[Benchlife] R9 480 (Polaris 10 >100w), R9

[Headfoot]

Might explain AMD adding the feature to HDMI?

The timing of all of it seems to make a lot of sense for this purpose. FreeSync over HDMI for TV <-> Console would be a great feature

 

[Glo.]

And why not? If PS4 Neo includes a Polaris 10 class GPU which was shown to be comparable to a Fury X then it can definitely do some games in 4k. The main point here is that comparison is not optimized. We are discussing Polaris 10 at Fury X speeds in a desktop windows environment. With much closer to metal programming and impending DX 12 performance tweaks this new chipset could absolutely run games @ 4k with medium-ish settings.



Yes smart move to just keep the cpu but tweaked up 30% clockspeed. With console optimizations this should continue to be sufficient in most applications.

Do not hype yourself too high. Fiji may still be on sale after Polaris release which would indicate that none of small AMD GPUs will be faster/more powerful than Fiji.

 

[coercitiv]

A PS4 pulls 130-140W at the wall while gaming, so doubling the CU count at 28nm while staying at the same power consumption seems unlikely.

Double the core count, increase clocks for CPU, GPU, RAM while TDP stays the same. Moore's law has clearly taken refuge in the lost art of 28nm kung fu.

 

[AtenRa]

Double the core count, increase clocks for CPU, GPU, RAM while TDP stays the same. Moore's law has clearly taken refuge in the lost art of 28nm kung fu.

Fiji Nano even with GDDR-5 would have higher perf/watt than Hawaii at 4K.

 

[swilli89]

Your expectations of Polaris seems to have reached new heights.

The PS4 today already struggles at 1080p. Adding twice the power wont do anything if you are to render a resolution 4 times greater.

Shintai regardless of my personal projections; you have to atleast agree there will be a performance increase in per CU or per SP. We are already doubling SP count with this proposed chipset so it will definitely be more than double performance.

Doesn't make a lot of sense to backport Polaris to 28nm when its 14nm native, and the economics of the consoles already dictate going to 14nm for #of die per wafer and power consumption reasons. I doubt it will be a 28nm, considering how touchy they are about thermal budgets after the PS360 generation fiascos

This^. No way would they spend money to backport to an old process on such a die size and energy efficiency focused application. 14nm capacity is very high with both GF and Samsung able to produce designs on the exact same process with no porting work needing done.

 

[swilli89]

Fiji Nano even with GDDR-5 would have higher perf/watt than Hawaii at 4K.

The problem is Sony would never be able to afford such a huge and complex die size, even on 28nm. It makes way more sense to utilize a Polaris 10 on 14nm and enjoy the small die size/power consumption/DX 12 opportunities.

 

[ShintaiDK]

Shintai regardless of my personal projections; you have to atleast agree there will be a performance increase in per CU or per SP. We are already doubling SP count with this proposed chipset so it will definitely be more than double performance.

But you are not doubling bandwidth for example.

 

[railven]

Yes exactly the ps4 has far more gpu power yet devs still manage it. I forgot to mention that thanks.

But they don't. Either it's MSFT money hatting or devs not bothering, the performance different between Xbone and PS4 is substantial (almost a whole GPU tier on the PC side) yet AAA console games often perform the same on both consoles.

In situations where the PS4 is given a little leeway, it often bites them in the ass due to frame pacing issues.

Just look at this recent example:
https://www.youtube.com/watch?v=YECC8u0nPO8

In that example the superior PS4 hardware is deemed inferior to the Xbone performance due to frame pacing issues.

And this example:
https://www.youtube.com/watch?v=UPoGhr-3DS8

The Xbone version is deemed "identical" to the PS4 version.

Now imagine a higher performance console thrown into the mix. Will it get hamstrung by the weaker consoles, or will it be used as the base with the weaker consoles getting shafted?

And now rumors of Nintendo NX being faster/stronger than PS4, it will be interesting to see if MSFT hurries something out to compete.

Console buyers aren't regulars in the hardware upgrade. People who keep using the mobile phones example aren't factoring phone upgrades are often baked into service payments. If Sony or MSFT wants to follow that structure, it would make more sense. Pay for Live/PSN+ and get the console "free."

 

[AtenRa]

The problem is Sony would never be able to afford such a huge and complex die size, even on 28nm. It makes way more sense to utilize a Polaris 10 on 14nm and enjoy the small die size/power consumption/DX 12 opportunities.

Im not implying they will use a 600mm2 die, i was just pointing out that Fiji with a bigger die at the same process has higher perf/watt than smaller die like Hawaii at 4K.

 

[coercitiv]

Fiji Nano even with GDDR-5 would have higher perf/watt than Hawaii at 4K.

How is this relevant? Fiji @ 28nm achieves a TDP of 175W while the entire PS4 uses 130-150W while gaming.

How do you expect to get more than 2x perf/watt improvement on 28nm (double core count, increased clocks, better cores) when it takes 14nm to get 2.5x perf/watt with Polaris vs. Hawaii?

 

[Erenhardt]

But they don't. Either it's MSFT money hatting or devs not bothering, the performance different between Xbone and PS4 is substantial (almost a whole GPU tier on the PC side) yet AAA console games often perform the same on both consoles.

In situations where the PS4 is given a little leeway, it often bites them in the ass due to frame pacing issues.

Just look at this recent example:
https://www.youtube.com/watch?v=YECC8u0nPO8

In that example the superior PS4 hardware is deemed inferior to the Xbone performance due to frame pacing issues.

And this example:
https://www.youtube.com/watch?v=UPoGhr-3DS8

The Xbone version is deemed "identical" to the PS4 version.

Now imagine a higher performance console thrown into the mix. Will it get hamstrung by the weaker consoles, or will it be used as the base with the weaker consoles getting shafted?

And now rumors of Nintendo NX being faster/stronger than PS4, it will be interesting to see if MSFT hurries something out to compete.

Console buyers aren't regulars in the hardware upgrade. People who keep using the mobile phones example aren't factoring phone upgrades are often baked into service payments. If Sony or MSFT wants to follow that structure, it would make more sense. Pay for Live/PSN+ and get the console "free."

There are plenty of examples where the difference in visuals between ps4 and xbone is jarring, where ps4 runs games much faster.

Sony does what devs want.

Ultimately, market decides. And market reaction will be interesting to watch.

Also, OCed jaguar makes sense. Some devs were saying something about fine tuning games for consoles, aiming for specifics such as CPU cache latency and capacity, claim a change in those would result in a degradation in performance due to broken optimization, regardless of generally faster CPU architecture. Keeping the core and clocking it higher ensures optimizations to work and improves performance in a linear fashion.

 

[swilli89]

But you are not doubling bandwidth for example.

This is the main question I have on my mind. So many memory technologies in play, I'm really curious about what would happen here.

Historically consoles make concessions when it comes to hardware performance as we all know. We are probably looking at the same 256-bit interface clocked up a certain amount with whatever memory management improvements AMD has incorporated into Polaris. In the end, it might end up being enough for 4k or it might not, too difficult to tell what will happen on the memory side at this point.

 

[railven]

There are plenty of examples where the difference in visuals between ps4 and xbone is jarring, where ps4 runs games much faster.

That's the issue. It should be ALL, not some.

EDIT: And when it was PS3 vs 360, the "journalist" and "reviewers" had no issue pointing out when the 360 version of a game was superior. When PS4 and Xbone launched these same talking heads spun it that resolution/IQ didn't matter "just the gameplay."

Sony does what devs want.

Ultimately, market decides. And market reaction will be interesting to watch.

Also, OCed jaguar makes sense. Some devs were saying something about fine tuning games for consoles, aiming for specifics such as CPU cache latency and capacity, claim a change in those would result in a degradation in performance due to broken optimization, regardless of generally faster CPU architecture. Keeping the core and clocking it higher ensures optimizations to work and improves performance in a linear fashion.

End of the day I feel Sony is risking the regained popularity they gained. As someone said here, they need to PR it just right. MSFT pretty much blew away all their cred with their always online shtick (witching reading forums like NeoGAF, seems them fools already expect it yet threw a fit when MSFT blatantly said this was the future).

Where MSFT fumbled was execution. Sony can easily piss off the consolers (if you think the flamer wars here are bad, woof, swing by any NPD thread over at Gaf) and suddenly find themselves back in the PS3 situation. Even more so if they expect to charge more per game (which I wouldn't put it pass them).

 

[swilli89]

That's the issue. It should be ALL, not some.

EDIT: And when it was PS3 vs 360, the "journalist" and "reviewers" had no issue pointing out when the 360 version of a game was superior. When PS4 and Xbone launched these same talking heads spun it that resolution/IQ didn't matter "just the gameplay."



End of the day I feel Sony is risking the regained popularity they gained. As someone said here, they need to PR it just right. MSFT pretty much blew away all their cred with their always online shtick (witching reading forums like NeoGAF, seems them fools already expect it yet threw a fit when MSFT blatantly said this was the future).

Where MSFT fumbled was execution. Sony can easily piss off the consolers (if you think the flamer wars here are bad, woof, swing by any NPD thread over at Gaf) and suddenly find themselves back in the PS3 situation. Even more so if they expect to charge more per game (which I wouldn't put it pass them).

The Xbox One die is 363 square millimeters, up from the PS4's 348 sq mm. The 5% additional space, despite having the smaller GPU core, is mostly due to RAM. The Xbox One contains a whopping 47MB of on-die RAM, and that pushes the die size up considerably.

This makes a huge difference...

Also remember this?: http://wccftech.com/amd-developing-20nm-ps4-xboxone-apu/

 

[AtenRa]

How is this relevant? Fiji @ 28nm achieves a TDP of 175W while the entire PS4 uses 130-150W while gaming.

How do you expect to get more than 2x perf/watt improvement on 28nm (double core count, increased clocks, better cores) when it takes 14nm to get 2.5x perf/watt with Polaris vs. Hawaii?

Fiji is ~40% bigger die than Hawaii.
Fiji has ~45% more Shaders
Fiji Nano at 175W TDP is 10-20% (or more faster with DX-12) at 4K than Hawaii at 250W TDP. That makes it have almost 2x the perf/watt of Hawaii at the same 28nm.

So essentially we could have a bigger die with 50% or more Shaders over PS4 and at lower TDP and still be faster at 4K and still manufactured at 28nm.

edit: Or we could have double the die, double the shaders and double the performance at 4K at the same TDP as current PS4 using the same 28nm process.
The difference here is we are using 4K perf/watt vs 1080p perf/watt

 

[coercitiv]

Fiji is ~40% bigger die than Hawaii.
Fiji has ~45% more Shaders
Fiji Nano at 175W TDP is 10-20% (or more faster with DX-12) at 4K than Hawaii at 250W TDP. That makes it have almost 2x the perf/watt of Hawaii at the same 28nm.

So essentially we could have a bigger die with 50% or more Shaders over PS4 and at lower TDP and still be faster at 4K and still manufactured at 28nm.

Except Fiji obtains better perf/watt by increasing core count and decreasing clocks (HBM is just the cherry on top). Meanwhile, if the specs are real, the new PS4 increases core count and also increases clocks as well.

Again, unless the new PS4 sees a significant jump in power usage, you are expecting more than 2x perf/watt jump by staying on the same process while AMD stated their 14nm products offer a 2.5x perf/watt jump relative to comparable 28nm products. Do you really expect 14nm to bring less than 1.25x perf/watt improvement relative to this iteration of 28nm?

 

[tential]

There are plenty of examples where the difference in visuals between ps4 and xbone is jarring, where ps4 runs games much faster.

Sony does what devs want.

Ultimately, market decides. And market reaction will be interesting to watch.

Also, OCed jaguar makes sense. Some devs were saying something about fine tuning games for consoles, aiming for specifics such as CPU cache latency and capacity, claim a change in those would result in a degradation in performance due to broken optimization, regardless of generally faster CPU architecture. Keeping the core and clocking it higher ensures optimizations to work and improves performance in a linear fashion.

It goes both ways erenhardt. Sometimes ps4 does better due to more gpu power other times the extra clockspeed of the Xbox one helps.

 

[JDG1980]

But you are not doubling bandwidth for example.

Assuming they don't use GDDR5X, no. But even with standard GDDR5, there's decent room for improvement. The current PS4 APU runs with a memory clock of 1375 MHz, which provides 176 GB/sec of bandwidth. The leaks indicate that bandwidth will be going up to 218 GB/sec, which would correspond to a memory clock of 1700 MHz - well within reason, since Nvidia has had Maxwell cards running at that rate for 2 years already.

And don't forget that the existing PS4 is GCN 1.1, which has no memory compression at all. AMD has had some form of memory compression tech since Tonga, and with Polaris it's likely that it will be at least as efficient as Maxwell's. The GTX 980 shows how far it's possible to go with just a 256-bit bus. And Anandtech's overclocking results indicate that GM204 isn't bottlenecked; pushing the core clock higher gives substantial benefits even without overclocking the RAM.

Assuming that the PS4 NEO APU is a Polaris-based product on 14LPP, and that existing leaks/rumors about number of shaders, bus width, and memory clocks are roughly accurate, I think we should expect the new console's GPU power to be about equivalent to a R9 290X. That's pretty impressive, considering that this is still a solid mid-range discrete card today. That should be good news for PC gamers, since there will no longer be the need to cripple everything for the sake of the consoles. I expect that PS4 NEO users will be given a choice between 1080p@60Hz or 4K@30Hz for most games going forward, and legacy PS4 users will get 1080p@30Hz.

 

[HurleyBird]

Anandtech's overclocking results indicate that GM204 isn't bottlenecked; pushing the core clock higher gives substantial benefits even without overclocking the RAM.

By itself, comparing memory OC to core OC does not tell you if a part is well balanced or bottlenecked. You need to take into account transistor and power budgets too.

To give a somewhat simplified example, if a core OC gives twice the performance increase of the same relative memory overclock, but the memory system is only using 1/5th of the transistors and power of the rest of the chip, then yes, you're being limited by the memory subsystem more than anything else - by a fairly large margin at that. Of course, getting all the necessary information isn't exactly straightforward, but that doesn't make ignoring it correct.

 

[Silverforce11]

This is a terrible way forward for consoles. Split dev priorities, compatibility issues, frequent upgrade cycles, etc. You can argue second rate isn't obsolete, but it's still second rate, and every PS4 currently in homes and stores just became it.

Not at all. Sony's guide to devs pretty much tells the story here. All they need to do is add a Neo mode, instead of 1080/30 fps frame-rate-lock (which most games have), either go with 1080p/60 fps or 4K/30 fps. How hard do you think it is to change the resolution? Real simple, lots of games even have dynamic resolution built in to hit a frame rate target.

They won't have to write a different code-base or porting their game. The final compiled game is x86 based and the GPU talks in the same API. With Polaris having an improved command processor/hardware scheduler, devs won't need to worry about writing GPU specific code, because the dynamic scheduler will decide which SIMD/ALU will run a certain shader most optimally.

 

[JDG1980]

By itself, comparing memory OC to core OC does not tell you if a part is well balanced or bottlenecked.

It does, however, indicate that memory bandwidth on a ~224GB/sec bus is not an insuperable obstacle to greater performance than stock GM204. That's the relevant issue here.

 

[Silverforce11]

I don't think the "new" consoles will be rendering games at 4K...

I think you will be very wrong.

I have a 4K setup and only an R290X to power it.

Was just playing some recent games, ie. Far Cry Primal. By running max textures, but other settings on normal, the game was locked at a 30 fps cap. It's actual frame rate was ~35 to 40 but it was smoother with a 30 fps lock.

Does it look good? Heck yes, looks 99% like Ultra, because I can run 4K textures.

This is effectively the same for most AAA titles I've tried. Older games will run at 60 fps with max textures and medium/normal settings.

On consoles, these are the kind of settings they will utilize, and with more optimizations AND Polaris 10 based GPU being faster than R290X, 4K/30fps will be easily achieved.

 

[raghu78]

Assuming they don't use GDDR5X, no. But even with standard GDDR5, there's decent room for improvement. The current PS4 APU runs with a memory clock of 1375 MHz, which provides 176 GB/sec of bandwidth. The leaks indicate that bandwidth will be going up to 218 GB/sec, which would correspond to a memory clock of 1700 MHz - well within reason, since Nvidia has had Maxwell cards running at that rate for 2 years already.

And don't forget that the existing PS4 is GCN 1.1, which has no memory compression at all. AMD has had some form of memory compression tech since Tonga, and with Polaris it's likely that it will be at least as efficient as Maxwell's. The GTX 980 shows how far it's possible to go with just a 256-bit bus. And Anandtech's overclocking results indicate that GM204 isn't bottlenecked; pushing the core clock higher gives substantial benefits even without overclocking the RAM.

yeah AMD has already said that memory compression on Polaris has been improved. This combined with a new memory controller and new L2 cache system is going to be the key to to avoid major bottlenecks due to bandwidth. I thinks its fair to say that AMD knows better than anybody of us internet forum arm chair critics on how to design a balanced GPU. Pitcairn was an excellent GPU and Polaris 10 looks all set to be an excellent successor.

Assuming that the PS4 NEO APU is a Polaris-based product on 14LPP, and that existing leaks/rumors about number of shaders, bus width, and memory clocks are roughly accurate, I think we should expect the new console's GPU power to be about equivalent to a R9 290X. That's pretty impressive, considering that this is still a solid mid-range discrete card today. That should be good news for PC gamers, since there will no longer be the need to cripple everything for the sake of the consoles. I expect that PS4 NEO users will be given a choice between 1080p@60Hz or 4K@30Hz for most games going forward, and legacy PS4 users will get 1080p@30Hz.

Thats very conservative. AMD has already hinted at a large performance increase with 4th gen GCN. The PS4 Neo GPU will easily more than double the PS4 GPU perf . Even on the desktop I think we will see Polaris 10 based GPUs easily beat Hawaii and match or exceed Fury X. I am thinking AMD has a lot of SKUs planned with Polaris 10. I would not be surprised if the fully enabled 2560 sp Polaris 10 sports GDDR5X and the 2304 sp Polaris 10 sport GDDR5 at 256 GB/s.

 

[Silverforce11]

But you are not doubling bandwidth for example.

Yes, yes you are. You need to apply some of your logic.

PS4 is GCN 1.1, 176GB/s.

Tonga has 40% effective memory compression tech, this is from AMD's presentations.

Polaris has an improved memory compression tech. How much better? Let's say a modest gain, up to 50% effective.

It's ~226GB/s with 50% compression is equivalent to 452GB/s.

Other uarch changes also improve bandwidth efficiency, by going less back & forth to memory with better pre-fetch and cache etc.

 

[Silverforce11]

Btw, PS4K Neo confirmed specs.

http://www.eurogamer.net/articles/digitalfoundry-2016-sonys-plan-for-playstation-4k-neo-revealed

https://www.youtube.com/watch?v=8JJbWo8y58M

These guys claimed they have seen the documentation that Sony have sent to developers (along with devkits).

PS4 and Neo will co-exist, Sony has tough guidelines that devs must adhere to to publish games on their platform. Basically games need to target both, PS4 quality must be like the current situation, target 1080p/30 fps. Neo extends that, devs can pick 60 fps or 4K/30 fps.

There's now leaks that MS is working on the Xbox Next, also going with Polaris APU.

All 3 major consoles are going to be on the same tech level, x86/GCN, Sony/MS will be backwards compatibility. Nintendo jumping onboard to the x86/GCN train. Devs will love this because it makes their cross-platform development much easier.

ETA? Sony is accepting "Neo" game submission in August. They require all games released after October to support Neo. This likely points to an October release for the PS4 Neo.