpu that size on 28nm. I really hope its faster than Titan and costs $599 or less so we can start getting some price adjustments from Nvidia's arrogant butt.

 

[cmdrdredd]

Dont think so,

http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/18

He specifically said GTX 7xx series and provided a link to a benchmark from yesterday. Nice try

 

[smackababy]

The impression I get from ocUK and xtremesystems is that the 290X beats the GTX 780 consistently and trades with the Titan. It certainly won't "crush" Titan as that wording would indicate clear wins across the board - In terms of trading with the Titan, it seems the Titan may win more (let's say, 60-40 favoring Titan) but that is fine as long as the 290X is priced correctly. Then again, both of these sites focused mainly on synthetic benchmarks. They haven't hinted directly at any gaming benchmarks, but they ran the gamut of Unigine, firestrike, and stuff along those lines. Regarding price, I still think it should be 550-600$ and that would be perfectly fine given the performance.

It is also rumored to overclock well on air (the 290X, that is). So generally speaking a 290X overclocked would match any GTX 780 overclocked in terms of performance, but there are many other overclocking variables which we don't know.

At ~$600, they would be competing with the 780 on price and the Titan on performance. That would score a huge win for AMD. And if the 290 was 60/40 against the 780, at <$500, it will be an amazing card.

 

[AtenRa]

He specifically said GTX 7xx series. Nice try

He was showing the GTX770 which is a GTX680 at higher clocks.
Also, AT choose to change the Tessellation Benchmark they used exactly when TITAN was released. :whiste:

 

[blackened23]

At ~$600, they would be competing with the 780 on price and the Titan on performance. That would score a huge win for AMD. And if the 290 was 60/40 against the 780, at <$500, it will be an amazing card.

I agree with this but i'd also say for the more extreme guys (like us) overclocking becomes very much a value added proposition. I'm hoping the 290X does overclock as well on air as some have hinted at, if that is the case then a 600$ 290X would definitely be a hell of a card. I've mentioned it before but the only real pricing target for the 290X are the pre-overclocked GTX 780 cards. I'd think differently if NV weren't better than AMD in terms of software, but it is what it is in that respect. AMD can't charge what NV can for the same performance because of software (IMHO).

I know I sound like a broken record at this point with regard to software, but AMD must get their software on par with nvidia's level if they want the same type of brand strength/recognition that they get. Right now AMD's software isn't quite there.

 

[HurleyBird]

In terms of trading with the Titan, it seems the Titan may win more (let's say, 60-40 favoring Titan)

The interesting thing is that the guy on XS thought the 290X was faster at stock until he went and averaged the scores, and found that Titan was just a wee bit faster. Sounds like the two are close enough that it comes down to benchmark selection and/or margin of error. That said, the fact that he first thought that the 290X was faster may mean that it was winning more prestigious or challenging benchmarks with high AA or high resolution and that's what he was focused on. Makes sense with the ROP and bandwidth advantage.

 

[LegSWAT]

Higher transistor density is possible due to GCN1.1/2.0 "uMA" compatible memory architecture. Memory access is completely changed in GCN1.1/2.0, it has less possible access paths (ie. less overwriting operations, less access modes) which mean simpler design in the GPU.

Or at least that is what I understand it to be given the architectural changes, this was the major change.

Sounds interesting. Where do you get that info from?

 

[raghu78]

That's a part of it.

HD6970 = 389mm2, 2640 million transistors or 6.79 million / mm2
HD7970/GE = 365mm2, 4313 million transistors or 11.82 million / mm2
R9 290X = ~425mm2, ~6000 million transistors or 14.12 million / mm2
GTX780 = 561mm2, 7100 million transistors or 12.66 million / mm2

The transistor density between 780 and R9 290X is pretty close.

btw Hawaii GPU die size is 438 sq mm. thats from an official AMD spokesperson

http://www.maximumpc.com/amd_r9_290x_will_be_much_faster_titan_battlefield_4

"We had a chance to sit down with AMD Product Manager Devon Nekechuck to see how AMD&#8217;s new top dog R9 290X stacks up against the green team&#8217;s best single-GPU offering. According to Nekechuck, even though the R9 290X uses a 438 square mm die, which is significantly smaller than the Titan&#8217;s 550 sq. mm GK110 offering, it &#8220;will definitely compete with the GTX 780 and Titan.&#8221; "

3dcenter has the same die size. also the transistor count is 6.2 billion transistors. thats has to be officially confirmed. we will know exact die size in 5 days time.

http://translate.google.com/translate?sl=auto&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&u=http%3A%2F%2Fwww.3dcenter.org%2Ffrontpage%3Fpage%3D1

so Hawaii 6200 mn transistors / 238 sq mm = 14.15 million transistors per sq mm.

Pitcairn = 2800 / 212 = 13.2
Bonaire = 2080 / 160 = 13

beats both these GPUs by a significant 1 - 1.2 million transistors per sq mm

AMD must have re-designed the GPU from scratch not just widened it. How can you have 2x the ROPs, 512-bit memory controller, nearly 40% more TMUs/steam processors but die size goes up only 16% from Tahiti XT? Something is definitely smaller inside Hawaii than Tahiti. The memory controller is presumably smaller than 384-bit one in Tahiti, but what else?

Hawaii is a from scratch GPU incorporating a lot of learning on the TSMC 28nm process. AMD have gone super aggressive on transistor density because the TSMC 28nm process variables are well understood and the maturity of the process is high. yields are also very good.

With 64 ROPs, I have a feeling the gap in performance between 780 and R9 290X will grow even more now in AMD GE titles.

And in games where NV was winning (FC3, Metro LL), AMD has now made up ground which means it will no longer be an "easy win" for 780.

I think the 780 will now lose to the R9 290X in both Metro LL and FC3/Blood Dragon games. In other GE titles like Tomb Raider, I don't think 780 has hope.

by now its almost pretty clear that R9 290x will edge ahead of even Titan at high resolution and high MSAA / SSAA .

 

[Grooveriding]

I really think AMD should of delayed this launch until October 29th and sent copies of BF4 with the cards to reviewers :sneaky:

For certain BF4 will be getting benchmarked heavily after it's out, but it would of made for good reviews for it to be available at launch. Sites won't even be able to use the beta because it's shutting down the day after the 290 cards are launching. Actually I guess they could as they bench these cards prior to release. Not sure how many will though.

 

[cmdrdredd]

He was showing the GTX770 which is a GTX680 at higher clocks.
Also, AT choose to change the Tessellation Benchmark they used exactly when TITAN was released. :whiste:

Yeah cause AT is bias now right? Geeze...

 

[raghu78]

I really think AMD should of delayed this launch until October 29th and sent copies of BF4 with the cards to reviewers :sneaky:

For certain BF4 will be getting benchmarked heavily after it's out, but it would of made for good reviews for it to be available at launch. Sites won't even be able to use the beta because it's shutting down the day after the 290 cards are launching.

http://techreport.com/review/25466/amd-radeon-r9-280x-and-270x-graphics-cards/5

BF4 beta benchmarked on an R9 280X review from TR. :thumbsup: you can bet this month end almost very site will have benchmarks of BF4 on all the high end and mid range cards from both vendors. and guess what R9 290X is most likely to be end up the fastest card for BF4.

 

[Cloudfire777]

I am still shocked they got all of that crammed into a gpu that size on 28nm. I really hope its faster than Titan and costs $599 or less so we can start getting some price adjustments from Nvidia's arrogant butt.

7970GHz: 2048 cores @1000MHz
R9 290X: 2816 cores @ 1000MHz

Bigger bandwidth on the 290X, but that was expected since the new cores need bigger lane.

2816/2048 = 1.375

83 * 1.375 = 114

Cores do however not scale perfectly linearly.

Since Titan is at 107, that is most likely where the R9 290X will be too. So no, it won`t be the "Titan killer" some people talked about earlier.

Unless, the cores have become more efficient on Hawaii than GCN and AMD have not used that advantage to lower the power draw.

 

[blackened23]

http://techreport.com/review/25466/amd-radeon-r9-280x-and-270x-graphics-cards/5

BF4 beta benchmarked on an R9 280X review from TR. :thumbsup: you can bet this month end almost very site will have benchmarks of BF4 on all the high end and mid range cards from both vendors. and guess what R9 290X is most likely to be end up the fastest card for BF4.

To be fair, it usually takes a couple of weeks for the dust to settle, so to speak, in terms of game performance. Post-driver fixed performance sometimes differs from launch performance - Tomb Raider is a good example of this. Another example being guild wars 2 - AMD did not have access to it until after launch, and the launch performance for AMD cards was below where it should have been (although it has been brought up to where it needs be with drivers).

I'm not saying that BF4 won't favor AMD (i'm not taking Mantle into consideration, only DX11), but again - when the dust settles with game specific drivers, that can change the performance dynamic. Nvidia aren't idiots. They know BF4 will be the biggest title for the PC this year, and i'm sure they'll do their thing with drivers to bring their cards up to snuff. Whether it catches AMD, who knows.

My main point is that it is a bit pre-mature to declare AMD to be the all out winner in terms of BF4 DX11 performance. Calm down there a bit - things can change post release.

 

[toyota]

7970GHz: 2048 cores @1000MHz
R9 290X: 2816 cores @ 1000MHz

Bigger bandwidth on the 290X, but that was expected since the new cores need bigger lane.

2816/2048 = 1.375

83 * 1.375 = 114

Cores do however not scale perfectly linearly.

Since Titan is at 107, that is most likely where the R9 290X will be too. So no, it won`t be the "Titan killer" some people talked about earlier.

Unless, the cores have become more efficient on Hawaii than GCN and AMD have not used that advantage to lower the power draw.

since the cores are faster clock for clock than before that is at least a 50% increase. so that along with the other improvements better yield more just a 25% performance increase or this is the most inefficient architecture ever.

 

[raghu78]

7970GHz: 2048 cores @1000MHz
R9 290X: 2816 cores @ 1000MHz

Bigger bandwidth on the 290X, but that was expected since the new cores need bigger lane.

2816/2048 = 1.375

83 * 1.375 = 114

Cores do however not scale linearly, not perfect that is.

Since Titan is at 107, that is most likely where the R9 290X will be too. So no, it won`t be the "Titan killer" some people talked about earlier.

you realize that Tahiti was a bottlenecked design with the same front end as Pitcairn but with 60% more shaders. with the doubling of front end the perf per sp is going to improve in Hawaii wrt Tahiti.

btw that perf chart has changed with 13.11 drivers

http://www.techpowerup.com/reviews/MSI/R9_280X_Gaming/26.html

MSI R9 280X . 1050 mhz clocks. same as HD 7970 GE. Titan is 24% faster. to know how it happened see Starcraft II perf. 30% improvement over HD 7970 GE with older drivers. that game was affecting the TPU perf average in a meaningful way.

 

[Grooveriding]

http://www.youtube.com/watch?v=hZQANNndqOA&feature=c4-overview&list=UUXuqSBlHAE6Xw-yeJA0Tunw

Saw the upvoted comment on Linus' 280X review video.

At 8:08 you can see a R9 290X or R9 290 on the test bench in the background. Guess if reviewers have these already we are getting reviews on the 15th ?

 

[boxleitnerb]

50%? No, definitely not. I would expect 35% tops.
7970 GHz clocks at 1050 MHz (although not always, but most of the time), R9 290X will clock lower. And the better frontend and backend will be only really useful at lower resolutions and with lots of MSAA (like 8x) respectively.

 

[Erenhardt]

Sounds interesting. Where do you get that info from?

Good luck with this one! :ninja:

 

[raghu78]

To be fair, it usually takes a couple of weeks for the dust to settle, so to speak, in terms of game performance. Post-driver fixed performance sometimes differs from launch performance - Tomb Raider is a good example of this. Another example being guild wars 2 - AMD did not have access to it until after launch, and the launch performance for AMD cards was below where it should have been (although it has been brought up to where it needs be with drivers).

I'm not saying that BF4 won't favor AMD (i'm not taking Mantle into consideration, only DX11), but again - when the dust settles with game specific drivers, that can change the performance dynamic. Nvidia aren't idiots. They know BF4 will be the biggest title for the PC this year, and i'm sure they'll do their thing with drivers to bring their cards up to snuff. Whether it catches AMD, who knows.

My main point is that it is a bit pre-mature to declare AMD to be the all out winner in terms of BF4 DX11 performance. Calm down there a bit - things can change post release.

BF4 is not just another title. It is the biggest PC title this holiday season. I am sure Nvidia and AMD have been working a long time on BF4 performance. there was an alpha performance review a few months earlier on gamegpu.ru and now a beta performance review on a lot of sites.

yeah after launch we need to revisit the situation. but I don't expect massive improvements on a title on which both AMD and Nvidia have been working closely with DICE for a long time.

 

[toyota]

50%? No, definitely not. I would expect 35% tops.
7970 GHz clocks at 1050 MHz (although not always, but most of the time), R9 290X will clock lower. And the better frontend and backend will be only really useful at lower resolutions and with lots of MSAA (like 8x) respectively.

where did anyone say a 50% performance increase? I was saying its at least 50% better in sp department.

 

[boxleitnerb]

where did anyone say a 50% performance increase? I was saying its at least 50% better in sp department.

What is that supposed to mean?
SP GFLOPs will increase by up to 40% (top clock bin is supposedly 1050-1070 MHz).

 

[ams23]

no it wasnt.

Yes it was. When the GTX 680 4GB was originally listed at newegg, it was somewhere around $629-649 while GTX 680 2GB was around $449-469. Anyway, this is a semantical argument because the price difference was relatively large and much more than $50 as asserted earlier.

 

[ams23]

they'll get more of them out of a wafer and should be able to sell them for less than both the 780 or Titan

Actually that is not necessarily correct. Due to having 4GB RAM in R9 290X instead of 3GB RAM as in GTX 780 SC, NVIDIA's RAM production cost will be lower. GTX 780 SC should be able to easily compete with R9 290X when it comes to overall perf. and perf per dollar.

 

[blackened23]

Yes it was. When the GTX 680 4GB was originally listed at newegg, it was somewhere around $629-649 while GTX 680 2GB was around $449-469. Anyway, this is a semantical argument because the price difference was relatively large and much more than $50 as asserted earlier.

I agree with Toyota on this one. Only the GTX 680 classified was advertised at that price, and it was widely panned because of the exorbitant price - the classified was priced at 650$ because of the improved PCB and was an extreme factory overclocked card (i'm sure you're aware of the EVGA classified line of products, they are expensive and not just because of VRAM). All of the other 4GB cards that I recall were in the 550$ range. MSI also had a 4GB GTX 680 in that price range, as well. I want to say that Gigabyte had a 4GB model near release that was right around 550$ - the only 600$+ 4GB card that I'm aware of was the classified. Again, the classified was priced as such due to being voltage unlocked with EV Bot at the time and was a card designed for world records and exotic cooling.

If i'm incorrect on this, by all means let me know. But i'm fairly sure that the classy was the only 600$+ 4GB 680 - there were definitely 4GB 680s in the 550$ range.