wccftechAMD Pirate Islands : R9 300 Series Alleged Specifications Detailed

[dangerman1337]

I hope you realize most games often have one or two graphics settings that destroys performance for very minimal image quality gains. You can disable that and get very high performance AND enjoy the benefits of higher clarity and more pixels.

I have never felt the need to max games just because, even when I have the hardware that can do it, features such as DoF, HDAO and Soft Shadows are the first on my list to be disabled. It means my power consumption is significantly less to maintain constant 60 fps.

Well, I do play BF4 with shadows, lighting, HBAO/SSAO and similar lowered and disabled at 1920x1080 (have MSAA 2x but wish there was a SMAA option) with my slightly OC'd 7950 but you do have to remember the future as well. TBH 4K is not remotely affordable, I'd agree that it'll be affordable in say Pascal or Volta (or AMD equivalent) but right now; not so much even if you lower settings that offer minimal image quality gains.

 

[raghu78]

Yes, 3D memory will proliferate the high end in 2016.

For AMD its most likely to happen with the next gen R9 390X in Q4 2014 . For Nvidia it will happen in 2016 according to their roadmaps revealed at GTC 2014.

http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

 

[ShintaiDK]

For AMD its most likely to happen with the next gen R9 390X in Q4 2014 . For Nvidia it will happen in 2016 according to their roadmaps revealed at GTC 2014.

http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

Stacked memory is nowhere near ready for mainstream usage. 2016+ for both parties is a good estimate. And Knights Landing will be the first product we see with it with Micron stacked memory.

 

[TreVader]

7970Ghz is about 10% faster than 680 and continued to be so for a while after Nov 2013 for sure. Even on release it beat the 680. Right now at 1600P, 770 is a competitor to the 7970Ghz not 680.
http://www.computerbase.de/2014-04/amd-radeon-r9-295x2-benchmark-test/5/

Not trying to derail the thread as 7970Ghz/R9 280X/770/680 has been beaten to death. We can conclude that it's pointless to compare NV vs. AMD specs such as CUDA cores and SPs on paper. Trying to formulate 880's performance based on specs or R9 390X's performance is pointless since both the 880 and R9 390X will have higher IPC courtesy of Maxwell and GCN 2.0 architectures.

What we need to know are prices too. If 880 is slower than R9 390X but it costs $499 while R9 390X costs $649 and is only 10% faster? All of this is just speculation. We can't say how powerful 880 will be since 750Ti kicks 650Ti's ass but on paper specs, it's not much better vs. it's real world performance increase.

I would give NV the edge next round since Maxwell has already shown a 2x increase in performance/watt and 35% increase in IPC and that was only on 28nm! I would never underestimate NV's Maxwell on 20nm. What pisses me off more is the notion of NV launching GM204 as 880 and trying to pass that off as flagship again.



Cannot accurately compare NV vs. AMD on specs alone when both will debut with new/significantly revised architectures and on newer node. We can't even compare 750Ti vs. 660 on paper specs but you are trying to do that for R9 390X vs. 880?

On paper 660 has 50% more CUDA cores, 100% more TMUs, 50% more ROPs, 67% more memory bandwidth (144 vs 86) against the 750Ti but is only 20% faster than 750Ti:
http://www.computerbase.de/2014-02/nvidia-geforce-gtx-750-ti-maxwell-test/5/

Therefore, you are just wasting your time trying to both extrapolate 880's performance relative to 780Ti and especially its performance relative to R9 390X. Comparing 7970 vs. 680 actually proves that comparing NV vs. AMD specs on paper is generally a shot in the dark. For starters, other bottlenecks exist in the GPU such as rasterization/geometry power and real world efficiency that imply comparison of 2048 Stream Processors to 1536 CUDA cores is a wasted effort. For example, R9 290X has 2816 SPs vs. 2880 CUDA cores for 780Ti and their performance is fairly close but yet 680 is very close to 7970 despite far inferior paper specs.

The 750ti boosts to 1350 stock so that's why it looks like it "magically" performs better than the 650ti. I'm sure maxwell architecture helps, but you can't discount a 50-60% increase in clock speed and pretend that it's all maxwell.

These specs may be wrong, but it's all the info we have. The one big advantage maxwell has is that it doesn't appear to need a huge amount of memory bandwidth, which will be necessary w/ that 256bit bus.

 

[RussianSensation]

The 750ti boosts to 1350 stock so that's why it looks like it "magically" performs better than the 650ti. I'm sure maxwell architecture helps, but you can't discount a 50-60% increase in clock speed and pretend that it's all maxwell.

AT reported 1150mhz as the max boost. Also, compare it to the 660 instead of the 650Ti to see that specs on paper would have meant that 660 should technically destroy the 750Ti but it doesn't.

http://www.anandtech.com/show/7764/the-nvidia-geforce-gtx-750-ti-and-gtx-750-review-maxwell/23

For AMD its most likely to happen with the next gen R9 390X in Q4 2014 . For Nvidia it will happen in 2016 according to their roadmaps revealed at GTC 2014.

http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

I don't see how it's possible for AMD to be more than a year ahead of NV with such cutting edge technology.

The way I look at it is NV's Kepler architecture's best chip (780Ti) beats AMD's newer GCN 1.1 chip (290X). Since Maxwell will bring about a 35% increase in IPC, unless NV makes a die much smaller than 550mm2 and/or AMD produces a similar increase in IPC with GCN2.0 and/or increases die size beyond 440mm2, NV will only extend the lead next generation.

It's like taking Kepler, adding 35% on top and making a 530mm2 die. How is AMD going to combat this exactly with a 450mm2 GCN 1.1 chip? I don't see it.

 

[TreVader]

AT reported 1150mhz as the max boost. Also, compare it to the 660 instead of the 650Ti to see that specs on paper would have meant that 660 should technically destroy the 750Ti but it doesn't.

http://www.anandtech.com/show/7764/the-nvidia-geforce-gtx-750-ti-and-gtx-750-review-maxwell/23



I don't see how it's possible for AMD to be more than a year ahead of NV with such cutting edge technology.

The way I look at it is NV's Kepler architecture's best chip (780Ti) beats AMD's newer GCN 1.1 chip (290X). Since Maxwell will bring about a 35% increase in IPC, unless NV makes a die much smaller than 550mm2 and/or AMD produces a similar increase in IPC with GCN2.0 and/or increases die size beyond 440mm2, NV will only extend the lead next generation.

It's like taking Kepler, adding 35% on top and making a 530mm2 die. How is AMD going to combat this exactly with a 450mm2 GCN 1.1 chip? I don't see it.

AMD already has much better shader density than nvidia, and it remains to be seen what maxwell will look like on an immature 20nm process. Also, you do realize that AMD can decide to make a 550mm die whenever they want, they would just have to cut margins (just like nvidia does).

The only reason Kepler is "faster" (it's actually something like 20% slower than hawaii at 4k with mantle) is it's a HUGE die.


AMD has consistently shown they can do more with less than nvidia. And these specs say that nvidia has already lost the performance crown. 3200sp, even with the IPC improvements, isn't going to come close to competing with nearly 2x the ROPs, 2x the bandwidth, and 1000 extra shaders.

 

[n0x1ous]

AMD already has much better shader density than nvidia, and it remains to be seen what maxwell will look like on an immature 20nm process. Also, you do realize that AMD can decide to make a 550mm die whenever they want, they would just have to cut margins (just like nvidia does).

The only reason Kepler is "faster" (it's actually something like 20% slower than hawaii at 4k with mantle) is it's a HUGE die.


AMD has consistently shown they can do more with less than nvidia. And these specs say that nvidia has already lost the performance crown. 3200sp, even with the IPC improvements, isn't going to come close to competing with nearly 2x the ROPs, 2x the bandwidth, and 1000 extra shaders.

So much wrong with this.... "cutting margins just like nvidia" Nvidia can make 550mm dies because of Tesla and Quadro. AMD doesnt have the professional or HPC business to support that investment.

The 880 specs we are seeing are clearly the successor to GK104 mid-range chip. We all know these numbers are not for the big daddy GM100/200 - so talk of losing the performance crown is ridiculously premature and probably not happening at all because AMD wont make a 500mm+ die..

 

[Makaveli]

AMD has consistently shown they can do more with less than nvidia. And these specs say that nvidia has already lost the performance crown. 3200sp, even with the IPC improvements, isn't going to come close to competing with nearly 2x the ROPs, 2x the bandwidth, and 1000 extra shaders.

Good points vader.

What RS was getting to before though is we still can't just look at the specs alone it may or may not be competitive. The whole ecosystem has to be looked at though not just the hardware.

Driver optimizations
Driver maturity
Game optimizations
etc

Are equally as important in the final outcome of performance.

I myself like to see a 50%-75% increase in performance before I upgrade so the R9 300 series maybe my next step but I will be waiting for performance reviews before I drop any cash.

 

[DiogoDX]

So much wrong with this.... "cutting margins just like nvidia" Nvidia can make 550mm dies because of Tesla and Quadro. AMD doesnt have the professional or HPC business to support that investment.

The 880 specs we are seeing are clearly the successor to GK104 mid-range chip. We all know these numbers are not for the big daddy GM100/200 - so talk of losing the performance crown is ridiculously premature and probably not happening at all because AMD wont make a 500mm+ die..

I heard the same about AMD not making a >300mm 28nm die.

 

[OCGuy]

For AMD its most likely to happen with the next gen R9 390X in Q4 2014 . For Nvidia it will happen in 2016 according to their roadmaps revealed at GTC 2014.

http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

Yea I don't think production is from Hynix is that close. They are barely getting their supply chain back to normal after the fire as it is.

 

[TreVader]

So much wrong with this.... "cutting margins just like nvidia" Nvidia can make 550mm dies because of Tesla and Quadro. AMD doesnt have the professional or HPC business to support that investment.

The 880 specs we are seeing are clearly the successor to GK104 mid-range chip. We all know these numbers are not for the big daddy GM100/200 - so talk of losing the performance crown is ridiculously premature and probably not happening at all because AMD wont make a 500mm+ die..

Do you know what the margins are? Serious question, because I doubt nvidia is LOSING money even when they sell a 780ti at 700 instead of a K6000 or whatever at 5k. I mean losing money outright, not losing profits.


As long as AMD makes even a small profit, it's worth it to go big. They're taking back a lot of marketshare, and with all the 290s hitting ebay now for as low as $310 each you'd have to be insane to buy a 780 or worse 780ti over 2x 290 for the same price.

 

[MrK6]

Specs on the 390X are impressive if true; a summer release would be fantastic.

 

[Silverforce11]

The way I look at it is NV's Kepler architecture's best chip (780Ti) beats AMD's newer GCN 1.1 chip (290X). Since Maxwell will bring about a 35% increase in IPC, unless NV makes a die much smaller than 550mm2 and/or AMD produces a similar increase in IPC with GCN2.0 and/or increases die size beyond 440mm2, NV will only extend the lead next generation.

It's like taking Kepler, adding 35% on top and making a 530mm2 die. How is AMD going to combat this exactly with a 450mm2 GCN 1.1 chip? I don't see it.

Latest drivers for both sides in a lot of games (but without inclusion of BF4 MP with Mantle!)
http://www.computerbase.de/2014-04/...-test/5/#diagramm-rating-2560-1600-4xaa-16xaf

The R290X is matching the 780ti, surprisingly. If they had included their BF4 MP data with Mantle, the R290X would actually come out on top.

GTX (Max) = power slider maxed, 100% fan, higher boost resulting in a massive power consumption increase (nearly the same as R290X).
http://www.computerbase.de/2014-04/amd-radeon-r9-295x2-benchmark-test/8/

Whereas AMD Max doesn't lift the top boost clocks beyond the default, it just prevents throttling.

Win 8.1 with i7 @ 4.8Ghz

Similar numbers to Computerbase.de's MP test at 4K for their CF R290X and 295x2.

Take it for what you will, but with the latest drivers, the R290X has caught up and to me, because BF4 is the main FPS game (unlike titles where NV leads such as AC), its actually the winner where it matters. So you could see it like so: Hawaii is a smaller chip thats matching a bigger GK110 chip. NV has no where to go in terms of bigger die size, AMD still have significant room left if they ever want a bigger chip.

 

[PPB]

The way I look at it is NV's Kepler architecture's best chip (780Ti) beats AMD's newer GCN 1.1 chip (290X). Since Maxwell will bring about a 35% increase in IPC, unless NV makes a die much smaller than 550mm2 and/or AMD produces a similar increase in IPC with GCN2.0 and/or increases die size beyond 440mm2, NV will only extend the lead next generation.

It's like taking Kepler, adding 35% on top and making a 530mm2 die. How is AMD going to combat this exactly with a 450mm2 GCN 1.1 chip? I don't see it.

GK110 is to GK10X what GCN 1.1 is to GCN 1.0. On both sides, there wasn't a big die GK100 or a Hawai-sized GCN 1.0 chip to compare it to, only AMD produced more than 1 different die for their refresh uarch of GCN.

So your comparison is flawed from the very beggining. The problem with people forgetting GK110 isnt the same as the rest of the GK10x lineup is that Nvidia this time couldn't even design a GK100 at a less-than-horrid yield rate, that's why they rushed the follow up instead.

If GCN 2.0 or whatever this will be named is to GCN 1.1 as Maxwell is to GK110, the complete beatdown in perf/mm will happen all over again.

 

[Enigmoid]

AMD already has much better shader density than nvidia, and it remains to be seen what maxwell will look like on an immature 20nm process. Also, you do realize that AMD can decide to make a 550mm die whenever they want, they would just have to cut margins (just like nvidia does).

The only reason Kepler is "faster" (it's actually something like 20% slower than hawaii at 4k with mantle) is it's a HUGE die.


AMD has consistently shown they can do more with less than nvidia. And these specs say that nvidia has already lost the performance crown. 3200sp, even with the IPC improvements, isn't going to come close to competing with nearly 2x the ROPs, 2x the bandwidth, and 1000 extra shaders.

Shader density is irrelevant without looking at performance per shader.

Hawaii really looks like its pushed to the max. Its also a bad example (GK110 vs Hawaii) when GK104 vs. Tahiti or GK 106 vs. Pitcairn show pretty much equality. 750 TI vs. Bonaire looks decidedly different.

The 780 TI has no trouble with its ROP deficiency vs. the R290X. 750TI is insanely powerful for the amount of BW it has. Shader for shader is difficult to compare when you don't know how powerful the shaders are.

Also assuming that Nvidia isn't doing what they did this time which is release the mid end first.

 

[Gloomy]

780 isn't that slow. Should be matching the 290, maybe 3-4% faster at stock.

...right?

 

[Saylick]

The way I look at it is NV's Kepler architecture's best chip (780Ti) beats AMD's newer GCN 1.1 chip (290X). Since Maxwell will bring about a 35% increase in IPC, unless NV makes a die much smaller than 550mm2 and/or AMD produces a similar increase in IPC with GCN2.0 and/or increases die size beyond 440mm2, NV will only extend the lead next generation.

It's like taking Kepler, adding 35% on top and making a 530mm2 die. How is AMD going to combat this exactly with a 450mm2 GCN 1.1 chip? I don't see it.

This. The 35% improvement in perf/core allows nVidia to hit the same performance target with less cores, and with the spare area they are able to cram more shaders onto the chip. It's not simply the 35% improvement in IPC that going to let nVidia get away with releasing GM204 as GTX 880; it's the double whammy of improved IPC + area efficiency that let's them have their cake and eat it as well. If I'm not mistaken, if the performance of 1 SMM is roughly 90% of the performance of 1 SMX while also taking up a less area, then you're looking at better than 35% scaling in terms of overall performance with a given die size.

 

[VulgarDisplay]

780 isn't that slow. Should be matching the 290, maybe 3-4% faster at stock.

...right?

Its hard to say really because I'm almost positive most reviews recycle numbers for every card except the one they are currently testing. It doesn't always give a very clear picture.

 

[n0x1ous]

Do you know what the margins are? Serious question, because I doubt nvidia is LOSING money even when they sell a 780ti at 700 instead of a K6000 or whatever at 5k. I mean losing money outright, not losing profits.


As long as AMD makes even a small profit, it's worth it to go big. They're taking back a lot of marketshare, and with all the 290s hitting ebay now for as low as $310 each you'd have to be insane to buy a 780 or worse 780ti over 2x 290 for the same price.

If its worth if for even a small profit then why haven't they ever done it?

 

[raghu78]

I don't see how it's possible for AMD to be more than a year ahead of NV with such cutting edge technology.

AMD has always been the driver of memory standards / technologies. Remember how the HD 4870 was first to GDDR5. Nvidia got there only 2 years later with GTX 480. AMD's lead in memory standards is well known and accepted.

The way I look at it is NV's Kepler architecture's best chip (780Ti) beats AMD's newer GCN 1.1 chip (290X). Since Maxwell will bring about a 35% increase in IPC, unless NV makes a die much smaller than 550mm2 and/or AMD produces a similar increase in IPC with GCN2.0 and/or increases die size beyond 440mm2, NV will only extend the lead next generation.

It's like taking Kepler, adding 35% on top and making a 530mm2 die. How is AMD going to combat this exactly with a 450mm2 GCN 1.1 chip? I don't see it.

Sorry but at the high end at 4k and multi monitor, the R9 290X CF actually is faster than GTX 780 Ti SLI, due to better scaling (XDMA gets credit for that)

also you don't know what AMD has on the cards with GCN 2.0. Nvidia has revealed Maxwell. So wait till year end to find out how actual products fare. Remember you were one of the most vocal people saying that AMD cannot beat GTX 780 / Titan with Hawaii. Look what happened.

AMD actually has the fastest solution for enthusiasts at the high end. Those 64 ROPs , massive 4GB VRAM and XDMA tech all contribute to excellent perf at the high end. Look at games like Crysis 3, BF4 to get an idea of what I am saying

most tested BF4 singleplayer which is useless. hardocp tested BF4 multiplayer with Mantle.

http://www.hardocp.com/article/2014/04/08/amd_radeon_r9_295x2_video_card_review/3

R9 295X2 crushes GTX 780 Ti SLI in min fps due to those 64 ROPs .

http://www.sweclockers.com/recension/18544-amd-radeon-r9-295x2/16#pagehead

http://www.sweclockers.com/recension/18544-amd-radeon-r9-295x2/15#pagehead

 

[x3sphere]

Its hard to say really because I'm almost positive most reviews recycle numbers for every card except the one they are currently testing. It doesn't always give a very clear picture.

The latest beta NV drivers gave a big boost in BF4 I believe, dunno if HardOCP tested with them, probably not.

 

[raghu78]

The latest beta NV drivers gave a big boost in BF4 I believe, dunno if HardOCP tested with them, probably not.

they used the latest 337.50 drivers.

http://www.hardocp.com/article/2014/04/08/amd_radeon_r9_295x2_video_card_review/2#.U0d751WSyE4

Video card configuration

GTX 780 Ti SLI - 876 / 928 Mhz / 7 Ghz (1019 Mhz real clock) - 337.50 beta

 

[3DVagabond]

The latest beta NV drivers gave a big boost in BF4 I believe, dunno if HardOCP tested with them, probably not.

They did.

Besides playing games at "max playable settings" the other difference with [H] reviews are the cards are well warmed up.

 

[x3sphere]

That's impressive then. I think once I pick up a 4K monitor I'll put the R9 290Xs that I'm using for mining into my gaming rig and sell off the 780s I have now, unless there's something better out by then. AMD definitely seems to be the way to go for high-res gaming.

 

[VulgarDisplay]

That's impressive then. I think once I pick up a 4K monitor I'll put the R9 290Xs that I'm using for mining into my gaming rig and sell off the 780s I have now, unless there's something better out by then. AMD definitely seems to be the way to go for high-res gaming.

at least when you're considering multicard setups. Nvidia is still faster overall for single cards.