Vega/Navi Rumors (Updated)

[Despoiler]

NVidia has a CUDA lock-in though. Even if there are better options, they can't run CUDA and it's probably more expensive to replace that code than it is to eat some extra cost.

There is no longer a CUDA lock in. AMD's HIP can port 99.6% of CUDA code. The rest is a few weeks of work, but given the flexibility of AMD solutions, ie potentially much lower cost, the window is open.

Also, Otoy CUDA cross compiler.
https://www.bit-tech.net/news/hardware/2016/03/11/otoy-cuda-compiler/1

 

[Valantar]

NVidia has a CUDA lock-in though. Even if there are better options, they can't run CUDA and it's probably more expensive to replace that code than it is to eat some extra cost.

Didn't AMD announce some sort of CUDA emulation layer or some such last year? Or am I hallucinating again?


Edit: Despoiler beat me to it. And I was apparently not hallucinating. Good to know.

 

[lobz]

Direct links to Videocardz's images are broken.

yep, sadly you have to remove the 's' from the end of 'https'

 

[Crumpet]

https://www.reddit.com/r/Amd/comments/6bklro/we_are_radeon_technologies_group_at_amd_and_were/

Raja ask me anything, 25 minutes left, go nuts.

 

[Karnak]

We’ll be showing Radeon RX Vega off at Computex, but it won't be on store shelves that week. We know how eager you are to get your hands on Radeon RX Vega, and we’re working extremely hard to bring you a graphics card that you’ll be incredibly proud to own. Developing products with billions of transistors and forward-thinking architecture is extremely difficult -- but extremely rewarding -- work. And some of Vega’s features, like our High Bandwidth Cache Controller, HBM2, Rapid-Packed Math, or the new geometry pipeline, have the potential to really break new ground and fundamentally improve game development. These aren’t things that can be mastered overnight. It takes time for developers to adapt and adopt new techniques that make your gaming experience better than ever. We believe those experiences are worth waiting for and shouldn’t be rushed out the door. We’re working as hard as we can to bring you Radeon RX Vega.

On HBM2, we’re effectively putting a technology that’s been limited to super expensive, out-of-reach GPUs into a consumer product. Right now only insanely priced graphics cards from our competitors that aren’t within reach of any gamer or consumer make use of it. We want to bring all of that goodness to you. And that’s not easy! It’s not like you can run down to the corner store to get HBM2. The good news is that unlike HBM1, HBM2 is offered from multiple memory vendors – including Samsung and Hynix – and production is ramping to meet the level of demand that we believe Radeon Vega products will see in the market.

https://www.reddit.com/r/Amd/commen...gies_group_at_amd_and_were/dhqo7ha/?context=3

 

[swilli89]

tldr; There's no faster RX (consumer) Vega coming than air-cooled Frontier Edition.

Stop dreaming and be realistic.

tldr: You're wrong.

Raja Koduri: "RX Vega will actually be faster than Frontier version!"

https://videocardz.com/69662/raja-koduri-explains-where-radeon-rx-vega-in-reddit-ama

 

[Bacon1]

1. Was the card used in the gaming demos a Frontier Edition? And if so, was it the water cooled one or the air cooled version? A: It was an air-cooled version / It was running on Frontier Edition
   
   
2. Does the card profit from DX12 a lot or is DX12 performance similar to DX11 performance? A: Our architecture is very well suited for explicit APIs such as DX12 and Vulkan. If a game or a game engine prioritizes low level access to the GPU, Vega will soar. At the same time we're optimizing Vega for legacy APIs as well as much as possible.
   
   
3. Is there a difference in performance/clock speed between the water and the air cooled version or is one just quieter/cooler? A: There will be a slight difference in clock speeds, and therefore performance as well.

 

[Elixer]

Use https://www.reddit.com/user/gfxchiptweeter/ to see all of Raja's replies to Vega questions on his AMA.

 

[Bacon1]

1. Does Frontier Edition use 4 stacks or 2 stacks of HBM2? A: Frontier edition employs 2 stacks of HBM2
2. 480 GB/s of memory bandwidth is slower than Fiji's 512 GB/s, and that was with first generation HBM. When HBM1 on Fiji can match or beat these speeds, it sort of makes you wonder, what even is the point of using HBM2 anyway if configurations don't surpass Fiji's memory bandwidth? Besides PCB space savings and latency. A: Both Fiji's and Vega's HBM(2) implementations offer plenty of bandwidth for all workloads.

 

[Cookie Monster]

https://i.imgur.com/A3DH7rr.jpg

I made the comparison a few weeks ago.

Nice comparison. The savings arent huge in terms of die size but still better than the convential GDDR technology.

V100 is kinda stupid though. Nvidia's strat is an expensive, monolithic, one size fits all chip. AMD's is scalable solutions. Google's TPU 2.0 is faster, more power efficient, and more flexible given that it's an ASIC. EPYC + Vega + GTPUs allows a customer to tailor the system top to bottom for their calculation needs. They won't have too much or too little of anything. AMD will do better in the long run.

This is very ignorant especially when the buisness case is there for nVIDIA to produce these GPUs even with horrible yield rates (we dont know for sure but assuming it will be very low). They have massive super computer contracts to meet.

Just take a look into their latest datacenter revenue numbers. They aren't stupid.

But this is purely off topic

 

[Bacon1]

Here is another big one:

6) Many argue that vega is just a refined polaris gpu, how would you respond to this ?

My software team wishes this was true

Vega is both a new GPU architecture and also completely new SOC architecture. It's our first InfinityFabric GPU as well

And:

The new geometry pipeline in Vega was designed for higher throughput per clock cycle, through a combination of better load balancing between the engines and new primitive shaders for faster culling. As a programmer you shouldn't need to do anything special to take advantage of these improvements, but you're most likely to see the effects when rendering geometrically complex scenes that can really push the capabilities of the hardware.

RX will be fully optimized gaming drivers, as well as a few other goodies that I can't tell you about just yet....But you will like FE too, if you can't wait

 

[Crumpet]

And let the hype flow again

 

[Cookie Monster]

All abroad the hype train!

 

[T1beriu]

tldr; There's no faster RX (consumer) Vega coming than air-cooled Frontier Edition.

Stop dreaming and be realistic.

tldr: You're wrong.

Raja Koduri: "RX Vega will actually be faster than Frontier version!"

You truncated the complete answer: "RK: Consumer RX will be much better optimized for all the top gaming titles and flavors of RX Vega will actually be faster than Frontier version!"

That doesn't mean that that RX version is higher clocked. I think there's a chance that 40% is because of better gaming drivers, 40% higher clocked HBM (512GB/s) and 20% that FE is cut down chip. But I think it's kinda silly to think that a cut down chip is more expensive than a full one and AMD would be ballsy to do that. But recently they have been pretty ballsy.

12 more days till we find out.

LE: Another quote.

RX will be fully optimized gaming drivers, as well as a few other goodies that I can't tell you about just yet....But you will like FE too, if you can't wait

 

[Crumpet]

I think it's simply that Vega FE is configured more towards machine learning, and whilst is obviously pretty capable at gaming, RX Vega will be configured towards gaming.

(Also RTG is now looking into the possibility of an RX Vega 16gb model because people kept asking for it, but obviously that would be some way in the future now)

 

[Valantar]

You truncated the complete answer: "RK: Consumer RX will be much better optimized for all the top gaming titles and flavors of RX Vega will actually be faster than Frontier version!"

That doesn't mean that that RX version is higher clocked. I think there's a chance that 40% is because of better gaming drivers, 40% higher clocked HBM (512GB/s) and 20% that FE is cut down chip. But I think it's kinda silly to think that a cut down chip is more expensive than a full one and AMD would be ballsy to do that. But recently they have been pretty ballsy.

See:

3. Is there a difference in performance/clock speed between the water and the air cooled version or is one just quieter/cooler? A: There will be a slight difference in clock speeds, and therefore performance as well.

Most likely we're talking a clock bump in the 50-100MHz range. But that still makes a difference, if a slight one. I'm more psyched about this:

Here is another big one:

The new geometry pipeline in Vega was designed for higher throughput per clock cycle, through a combination of better load balancing between the engines and new primitive shaders for faster culling. As a programmer you shouldn't need to do anything special to take advantage of these improvements, but you're most likely to see the effects when rendering geometrically complex scenes that can really push the capabilities of the hardware.

 

[T1beriu]

Is there a difference in performance/clock speed between the water and the air cooled version or is one just quieter/cooler?
A: There will be a slight difference in clock speeds, and therefore performance as well."

I don't think Raja can't say white lies like that "Oh, when I said that some RX versions are faster than FE I was actually saying that it's only faster than the air cooled one, but not the water cooled version. I thought of this myself, but I can't think he would actually lie about it.

Again, why there's a water-cooled FE if it's not pushed to its limits like the Fury X?

 

[Malogeek]

My excitement level just soared again. Neeeeeeeed Vega.

 

[CatMerc]

I have thought of an odd possibility. The gaming cards might be cut down not to make them weaker, but rather to allow more thermal headroom for higher clocks.
Higher clocks with less shaders would mean that your shader to fixed function performance ratio is higher than FE. And shaders cost energy, so it might be a balancing act there.

So:
Frontier Edition: 1x Shaders, 1x Clockspeed, 1x TFLOPS, 1x Geometry, 1x Rasterization, 1x Watt

RX Vega: 0.875 Shaders, 1.11x Clockspeed, 0.97x TFLOPS, 1.11x Geometry, 1.11x Rasterization, 1.05x Watt



I thought of this possibility mainly because of how the balance of ROPS, Geometry Engines, and other such hardware to shaders remained the same as with Fiji (even if they recieved targeted improvements). And we know that in gaming the fixed function parts of Fiji were usually the weak points.

I don't expect this to happen, but it's just a random thought I've had.

 

[Paratus]

I liked this comment:

Infinity Fabric allows us to join different engines together on a die much easier than before. As well it enables some really low latency and high-bandwidth interconnects. This is important to tie together our different IPs (and partner IPs) together efficiently and quickly.
It forms the basis of all of our future ASIC designs.

We haven't mentioned any multi GPU designs on a single ASIC like Epyc, but the capability is possible with Infinity Fabric

Note: They didn't mention it. Not they didn't do it.

 

[Maverick177]

So no confirmation or at the very least an approximation of launch date yet again.

 

[swilli89]

You truncated the complete answer: "RK: Consumer RX will be much better optimized for all the top gaming titles and flavors of RX Vega will actually be faster than Frontier version!"

That doesn't mean that that RX version is higher clocked. I think there's a chance that 40% is because of better gaming drivers, 40% higher clocked HBM (512GB/s) and 20% that FE is cut down chip. But I think it's kinda silly to think that a cut down chip is more expensive than a full one and AMD would be ballsy to do that. But recently they have been pretty ballsy.

12 more days till we find out.

LE: Another quote.

When he says "flavors of RX Vega will be faster" I take that to assume certain models will in fact have a higher clockspeed.. otherwise if its just optimization for RX then we would probably assume all RX would be faster?

But yes.. we will find out in 12 days.

 

[JDG1980]

There is no longer a CUDA lock in. AMD's HIP can port 99.6% of CUDA code. The rest is a few weeks of work, but given the flexibility of AMD solutions, ie potentially much lower cost, the window is open.

Also, Otoy CUDA cross compiler.
https://www.bit-tech.net/news/hardware/2016/03/11/otoy-cuda-compiler/1

This is good, but AMD is going to need to be proactive. It's not enough to just throw the tools out there and hope they get used, they have to actually send engineers out to customer sites to help them port their code. As the underdog, they've got to try harder if they want to pick up market share.

 

[PhonakV30]

Q: Many argue that vega is just a refined polaris gpu, how would you respond to this ?
A: My software team wishes this was true
Vega is both a new GPU architecture and also completely new SOC architecture. It's our first InfinityFabric GPU as well

So It's not like GCN just NCU

 

[tential]

So they're seriously using the same fe moniker.... Lol....

Sent from my Nexus 5X using Tapatalk