Poll: ** The AMD VEGA Thread **

[DONOHUE07]

According to Pcper it's not, someone else mentioned it may be because of the memory latency. I'm not sure why you think it won't sell, The only thing that'll stop it selling is if AMD ask silly money for it.

Which they will do.
the rx480 = the year old 970, same price bracket
FuryX = the year old 980ti, same price bracket
Vega won't be any different, the days of AMD being value for money are over, there just a year behind.
And I have only ever bought AMD purely because of there price to performance ratio, I hope I'm wrong as I don't really want to buy NVIDIA if I can help it.

 

[N19h7m4r3]

When that Doom test was run wasn't it shown that an overclocked Fury X quite close? Plus it may of been beating the 1080 but that is in Vulcan where we have a significant advantage. In regular DX11 games going tit for tat it would have been slower.

The engineering sample was besting the 1080 in doom Vulkan yes. Now Frontier Edition is losing to it in doom Vulkan with gamer nexus tests. No wonder it can barely match 1070 in DX11 tests.

This Frontier Edition makes no sense at all.

 

[N19h7m4r3]

AMD better hope Vega does good at mining because gamers aint buying this **** lol.

PcPer tests show it on par with an RX 580 :|

Which they will do.
the rx480 = the year old 970, same price bracket
FuryX = the year old 980ti, same price bracket
Vega won't be any different, the days of AMD being value for money are over, there just a year behind.
And I have only ever bought AMD purely because of there price to performance ratio, I hope I'm wrong as I don't really want to buy NVIDIA if I can help it.

The Fury X launched two months after the 980ti.....

 

[WeSbO]

40% faster than the FuryX seems ok to me, it just needs to be priced right. It will also improve over time as all the other AMD cards have. The days of 100% performance boost each gen are long gone.

Nowhere near 40% faster than a Fury X (the only ones that put it up against an FX), well not out of the box on the PCper review, it averages out at 25% faster if you include ROTTR (which looks like an outlier) and 20% if you remove ROTTR, all that with a 37% ish clock GPU boost (1440mhz) and a 89% higher VRAM clock (945mhz), but lower bandwidth (483 gb/s) and bus size (2048 bit), because of only having 2 stacks

 

[TNA]

Is the rollercoaster on the way down at the moment? The pages are moving to fast to keep up

I still hold on to hope. RX Vega cannot be as bad as the FE. If so RTG are doomed until Navi comes out. They would have to sell it for £300 if they wanted my money for such performance.

If they price Vega wrong, they will lose out on a lot of sales and miss the change to improve their market share. I would rather wait another 5-6 months and buy 1080Ti performance for £400 or less from a brand new card with Volta.

 

[LoadsaMoney]

so its a smaller die than the FE, must be different then, also still saying, it is faster than the FE at gaming, and also has extra goodies the FE doesn't, so im presuming, these extra goodies, are for gaming.

AMD’s Raja Koduri Confirms Vega 10 Die Size At 484mm² – RX Vega To Be Faster In Gaming Than Frontier Edition

AMD’s Jason Evangelho also went on record to state that gamers should not draw conclusions about Vega’s gaming performance based on Vega Fonrtier Edition benchmarks. Vega Frontier Edition was designed and optimized for a vastly different set of tasks and use cases aimed at the professional market. On the other hand RX Vega graphics cards coming at the end of the month are specifically designed and optimized for games.

http://wccftech.com/amds-raja-koduri-confirms-vega-10-gpu-is-484mm²-large-rx-vega-gaming-cards-july-launch-confirmed/

 

[CAT-THE-FIFTH]

In the Gamer Nexus video the gaming relevant tests start at 16 min 58 sec.

The memory downclocking meant the card was no good for overclocking and they only included one example and that's the first test (Firestrike) at the time stamp above.

This may be why other's results are so poor, I don't recall Pcper mentioning it during the livestream though, that said I was tired at the time so I could have missed it. You'd of thought that they'd of noticed it if it is an issue we'll all experience with the card. Personally I hope it is an issue across all the FE cards as it will give us a reason for the dire results and the potential for RX to be a lot better than the FE makes it look.

Hmm,if the memory is doing that it would not help performance at all.

Back on the hype train we go!!

 

[Panos]

So it is replacing GCN all be it based from GCN?

NCU is based on the GCN ideas, but is new core designed from ground up.

Also if it was the same as the GCN cores, like Hawaii/Fiji/Polaris, he wouldn't say that the software engineers are in despair.
They could have released a new iteration of the drivers with the new stuff bolted.
But is not. Is new.

And all those WCC<whatever> rumour mills "is an upgrade to GCN iteration", where everyone else takes as gospel in here; Are the same who were writing about Zen been an upgraded iteration to Bulldozer or Piledrive up to March, not a brand new core design. Even if Lisa Su had said so back in December, that is brand new architecture, they kept the same mantra until we had the products on our hands.

Ofc after that they completely brushed under the carpet their previous statements, and given the goldfish memory timespan many of their readers have today, they continue taking their trashwriting as gospel.

 

[Freddie1980]

so its a smaller die than the FE, must be different then, also still saying, it is faster than the FE at gaming, and also has extra goodies the FE doesn't, so im presuming, these extra goodies, are for gaming.

AMD’s Raja Koduri Confirms Vega 10 Die Size At 484mm² – RX Vega To Be Faster In Gaming Than Frontier Edition



http://wccftech.com/amds-raja-koduri-confirms-vega-10-gpu-is-484mm²-large-rx-vega-gaming-cards-july-launch-confirmed/


http://wccftech.com/amds-raja-kodur...e-rx-vega-gaming-cards-july-launch-confirmed/

Maybe Vega 10 is the FE edition intended for entry level HPC tasks and Vega 11 and 12 will be the RX editions which will be the initial base for Vega for optimising the software and driver code. Latter on they might give us a full fat 8Gb RX Vega 10 version.

 

[beany_bot]

Vega = 590

At best.

 

[zornyan]

Agreed, I think Pascal will have a far longer shelf-life compared to Maxwell

funny actually , Nvidia just updated Fermi and older cards to be dx12 compatible so they can use newer features.

also for anyone reading, Nvidias newest driver let's you stream Netflix in 4k, you no longer need a kabylake cpu for it if you have an Nvidia card

 

[WeSbO]

so its a smaller die than the FE, must be different then, also still saying, it is faster than the FE at gaming, and also has extra goodies the FE doesn't, so im presuming, these extra goodies, are for gaming.

AMD’s Raja Koduri Confirms Vega 10 Die Size At 484mm² – RX Vega To Be Faster In Gaming Than Frontier Edition



http://wccftech.com/amds-raja-koduri-confirms-vega-10-gpu-is-484mm²-large-rx-vega-gaming-cards-july-launch-confirmed/

That article is laughable, Raja did not confirm 484mm2 die size (nowhere at all), he confirmed that the whole package was a perfect square (HBM+die+interposer) and that PCper was the closest with his measurements (513mm2) that die is not 484mm2 well not on the vega FE in all cases.

 

[CAT-THE-FIFTH]

That reddit is laughable, Raja did not confirm 484mm2 die size (nowhere at all), he confirmed that the whole package was a perfect square (HBM+die) and that PCper was the closest with his measurements (513mm2) that die is not 484mm2 well not on the vega FE in all cases.

https://twitter.com/GFXChipTweeter/status/881370308694822913

So that is between 484MM2 ~ 529MM2.

 

[KompuKare]

Well, it's good that someone from AMD is trying to do some damage limitation I guess.

Vega Frontier Edition was designed and optimized for a vastly different set of tasks and use cases aimed at the professional market. On the other hand RX Vega graphics cards coming at the end of the month are specifically designed and optimized for games.

There is only one die though, so 'designed' is a very poor word to use. About the only ways one die could be said to have designed for different purposes would be if some of the compute features could be fused off in such a way that the remain areas of the chip could clock higher or use (significant) less power; or if via Infinity Fabric they were able to offload compute features to a separate chip.

We know the second possible is not the case since we have seen Vega FE disassembled and it only has the main chip and the two HBM2 chips. Something like that might be possible when Navi comes out.

 

[Freddie1980]

Hmm,if the memory is doing that it would not help performance at all.

Back on the hype train we go!!

Primative shaders are still to be enabled. There's a ton of efficiency to gained with these enabled as currently GNC will render a scene that may have 220m polygons with only 2m visible to the player, the primative shaders stops these polygons going through the geometry pipeline freeing up capacity.

 

[Mauller]

The engineering sample was besting the 1080 in doom Vulkan yes. Now Frontier Edition is losing to it in doom Vulkan with gamer nexus tests. No wonder it can barely match 1070 in DX11 tests.

This Frontier Edition makes no sense at all.

considering the focus on pro workloads, there is a high chance that the vega FE drivers, as mentioned on reddit, are not using many if any of the latest architecture features available with Vega yet. Since those features are not considered code stable for the use in professional workloads. And since it was confirmed that it uses the same driver for gaming, then games won't see the advantages either.

Someone on reddit also looked into the vega FE drivers, he seemed to find that Vega FE is using a compatibility path to essentially emulate a fiji part, with little in the driver coding for any of the newer vega features.

 

[WeSbO]

https://twitter.com/GFXChipTweeter/status/881370308694822913

So that is between 484MM2 ~ 529MM2.

Yep, somewhere between 500 and 512 seems to be the best bet (so Fiji would be somewhere around 16 to 20% ish bigger), Why the heck would AMD not just give the bloody dimensions ???? lol

 

[Mauller]

Yep, somewhere between 500 and 512 seems to be the best bet (so Fiji would be somewhere around 16 to 20% ish bigger), Why the heck would AMD not just give the bloody dimensions ???? lol

it will be closer to 484, since he did mention it was under 500mm^2 previously, so anywhere between 484 and 500. when i looked at PCPers teardown, he was out by 1mm when measuring the cores length, which to me put it around 493mm^2.

 

[CAT-THE-FIFTH]

Yep, somewhere between 500 and 512 seems to be the best bet (so Fiji would be somewhere around 16 to 20% ish bigger), Why the heck would AMD not just give the bloody dimensions ???? lol

He said under 500MM2 before,so its probably why people think it was 484MM2,but at least its not massively bigger than the GP102.

If AMD can get clockspeeds up with an AIO water cooler in the top model,and get the gaming drivers in order,as long as they can beat the GTX1080 convincingly then it will be enough IMHO.

 

[N19h7m4r3]

considering the focus on pro workloads, there is a high chance that the vega FE drivers, as mentioned on reddit, are not using many if any of the latest architecture features available with Vega yet. Since those features are not considered code stable for the use in professional workloads. And since it was confirmed that it uses the same driver for gaming, then games won't see the advantages either.

Someone on reddit also looked into the vega FE drivers, he seemed to find that Vega FE is using a compatibility path to essentially emulate a fiji part, with little in the driver coding for any of the newer vega features.

That's just another issue so far. The engineering sample was running Debugging Fiji drivers, on debugging usb monitoring PCB, and what seems to be 1200Mhz. Yet performed better than Frontier Edition.

So many unanswered questions and issues that need to be addressed.