AMD Radeon RX VEGA to Launch at SIGGRAPH 2017

[Patrik1]

If it is 1080 performance with a very good price-tag, it`s not bad.. Not bad at all, because CPU:s is their main business, and the CPU:s they got today are actually Fantastic! in my opinion.

[PCGuy_5960]

37 minutes ago, Andrew Sh said:

Because AMD were showing off Vega beating Nvidia a looong time ago. 

Nope, in their demos Vega was on par/slightly worse than a 1080

[GDRRiley]

drivers if done correctly and a bit higher clock will give it a 10-15% boost. AMD is known for having slowly developing drivers. 

Edited July 2, 2017 by GDRRiley

[AlwaysFSX]

34 minutes ago, Andrew Sh said:

Also don't forget that it's supposed to be faster than the Vega FE since that isn't even a gaming gpu.
It's like comparing a Quadro with 1080ti.

You know, 5 years ago that statement might have held some weight, not much. But you're wrong.

 

Also for your second statement, that's been done.

 

[PCGuy_5960]

Just now, GDRRiley said:

drivers if done correctly and a bit higher clock will give it a 10-15% boost. AMD is known for having slowly developing drivers. 

Which would put it on par with a 1080. I really hope that they get the pricing right

[GDRRiley]

2 minutes ago, PCGuy_5960 said:

Which would put it on par with a 1080. I really hope that they get the pricing right

in some games it would beat it. It may be my new card if I want to get some more power. 

[PCGuy_5960]

2 minutes ago, GDRRiley said:

in some games it would beat it. It may be my new card if I want to get some more power. 

Yep, I believe that in games like GTA V, the 1080 will be better and in games like Doom the RX Vega will be faster. In other words, the 1080 and RX Vega will be trading blows

[DXMember]

6 minutes ago, PCGuy_5960 said:

Yep, I believe that in games like GTA V, the 1080 will be better and in games like Doom the RX Vega will be faster. In other words, the 1080 and RX Vega will be trading blows

so Raja was saying infinity fabric has played a major role in Vega - like a all the stuff like memory subsystem, schedulers, execution units are all interconnected with infinity fabric...

now this is what I purpose - PLOT TWIST - RX VEGA IS 2x VEGA FE ON ∞-FABRIC

[Agosto]

3 minutes ago, DXMember said:

now this is what I purpose - PLOT TWIST - RX VEGA IS 2x VEGA FE ON ∞-FABRIC

1.21 JIGOWATTS

[Terryv]

21 minutes ago, PCGuy_5960 said:

Nope, in their demos Vega was on par/slightly worse than a 1080

Source?

 

As far as I know, that demo was locked at 60fps on both cards without any fps counter displayed.

[PCGuy_5960]

Just now, Terryv said:

Source?

 

As far as I know, that demo was locked at 60fps on both cards without any fps counter displayed.

Their Doom Demo, it was on par with a 1080, and the Battlefront Demo, it was slightly worse than a 1080 iirc

[Carclis]

I'm not really buying the whole Vega having GTX 1070 performance story. PCPer and others have found that game performance is typically 15%, but up to 50% faster than the Fury X with a 60% higher clock speed and on top of that a lot of the touted new and improved features just aren't working. Either AMD is sandbagging, their driver software isn't up to snuff yet or this is the biggest disaster of a GPU release in AMD's history. Hopefully it'll all be clear after SIGGRAPH.

[Terryv]

9 minutes ago, PCGuy_5960 said:

Their Doom Demo, it was on par with a 1080, and the Battlefront Demo, it was slightly worse than a 1080 iirc

I'll concede on doom as it had performance metrics (although this s was very early silicon) but there was no such information for Battlefront.

[tiggar47]

4 hours ago, L.Lawliet said:

So this is it. Disappointment or one hell of a gpu.

Don't mix GPU with Graphics card, GPU is not everything that makes a good Graphics Card.

[Dash Lambda]

8 minutes ago, tiggar47 said:

Don't mix GPU with Graphics card, GPU is not everything that makes a good Graphics Card.

But AMD makes the GPU, and the specs are pretty much consistent across cards because the chip itself isn't made to support that varied a platform.

So we're waiting for information on AMD's GPU, and the graphics cards are what come from that. Not a lot of meaningful confusion happening here.

[Taf the Ghost]

1 hour ago, LAwLz said:

Just some thoughts that I think people should be aware of:

1) It might just be a paper launch. Don't expect cards to be in stores straight away. 

2) Vega will not compete against Pascal. Pascal is old at this point. Over a year old. Vega will have to compete against Volta, which should be out fairly soon. 

3) The workstation cards have never been that much worse than the gaming cards for gaming, so stop with the "it's not a gaming card!" excuses. They might improve the drivers in the future but that's about it. Don't expect RX Vega to be magically better than Vega FE for gaming. 

4) Do not trust AMD's own marketing and benchmarks. Don't take AMD's word for how it will perform. Wait for independent benchmarks to come out. 

1) Given the timings, probably late August might be about right. We might actually blame Apple for this issue more than HBM2 at this point.

2) It's going against Pascal, that's a given. Whether it's against the top-tier GPUs is a different issue. 

3) For Nvidia cards, generally true, but AMD has tended to release their Workstation cards last, while Nvidia is generally the other way around. Vega FE is going to be in the ballpark of the RX Vega, we just don't know how much.  20% uplift and it's a 1080 Ti competitor. 

4) Agreed, though the Vega hype seems to be mostly community generated.  Plus some paid Intel shilling. (Not normally here, but seen some tell-tale signs of it during the whole Computex/"I'm buying a 1080" situation.)

 

As with all tech, wait until we actually have it. Vega FE's most interesting issue is the low-level drivers. Are they built in such a way to allow Gaming & Pro mode more easily? Without a detailed understanding of that aspect, we don't know if they're operating the same way the Titan cards are. There are always trade-offs in hardware design, so I'm quite curious if the approach is different. (This is more praise for Nvidia's engineers than an issue with Vega FE.)

 

The last and kind of key issue no one is talking about yet: RX Vega + Ryzen. Will the scheduler & drivers handle spanning across multiple cores better? This matters, especially for DX12, as there's issues with the Nvidia driver + 1080/Ti cards + Ryzen. Something like Hitman benchmarks are going to be really interesting.

[DoctorZeus]

this isnt a gaming card they say... neither is the Titan Xp or the Quadros. yet they game well...

 

thus a pro card they say.... that wont receive pro certified drivers...

 

This is slower than two 580s they say.... a 1080 ti is slower than two 1070s... Nvidia shouldn't have released it then.... I'm sure the real gaming pascal card is coming soon....

 

Drivers will fix performance.... Volta will release before amd has good drivers....

 

RX vega will perform between a 1060 and above titan Xp.... and i can run somewhere  between 0 and 1000 miles per hour...

 

I know releasing an equalish card for cheaper is nice [Titan x pasc >> 1080 ti] but that was 500 dollars cheaper. 

 

1080 performance a year and a half later with the super duper awesome (manufacturer delayed) hbm 2 would also be sad. good value? probably. exciting? dont think so? too much attacking and apologizing on the wrong card but hey. amd been taking days to release so....

sad life.

 

 

 

[Vode]

2 hours ago, DXMember said:

the weird thing is that right now Vega FE is doing like +15% more gaming at most of the Fury X at +60% clock speeds, which is confusing as hell since it has the same amount of shader cores and an architectural advantage going for it 

They will probably enable Tiled Rasterizing for the RX Vega gaming cards and then it will perform as expected. Between 1080 and 1080Ti but closer to the former.

[Taf the Ghost]

1 hour ago, DXMember said:

so Raja was saying infinity fabric has played a major role in Vega - like a all the stuff like memory subsystem, schedulers, execution units are all interconnected with infinity fabric...

now this is what I purpose - PLOT TWIST - RX VEGA IS 2x VEGA FE ON ∞-FABRIC

So, I had a nice response that got eaten.

 

Vega is the step before they're at that point. The 2018 release for the Vega Pro Duo could very well see a fully attached packages, but the dual GPU stuff comes with Navi. That will be a big game changer, especially given where AMD's design prowess in GPUs actually is. (They're best at iGPU tech, which is why scaling that up eats more power than Nvidia's approach.)

[DXMember]

2 minutes ago, Taf the Ghost said:

So, I had a nice response that got eaten.

 

Vega is the step before they're at that point. The 2018 release for the Vega Pro Duo could very well see a fully attached packages, but the dual GPU stuff comes with Navi. That will be a big game changer, especially given where AMD's design prowess in GPUs actually is. (They're best at iGPU tech, which is why scaling that up eats more power than Nvidia's approach.)

that's what they want you to believe

July 30th - the game changer. complete 540*, OUT OF BLUE - DOUBLE VEGA ON ∞-FABRIC

[LAwLz]

31 minutes ago, Taf the Ghost said:

2) It's going against Pascal, that's a given. Whether it's against the top-tier GPUs is a different issue. 

But it's not... Volta will be released soon, so there will only be a few months at most where Vega is competing against Pascal. Once Volta is out, it will be Vega vs Volta until AMD releases a new architecture, which won't be for like another year.

So Vega will go up against Pascal for like ~2 months maybe, and it will be competing against Volta for some 10 months.

Vega will need to be as good or better than Volta, not Pascal.

 

33 minutes ago, Taf the Ghost said:

Vega FE is going to be in the ballpark of the RX Vega, we just don't know how much.  20% uplift and it's a 1080 Ti competitor. 

20% uplift is not "ballpark numbers".

[Taf the Ghost]

1 minute ago, DXMember said:

that's what they want you to believe

July 30th - the game changer. complete 540*, OUT OF BLUE - DOUBLE VEGA ON ∞-FABRIC

The one Ryzen tech that no one really talks about is the ARM processor on the die. It's there for AES work and is stupidly fast at it. I know about GPU rendering pipelines to know that I can't comment deeply, but the use of Infinity Fabric brings up some really interesting discussions about things you could place on the interposer massively accelerate specific tasks for a GPU.  Not just for gaming, but for professional work as well. 

 

Graphics cards are still a form of general computing unit, though highly focused, which is why CUDA ended up being so important for Nvidia. Infinity Fabric brings up a lot of new potential applications we haven't even thought much about yet. The obvious power is to eventually have opaque Dual GPU boards. (If the heat/power could be handled, 4x GPU cards would also make a lot of sense.) 

[DXMember]

14 minutes ago, Taf the Ghost said:

The one Ryzen tech that no one really talks about is the ARM processor on the die. It's there for AES work and is stupidly fast at it. I know about GPU rendering pipelines to know that I can't comment deeply, but the use of Infinity Fabric brings up some really interesting discussions about things you could place on the interposer massively accelerate specific tasks for a GPU.  Not just for gaming, but for professional work as well. 

 

Graphics cards are still a form of general computing unit, though highly focused, which is why CUDA ended up being so important for Nvidia. Infinity Fabric brings up a lot of new potential applications we haven't even thought much about yet. The obvious power is to eventually have opaque Dual GPU boards. (If the heat/power could be handled, 4x GPU cards would also make a lot of sense.) 

like the multimedia accelerators

or are you suggesting embedded on board hash accelerators for faster mining?

[Taf the Ghost]

48 minutes ago, DXMember said:

like the multimedia accelerators

or are you suggesting embedded on board hash accelerators for faster mining?

I kind of want to hit you for the Hash Accelerator comment, haha. 

 

But they could, or other things along those lines. Could lead to significantly cheaper Hardware Decoding, or even more stuff in that direction.  Little 20 cent chips on the board around the GPU could handle specific tasks without having to put stuff on-die. We could also see a migration of tasks off-die as as well, to keep down die space and design complications. We don't know, as while everyone has gone the SoC route, AMD might actually be heading in a completely different direction.

 

Think would would happen if rather than big, 64 CU cores, AMD only produces 16 CU cores, then can stack them from 8 CU (so a 1/2 cut) all the way to 128 CU. They're all one "package", but it ends up being up to 8 real dies.  Lots and lots of options, which is why AMD is headed that way.

[DXMember]

5 minutes ago, Taf the Ghost said:

I kind of want to hit you for the Hash Accelerator comment, haha. 

 

But they could, or other things along those lines. Could lead to significantly cheaper Hardware Decoding, or even more more stuff in that direction.  Little 20 cent chips on the board around the GPU could handle specific tasks without having to put stuff on-die. We could also a migration of tasks off-die as as well, to keep down die space and design complications. We don't know, as while everyone has gone the SoC route, AMD might actually be heading in a completely different direction.

 

Think would would happen if rather than big, 64 CU cores, AMD only produces 16 CU cores, then can stack them from 8 CU (so a 1/2 cut) all the way to 128 CU. They're all one "package", but it ends up being up to 8 real dies.  Lots and lots of options, which is why AMD is headed that way.

competition is green with envy, having to produce 800mm² dies