AMD: Speculation, Rumors, and Discussion (Archive)

Status
Not open for further replies.
I still ask me if it is a good idea to come with a ( 199$ ) gpu's without having launch the "big gun" before. Strangely, peoples who tend ot buy thoses cards, seems offtly just take reference of the high end one.
84% of reasons. if the number AMD gave were correct.
 
Things like compute software? yeah its their on their site. That is what they are talking about with the 2.8x

http://www.amd.com/en-us/innovations/software-technologies

here are you AMD technologies

its just like nV stating 10x more performance, they are talking about specific software.

That's not exactly how I read it.
nj0ek4xi.2of.jpg



70% uplift from process. A further 110% from the somewhat vague 'Arch+Features'. So 280% may rarely be reached, but its certainly more than the up to 170% from process alone.
 
This WILL be my first dGPU since a long time. My last dGPU is GeForce 400MX.....
Hopefully the 8GB variant is closer to 200 than 300 ($250 would be fine).
 
Also. if the duel CPU were only at 51% utilization. and with the scaling not been 100% that means that a single P10 in that game is equal to a 1080?
 
480 may not be the fully enabled chip. 36CU seems low to me.
I wonder if there will be a Rx 480x, name would sound terrible. Maybe Rx 485.
 
That's not exactly how I read it.
nj0ek4xi.2of.jpg



70% uplift from process. A further 110% from the somewhat vague 'Arch+Features'. So 280% may rarely be reached, but its certainly more than the up to 170% from process alone.

Effectively, what i read there is " 1.7 from " process change " ( in yellow labeled, technology ), Arch + features change ( arch improvement, + features ).

Anyway.
The AMD Polaris GPUs are based on the latest GCN 4.0 architecture which makes use of the FinFET architecture to deliver low power consumption, lesser voltage leak and better performance. AMD deploys a completely new architecture for Polaris chips with enhanced features listed below:

  • Primitive Discard Accelerator ( conservative rasterization. )
  • Hardware Scheduler
  • Instruction Pre-Fetch
  • Improved Shader Efficiency
  • Memory Compression
 
Also. if the duel CPU were only at 51% utilization. and with the scaling not been 100% that means that a single P10 in that game is equal to a 1080?

Nah. Just means they and the devs have more work to do with multi-GPU performance improvements.

I would guess they are seeing maybe 60-70% CF scaling, which would put a single Rx480 ~15-25% less than GTX1070.
 
That's not exactly how I read it.
nj0ek4xi.2of.jpg



70% uplift from process. A further 110% from the somewhat vague 'Arch+Features'. So 280% may rarely be reached, but its certainly more than the up to 170% from process alone.


They have 100% increase from base line there, where does that fit in with your 70% uplift from process?

The whole thing on the right doesn't' fit into the 1.7x at all. Are they constituting technology and baseline as finfet? Architecture and features on top of that? doesn't make much sense.

Anycase AMD technologies has a super scripted 1 next to it, its going to be specific applications.
 
2 polaris chips at 51% utilization beat an 1080 at 98% utilization, according to them.

Any chance polaris can gain similar to 1080 levels of utilization in future dx12 titles? If so a single polaris 480 would be matching a 1080, at least in some dx 12 titles.
@ what resolution or settings?
That 51% utilization is probably the limit of the Ashe's engine to distribute load on multiple GPUs, I bet 2 GTX 1080s would post roughly the same utilization rate.

gtx-1080-sli-aots.jpg

http://www.hardwarezone.com.sg/feature-preview-nvidia-geforce-gtx-1080-sli-benchmarked
 
2 polaris chips at 51% utilization beat an 1080 at 98% utilization, according to them.

Any chance polaris can gain similar to 1080 levels of utilization in future dx12 titles? If so a single polaris 480 would be matching a 1080, at least in some dx 12 titles.
We've already seen DX11 games where AMD managed 100% scaling, so it seems reasonable some games would do better. May be more an issue of AFR (high latency/FPS) vs SFR (low latency/moderate FPS gains). VR of course would be easy, but there are likely low utilization applications where the gains are worthwhile considering the price.

While maybe not 100% utilization, the big difference maker could be a game running a physics engine, AI, etc asynchronously. Not just async compute, but accelerating the engine at a different framerate than the rendering. Run collision every 3rd frame sort of arrangements or pathfinding in the background. They could accelerate far more complex systems that are currently used, but utilization would look strange. That sort of use would not work well with Nvidia cards asynchronously.

Does anyone think they're going to announce a bigger Polaris card in the near future? I'm wondering how long I should hold off.
Seems likely considering partners are considering water cooling options beyond the dual GPU configurations. A water cooled GPU could push clocks significantly as it would help with the thermal density issues. Consider a Fury running 50C as opposed to air cooled cards ~70C. The current 480 card they've shown lines up much better with a cut down Polaris 10 version, so there is room for an improved version.

480 may not be the fully enabled chip. 36CU seems low to me.
I wonder if there will be a Rx 480x, name would sound terrible. Maybe Rx 485.
I've wondered the same thing. Normally the xx5 series was already the off-year refresh. The other option may be the full chip being 490 series with Vega being a Fury/Rage tier with a big chip and maybe professional for the 2nd variant. A small Vega MCM, considering HBM2 and an interposer already, could be a possibility.
 
They have 100% increase from base line there, where does that fit in with your 70% uplift from process?

The whole thing on the right doesnt' fit into the 1.7x at all. Are they constituting techonlogy and baseline as finfet? Architecture and features on top of that? doesn't make much sense.
100% baseline + 70% uplift from 'technology' = 1.7x performance/w with finFET
 
They have 100% increase from base line there, where does that fit in with your 70% uplift from process?

The whole thing on the right doesn't' fit into the 1.7x at all. Are they constituting technology and baseline as finfet? Architecture and features on top of that? doesn't make much sense.

Anycase AMD technologies has a super scripted 1 next to it, its going to be specific applications.

So you don't understand percentages and are somehow making claims that they are lying?
 
So you don't understand percentages and are somehow making claims that they are lying?


its not a lie (its confusion the way they broke it down), its specific applications hence the super script. What did nV lie about 10x the performance? No, its specific applications.
 
CEO just mentioned in her summary 'RX GPUs for $100-$300 price points' if I heard correctly.
Edit: Rewatched from -2:36 minutes and that is definitely what she said.
 
OMG... they are stating upto 2.8x perf/w increase due to 14FinFet, Polaris architecture and features.


yes and it will in specific applications, that is what is noted on the slide, AMDtechnologies which I just linked you to what those were.

What happened to the 60% decrease in power usage with FinFets, that is the same as 2.0 perf/watt increase. Its no longer that.
 
Well, I'm and idiot. Then again so is AMD.

This was supposed to be the year they made a turnaround. The stock market has been betting on it. How the fuck is this a turnaround? Dual GPU for a low end card, not a mid range, but a low end thing? It's mind blowingly stupid for any company, especially one that was supposed to be on the right track.
 
Status
Not open for further replies.
Back
Top