Vega/Navi Rumors (Updated)

[ShintaiDK]

Sorry it was you who said they lied and now you want ME to prove they didnt ??

There is no single review that backs up AMDs claim at the chili event. But it sure lead to all kind of silly hype predictions that ended up far from reality, including from yourself

And now the Vega hype is already starting.

 

[kraatus77]

And with the same setup and what not? No, because AMD never said what they actually used with good reason

But lets face it, with the RX 460 reviews. Its not going to happen. They simply fooled people no matter how much you try and deny it.

They did state their test systems "Intel Core i7 4790K with 4x4GB DDR4-2600 MHz memory, Windows 10 64bit.
Star Wars Battlefront, X-wing Training
Med Preset @1080p. AMD Polaris GPU tested with 16.10 beta drivers . The GTX 950 tested with 359.06 drivers."

http://www.anandtech.com/show/9886/amd-reveals-polaris-gpu-architecture

so you have no proof of those tests being false results but still trying to make something out of nothing? remember 2.8x was up to figure and they also gave 1.8x figure without amd technologies ? which is correct. but nobody talks about that for some obvious reasons. anyway it is irrelevant here.

no one is hyping vega (the guy merely compared pol10 to 4770) or anything here and infact we are doing exact opposite. ( even i said before don't trust wccftech and don't fell for hype) but i didn't bring useless bait.

 

[ShintaiDK]

They did state their test systems "Intel Core i7 4790K with 4x4GB DDR4-2600 MHz memory, Windows 10 64bit.
Star Wars Battlefront, X-wing Training
Med Preset @1080p. AMD Polaris GPU tested with 16.10 beta drivers . The GTX 950 tested with 359.06 drivers."

http://www.anandtech.com/show/9886/amd-reveals-polaris-gpu-architecture

so you have no proof of those tests being false results but still trying to make something out of nothing? remember 2.8x was up to figure and they also gave 1.8x figure without amd technologies ? which is correct. but nobody talks about that for some obvious reasons. anyway it is irrelevant here.

no one is hyping vega (the guy merely compared pol10 to 4770) or anything here and infact we are doing exact opposite. ( even i said before don't trust wccftech and don't fell for hype) but i didn't bring useless bait.

So why cant review sites show the same? Why isn't this data in line with what Average Joe will experience?

And GTX 950 is also the faster card in that game.

 

[AtenRa]

So why cant review sites show the same? Why isn't this data in line with what Average Joe will experience?

Perf/watt is better on 950 than it is on Polaris 11.

60fps cap, 60fps cap, 60fps cap

No review ever used 60fps cap and measured power consumption between Polaris vs GTX 950 in Starwars or any other game (especially not in DX-12).

 

[ShintaiDK]

60fps cap, 60fps cap, 60fps cap

No review ever used 60fps cap and measured power consumption between Polaris vs GTX 950 in Starwars or any other game (especially not in DX-12).

60 FPS cap would benefit the faster card more wouldn't it

And yes, we heard so many of your DX12 stories. Just a shame they didn't match reality

We are to believe the 950 system drew 56W more. Or 65% more power than the 460. Just no. Specially since there would be a lot of static power draw for both systems. Storage, memory, board, CPU.

 

[kraatus77]

So why cant review sites show the same? Why isn't this data in line with what Average Joe will experience?

Perf/watt is better on 950 than it is on Polaris 11.

Go ask those websites as to why nobody is testing same thing, and p/w is better on 950 because unlock fps. lock their fps to 60 and tone down settings in that specific game and you will get same results as amd. until then you can't say they lied and faked that demo.

950 is faster in ultra quality, not medium. in that same map and cpu.


and for the last time, why did you bring up those things here again ? saying pol10 hype failed is enough. everyone knows that. also reported.

 

[AtenRa]

We are to believe the 950 system drew 56W more. Or 65% more power than the 460. Just no.

You know it was RX 460 (896 shaders) ???

 

[Sweepr]

No review ever used 60fps cap and measured power consumption between Polaris vs GTX 950 in Starwars or any other game (especially not in DX-12).

Try finding a single test where Polaris draws 54W less than a (reference) GTX 950. That would make its performance per watt leaps and bounds above Maxwell, probably edging Pascal as well. Hard to believe. Anyway, this is the Vega thread so let's save this discussion for the other (existing) threads.

 

[Krteq]

And again... It was NOT RX 460 in AMD presentation. It was Polaris 10 GPU capped to 60 FPS.

 

[ShintaiDK]

And again... It was NOT RX 460 in AMD presentation. It was Polaris 10 GPU capped to 60 FPS.

And it was running at what, 400-500Mhz?

 

[kraatus77]

Another thread gone down the lane. why don't we just ban these wccftech so that this won't happen ?

one guy says something another guy misunderstands and revision/flame-wars starts ? there are ways to say something without flame-bait but is it too much to ask for ?

 

[AtenRa]

Try finding a single test where Polaris draws 54W less than a (reference) GTX 950.

1. We dont know which Polaris GPU was used
2. There is no review with 60fps cap

personal estimation.
At the time of the test (CES 2016 January) , GTX 950 2GB was selling at $150-170. So the test could have been made with an RX 470. I dont have the GTX 950 or any Polaris to test it so unless a review site does, nobody can say that AMD lied until proved otherwise.

 

[Sweepr]

Trying to get back on track, have you guys read this?

Magnum is a unique chip, it features a matrix of logic blocks that can be configured and programmed individually for any desired application or program. In other words, it’s the company’s first ever FPGA and its greatest attempt yet to expand its penetration into the high performance embedded market.

http://wccftech.com/amd-vega-10-vega-11-magnum

Thoughts?

 

[PhonakV30]

Mod , Where are you ?!
Is thread about Polaris's Power Consumption or Vega/Navi ?
Is someone allowed to say "Polaris is a big failure Architecture" ?

 

[Krteq]

Trying to get back on track, have you guys read this?



http://wccftech.com/amd-vega-10-vega-11-magnum

Thoughts?

This have been taken from AMD patent filed in Sep 2014 - Heterogenous Function Unit Dispatch in a Graphics Processing Unit

There is some chance to see this in Vega, but I think it is scheduled for Navi

 

[Glo.]

Trying to get back on track, have you guys read this?



http://wccftech.com/amd-vega-10-vega-11-magnum

Thoughts?

It may mean that next generation of AMD GPUs is made from higher number of smaller Shader Engines.

Fiji has 4 Shader Engines with 16 CU's in each. Much better idea for power efficiency and power gating is using 8 shader engines with 8 CU's in each one.

There are two patents, both have been posted in this thread. Both would benefit from beefier CU's. So there is another clue for Vega.

One of the most questionable and interesting things is the possibility of only 2048 Bit HBM memory controller for 14 nm Vega GPU.

Also the patents imply that architecture could do massive number of micro operation in out of order way to keep the pipeline completely fed, with cross line operations.

It is pretty much the same GCN architecture in terms of wave front, but the execution appears to be different, if I understand correctly the technology which is on GPUopen site, and those two patents. It is hard to predict how it will behave.

 

[maddie]

It may mean that next generation of AMD GPUs is made from higher number of smaller Shader Engines.

Fiji has 4 Shader Engines with 16 CU's in each. Much better idea for power efficiency and power gating is using 8 shader engines with 8 CU's in each one.

There are two patents, both have been posted in this thread. Both would benefit from beefier CU's. So there is another clue for Vega.

One of the most questionable and interesting things is the possibility of only 2048 Bit HBM memory controller for 14 nm Vega GPU.

Also the patents imply that architecture could do massive number of micro operation in out of order way to keep the pipeline completely fed, with cross line operations.

It is pretty much the same GCN architecture in terms of wave front, but the execution appears to be different, if I understand correctly the technology which is on GPUopen site, and those two patents. It is hard to predict how it will behave.

Why do you see this as questionable?

The memory controllers on the GPU Die are organized as 1 [1024 bit] controller/HBM stack. What's strange?

It certainly will lower costs as the HBM assembly appears to be a major cost factor.

 

[railven]

Dat Polaris downgrade! From HD 4870 to HD 4770.

AMD just needs to be competitive with current GTX 1080 for 2017. Judging even once AMD faithful's opinions (outside of here of course), it seems even they don't expect AMD to compete 1:1 with NV on the enthusiast/Halo tiers. And it seems they are okay with that. Perhaps the CPU mentality has caught on. Outside of the "10% of GTX 1080 in DX12" optimists, you see people now just expecting current mid-range performance at a more reasonable price opinions. I'm starting to see it more often in Zen related threads. AMD doesn't need to beat Intel/Nvidia to keep it's current (and possible future) buyers happy. Just be more competitive at a reasonable price. (I mean only a fool would think AMD would deliver something better than their rivals at a much lower price.)

I personally don't think Vega will compete with big pascal. Even big Vega will probably just be either a bit faster or on par with GP104.

I see Intel/Nvidia continue to price their bigger/faster products to the moon. They have found an audience there. For those not in that buying bracket, AMD can cater to them while not making Intel/NV margins, making enough to not become VIA/Matrox.

 

[AtenRa]

I personally don't think Vega will compete with big pascal. Even big Vega will probably just be either a bit faster or on par with GP104.

If by big Vega you mean 300-400mm2, it is a no brainer this Vega chip will only compete against GP104 (314mm2).

 

[Glo.]

I personally don't think Vega will compete with big pascal. Even big Vega will probably just be either a bit faster or on par with GP104.

May I know on what you predict this?

Why do you see this as questionable?

The memory controllers on the GPU Die are organized as 1 [1024 bit] controller/HBM stack. What's strange?

It certainly will lower costs as the HBM assembly appears to be a major cost factor.

Questionable from the idea of the HBM. As far as I know, from CPU design, having access to system memory in quad channel rather than dual channel makes it faster. The only way 2048 bit memory controller design can be enough is because the data can be accessed more frequently, and executed more frequently. Which is pretty much in line with the patents, about the variable wavefronts in each SIMD units.

One of the things that are becoming apparent for me from the patents we have seen is that all of them are related more to power efficiency rather than raw performance. For that, AMD would need beefier CU designs. But lets see what they will come up with.

 

[linkgoron]

Dat Polaris downgrade! From HD 4870 to HD 4770.

AMD just needs to be competitive with current GTX 1080 for 2017. Judging even once AMD faithful's opinions (outside of here of course), it seems even they don't expect AMD to compete 1:1 with NV on the enthusiast/Halo tiers. And it seems they are okay with that. Perhaps the CPU mentality has caught on. Outside of the "10% of GTX 1080 in DX12" optimists, you see people now just expecting current mid-range performance at a more reasonable price opinions. I'm starting to see it more often in Zen related threads. AMD doesn't need to beat Intel/Nvidia to keep it's current (and possible future) buyers happy. Just be more competitive at a reasonable price. (I mean only a fool would think AMD would deliver something better than their rivals at a much lower price.)

I personally don't think Vega will compete with big pascal. Even big Vega will probably just be either a bit faster or on par with GP104.

I see Intel/Nvidia continue to price their bigger/faster products to the moon. They have found an audience there. For those not in that buying bracket, AMD can cater to them while not making Intel/NV margins, making enough to not become VIA/Matrox.

Without talking about perf/watt - I find it hard to believe that big Vega will be on par with GP104, that would be a huge failure. I could easily see small Vega being Fury perf for cutdown (i.e. a step up from the 480) and between 1070 and 1080 for the non-cut. Cut Big Vega should be on par with 1080 and full Big Vega should be 15%+ from 1080. At least that's what I think, I just don't see how small Vega can be any worse given Polaris performance, and how four cards fit between the 480 and 1080.

 

[railven]

May I know on what you predict this?

Gut feeling.

Without talking about perf/watt - I find it hard to believe that big Vega will be on par with GP104, that would be a huge failure. I could easily see small Vega being Fury perf for cutdown (i.e. a step up from the 480) and between 1070 and 1080 for the non-cut. Cut Big Vega should be on par with 1080 and full Big Vega should be 15%+ from 1080. At least that's what I think, I just don't see how small Vega can be any worse given Polaris performance, and how four cards fit between the 480 and 1080.

It's probably best to say I'd rather put my expectations lower than get carried away with the hype train that seems to almost always take off around here. I didn't see Polaris as bad as some others did (even from the AMD camp) but then again I made sure to disregard the halo wishes of some posters and article writers.

Even if Big Vega is on par with big Pascal, the perf / watt numbers probably won't favor it. Therefore I don't see AMD trying to compete with NV. Aim for another Nano sans price. Give it moderate power consumption with performance but price it a little more reasonably.

I see Small Vega being as indicated, a Polaris replacements. Probably +10% on current Polaris but with better power/thermal/RAM options. I don see Small Vega beating GP104, especially OC'ed GP104. And it's why I see Big Vega being at OC'ed GP104 or a little higher, but not GP102 levels. I don't think AMD wants to run with the power hungry / hot trophy anymore.

 

[linkgoron]

I see Small Vega being as indicated, a Polaris replacements. Probably +10% on current Polaris but with better power/thermal/RAM options.

That would be a major disappointment if true. If full small Vega is Fury performance (that's 15% higher than the 480 according to TPU), that would be very disappointing (at least for me), and in that case big Vega will probably really be on par with the 1080. However, I find it unlikely (but not impossible, AMD haven't been performing really well lately).

I don't see Small Vega beating GP104, especially OC'ed GP104. And it's why I see Big Vega being at OC'ed GP104 or a little higher, but not GP102 levels. I don't think AMD wants to run with the power hungry / hot trophy anymore.

Yeah, I agree with you that full small Vega will not beat GP104, but I can see the uncut version between the 1070 and 1080 (closer to 1070), and the cut version being 15% higher than the 480.

It's probably best to say I'd rather put my expectations lower than get carried away with the hype train that seems to almost always take off around here. I didn't see Polaris as bad as some others did (even from the AMD camp) but then again I made sure to disregard the halo wishes of some posters and article writers.

In the earlier days, people were hyping big Vega as a GP102 competitor. Currently, that's unrealistic, but I don't think that big Vega beating the 1080 by 10%-15% is hype. Maybe a but optimistic, but hopefully given process improvements and new arch changes it's not impossible.

Even if Big Vega is on par with big Pascal, the perf / watt numbers probably won't favor it. Therefore I don't see AMD trying to compete with NV. Aim for another Nano sans price. Give it moderate power consumption with performance but price it a little more reasonably.

Given what we got with Polaris, I don't really have high hopes for perf/watt - but maybe that HBM magic sauce and new arch improvements will help. Also, even if AMD manage to beat GP104 they're are not even close to really being competitive with nvidia. Recall that even the new Titan X is a cut chip, and nvidia can easily either release a new flagship or just release a 1080ti.

 

[Glo.]

I see Small Vega being as indicated, a Polaris replacements. Probably +10% on current Polaris but with better power/thermal/RAM options. I don see Small Vega beating GP104, especially OC'ed GP104. And it's why I see Big Vega being at OC'ed GP104 or a little higher, but not GP102 levels. I don't think AMD wants to run with the power hungry / hot trophy anymore.

http://videocardz.com/63715/amd-vega-and-navi-roadmap
And direct quote:
Like I said yesterday, full details of VEGA 11 were not yet disclosed. However, I did tell you that Polaris 10 will be replaced by Vega 11 next year. Of course, what I meant was the professional market. It does not mean there won’t be Polaris 10-based Radeons next year. It means that more powerful solutions will be offered in Radeon Pro series with VEGA GPU.

 

[ShintaiDK]

I see Small Vega being as indicated, a Polaris replacements. Probably +10% on current Polaris but with better power/thermal/RAM options. I don see Small Vega beating GP104, especially OC'ed GP104. And it's why I see Big Vega being at OC'ed GP104 or a little higher, but not GP102 levels. I don't think AMD wants to run with the power hungry / hot trophy anymore.

I doubt Polaris chips are going to be replaced the next many years. Else its going to be a direct financial loss. Also the chips are going to be quite big, not to mention very expensive with HBM.

I see no way Vega is going to reach OCed GP104 performance. They be lucky if they even reach stock 1080 performance. I wouldn't be surprised if they end up with custom 1070 cards as the competition.

Vega 10=Fiji+Tessellation fix+VRAM increase+FP64+slight clock increase.