videocardzAMD Radeon R9 490X and R9 490 launches in June/Pro Duo launches on April

[Silverforce11]

If Polaris uses the improved GCN in the patent, base clock will be meaningless because it's ALUs can turbo boost individually, based on the workload. Think of it like a finer-grained version of NV's turbo boost tech, since it's dynamic on a local SP/SIMD level rather then entire chip.

 

[flopper]

If Polaris uses the improved GCN in the patent, base clock will be meaningless because it's ALUs can turbo boost individually, based on the workload. Think of it like a finer-grained version of NV's turbo boost tech, since it's dynamic on a local SP/SIMD level rather then entire chip.

if and its a big if AMD pulled such a thing off then its a totally new enviroment for gpus

 

[flopper]

AMD want to keep power consumption down. Therefore the base clocks will likely be quite low. Various aspects of the GPU will boost based on load instead (like the ALUs).

A 1,150MHz base clock says nothing about the overclocking headroom. The GPU could overclock up to 1,600MHz for all we know.

NVIDIA need to be aggressive with their base clocks in order to compete under compute loads.

cant satisfy the wet dog Pascal crowd.
Polaris is awesome.

 

[gamervivek]

Just to say but:

Base clock of 980 is 1126MHz
Base clock of 980ti is 1000MHz

so having a base clock of 1150MHz is not exactly low given it is higher than any of the current cards.

Since we don't know the boost clock (assuming this information is remotely correct) it doesn't tell much where it will fall in terms of clockspeeds.

AMD doesn't have nvidia like boost clock. Their 1050Mhz clock on Fury X is 'upto' and not 'at least' like nvidia's.

980's official boost clock is 1216Mhz and it can boost further to 1265Mhz while 980Ti's official boost clock is 1075Mhz and it can boost to 1200Mhz easily.

Overclocking or custom cards can do 200-300Mhz over that. And AMD aren't that much better at engineering than nvidia to overcome that massive clockspeed deficit.

 

[el etro]

I would vote for clockspeeds 10%-15% above Hawaii, Finfet process being the bigger responsible. Shader core is stronger than was in older GCNs.

 

[Vaporizer]

Baffin will use powergating of CU like described in the patent.

 

[MrTeal]

Baffin will use powergating of CU like described in the patent.

Source?

 

[Vaporizer]

Www.planet3dnow.de/cms/23084-polaris-mit-bis-zu-1154-mhz-und-neuer-stromsparfunktion
German site
Was found in some linux graphicsdriver mailing list

 

[Kenmitch]

Overclocking or custom cards can do 200-300Mhz over that. And AMD aren't that much better at engineering than nvidia to overcome that massive clockspeed deficit.

Although clock speed plays a part, it's the actual architecture that dictates the performance in the end. You'll always get the point of diminishing returns, crazy power draw, etc. In the end it really depends on how much performance is baked in the design in the 1st place. Looking at the benchmarks from recent released games it pretty much proves the point.

 

[redvapor]

Www.planet3dnow.de/cms/23084-polaris-mit-bis-zu-1154-mhz-und-neuer-stromsparfunktion
German site
Was found in some linux graphicsdriver mailing list

Google translate:
Our resourceful users Bomby has found some interesting information about AMD's upcoming Polaris graphics cards. In the mailing list to AMD's Linux graphics driver an employee has submitted a patch, the first support for the Polaris family to offer. In it you will find some interesting information.

Firstly, it is confirmed that the next AMD graphics cards can display 16 bits per color channel. Thus, the Polaris-graphics cards may also be suitable for demanding applications. Furthermore you can find information to RAM. In Ellesmere 8 channels and 8 banks of 32 bits can be specified, Baffin is designed to provide only 4 channels, but can be fitted with 8 banks. From the specifications can be seen that AMD will initially continue to rely on GDDR5 memory. With four channels logically produces a 256-bit memory interface, eight channels of 512 bits. In both cases, a maximum memory configuration of 8 GiB GDDR5. As maximum clock a very high value of 6000 MHz is specified in each case. This corresponds to the effective clock speed of the GDDR5 memory and is already in models like AMD's Radeon R9 380 or GeForce GTX see 960th To what extent the currently specified maximum clock speed of 1154 MHz, which is valid for both chips, is also so incorporated into the series remains to be seen. At first engineering samples (prototypes) most other clock rates than the series production apply.
Furthermore Baffin is probably up to 5 displays and Ellesmere those able to control. 6 In another post Alex Deucher writes something to a new power saving mechanism which first times shall represent only Baffin available. It looks like AMD has in the new GCN generation Powergating for individual CUs (Compute Units), which bundle multiple shaders. This feature is likely to find some friends especially in the mobile segment. By adjusting the active computing cluster, the energy consumption control well and extend the battery life.

 

[Tuna-Fish]

His math on memory channels is incorrect, 8 x 32 = 256, 4x32 = 128.

That "leak" would strongly imply that the recent patent on CU organization only referred to Vega.

 

[Head1985]

Polaris 10 and 11 are 480/470
http://videocardz.com/59065/amd-radeon-r9-470-and-radeon-r9-480-to-be-released-at-computex

 

[AtenRa]

Www.planet3dnow.de/cms/23084-polaris-mit-bis-zu-1154-mhz-und-neuer-stromsparfunktion
German site
Was found in some linux graphicsdriver mailing list

nice thanks,

Individual ALU power gating was too much.

 

[Qwertilot]

Just to come back to something from further up thread - I guess that given AMD's recent history there has to be a decent chance they'll try and push out some madly (effectively over) clocked version of this to be the top tier.

I rather hope they don't actually as it'd obviously wreck the power efficiency.

 

[coercitiv]

I rather hope they don't actually as it'd obviously wreck the power efficiency.

Yeah, and most enthusiasts underclock and undervolt their rigs for that insane power efficiency.

 

[tential]

Just to come back to something from further up thread - I guess that given AMD's recent history there has to be a decent chance they'll try and push out some madly (effectively over) clocked version of this to be the top tier.

I rather hope they don't actually as it'd obviously wreck the power efficiency.

You want to cite some examples of this in recent amd history?

 

[Qwertilot]

Not hard FuryX most obviously and potentially quite anologously, but aren't the 390/X also clocked rather past the point where they'd be most efficient?

Can't stop the OEM's going crazy of course, but can at least ship the references at sane clock speeds so you can see where that should be.

Suspect they will revert to doing that this time, what with the efficiency being the principal benefit of these cards vs their last generation stuff.

Still, if they are say a chunk behind P104 in raw performance then they'll maybe be tempted to try and push it for at least SKU to have something close.

 

[tential]

Not hard FuryX most obviously and potentially quite anologously, but aren't the 390/X also clocked rather past the point where they'd be most efficient?

Can't stop the OEM's going crazy of course, but can at least ship the references at sane clock speeds so you can see where that should be.

Suspect they will revert to doing that this time, what with the efficiency being the principal benefit of these cards vs their last generation stuff.

Still, if they are say a chunk behind P104 in raw performance then they'll maybe be tempted to try and push it for at least SKU to have something close.

So what are efficient clocks for the Fury X and the 390x since these are by your definition "madly overclocked".

 

[USER8000]

So what are efficient clocks for the Fury X and the 390x since these are by your definition "madly overclocked".

They do seem past the most efficient clockspeed range though when compared to commericial cards and stuff like the Fury Nano.

 

[tential]

They do seem past the most efficient clockspeed range though when compared to commericial cards and stuff like the Fury Nano.

So again, I ask, what is the most efficient clockspeed for the cards?

And when compared to what commercial cards?

The R9 nano is an HTPC product that fits into ITX cases... so you're saying we should compare these products meant for desktop class products for the HTPC Market.....

K..... I don't see any reason why that should be done, but hey if that's what you want to do, I guess it's for others to judge whether that seems to be a remotely reasonable thing to do.

Also, I'm curious as to how underclocking the Fury X would have benefited AMD. If Qwertilot would explain that part of his post that would be lovely.

Summary:

How would releasing Fury X in a not "madly overclocked" state benefited AMD?
What are the not "madly overclocked" clockspeeds of the Fury X?

 

[Silverforce11]

https://benchlife.info/computex-2016-pre-show-with-amd-arm-intel-nvidia-04142016/

Polaris 10 will be 480/X: ~110-135W
Polaris 11 will be 470/X: ~50W

Called it earlier for these TDP.

If the 2.5x perf/w claims are accurate... time to do some basic maths.

 

[Adored]

Well I called the names at 470 and 480. Now if only Rage will be true for Vega...

135W should be over Fury X at 2.5x that but my guess is Tonga is the benchmark, somebody else can figure that one out. Surprised to see P11 doesn't have a faster version though and tbh I was expecting a dual 6-pin version of 10 as well. Looks like Nvidia will be a bit ahead in raw performance.

 

[Silverforce11]

Well I called the names at 470 and 480. Now if only Rage will be true for Vega...

135W should be over Fury X at 2.5x that but my guess is Tonga is the benchmark, somebody else can figure that one out. Surprised to see P11 doesn't have a faster version though.

When AMD launched Fury X, they claim perf/w improvements based on Hawaii as it was the top dog.

Remember the Nano 2x perf/w claims? It was ~there versus Hawaii.

I think they would put their improvement claims vs their most current, and that's Fiji, potentially Fury X as it's the top dog.

Polaris 10 should NOT be able to compete with GP104 unless NV messed up or Polaris is just very good. It's not a real mid-range chip, that's Vega 11. Think about it, tiny size, GDDR5, it should not reach GP104 with GDDR5X.

 

[Adored]

It's possibly Fiji but I don't like that due to HBM vs GDDR5. That said I don't like the Hawaii comparison either for the same reason and also the lack of compression.

Regardless of Tonga or Fiji though, it should be able to match or beat Fury X at 2.5x per/Watt and 135W, using the 380X or Fury X as the benchmark.

I would guess at 50W those low looking Polaris 11 clocks are real as well. Gotta feel there is a trick being missed there, surely that chip can be pushed to 90W or so.

 

[tential]

By your logic the 980ti shouldn't have been faster than hbm fury x. Memory isn't everything

I'll wait and see how the architectures actually play out in games rather than make judgements so hastily.