New Upcoming ATI/AMD GPU's Thread: Leaks, Hopes & Aftermarket GPU's

I didn't think tile based renderer does culling - it just batches the tiles up and keeps them in cache instead of pushing the entire frame too and from memory - which decreases the needed bandwidth and obviously saves power. Like maybe the same hardware unit that does the tiling also does culling - but I don't think the two are completely related.

Tom Peterson also mentioned that it uses GPU cycles, so it actually slightly lowers performance. They also turn it on/off in driver profiles, because it isn't worth doing in some titles.

Also David Kanter's company has provided consulting services to AMD/Nvidia and other major tech companies. He's not like a random youtuber or something - he works in the industry.

 

Raja Koduri explains where is Radeon RX Vega in Reddit AMA

Here are the most interesting questions and answers from Reddit’s AMA -> https://www.reddit.com/r/Amd/comments/6bklro/we_are_radeon_technologies_group_at_amd_and_were/

Raja Koduri: I want to start things off today but saying thank you to everyone for all of your excitement, energy and enthusiasm for all things AMD, and in particular, for Vega.

Earlier this week we were thrilled to launch Radeon Vega Frontier Edition. We think it will have a big impact on machine intelligence and content creators. I also know some of you are disappointed that we didn’t launch RX Vega as well.
I wanted to hold this AMA and have an open discussion with you about our Vega launches. And while we’re not launching RX Vega today — so I won’t be talking about pricing or launch date — there are lots of rumors and innuendo I want to put to bed, and there are plenty of questions I can answer.
I know you guys can’t wait to see Radeon RX Vega. I know that a lot of you guys obsess over when you’ll be able to game on Vega. Your passion for all things Radeon is what drives every single person in the Radeon Technologies Group to push hard on all fronts – hardware and software engineering, display technologies, and successful developer relationships.
Everyone at AMD sees how much you guys talk about Vega and how eager you are to get your hands on it, and it fuels us.

Please know that we’re working incredibly hard on Radeon RX Vega. You are the lifeblood of Radeon, and you all deserve a graphics card that you’ll be incredibly proud to own.

Elmnator: Will the consumer RX version be as fast at the Frontier version?

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!

RaverendCatch: Raja, are you not shaving your beard until Vega is launched?

RK: Yes

TitanicFreak: My first question for you is what is your vision with Vega? What do you see Vega excelling in? Specifically in the consumer market. Do you see people using Vega similar to how its predecessor (Fiji) was used? Where both the Fury X and Nano excelled in m-ITX builds. Do you see Vega continuing that?
My other question would be what is the difference between the Blue and Gold variants of the Frontier Edition. Do they share a similar TDP and clock speed with the Gold edition merely being more quiet due to it’s liquid cooling? Or is there something more separating the two apart?

RK: Primary vision with Vega was to establish our next generation architecture that is capable of dealing with large data-sets (tera, peta,exa etc)
From a gaming perspective we wanted to build a product that tackles the challenging 4K@60Hz for AAA gaming…
Like Fiji Vega will excel in small form factors etc due to HBM2 advantages
Yes – the gold version may have more thermal headroom that could help in some scenarios

nas360: Why the official Frontier page shows renders of the card with 2x8pin connectors and the one you were holding at the presentation has 1×8 + 1×6 connectors?

RK: I grabbed an engineering board from the lab on the way to the Sunnyvale auditorium, and that boards works well with a 6 and an 8 pin. We decided to put two 8 pin connectors in the production boards to give our Frontier users extra headroom

Pepri: 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?

RK: It was an air-cooled version

Does the card profit from DX12 a lot or is DX12 performance similar to DX11 performance?

RK: 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.

Is there a difference in performance/clock speed between the water and the air cooled version or is one just quieter/cooler?

RK: There will be a slight difference in clock speeds, and therefore performance as well.

RA2lover: How difficult is the process of developing drivers for a GPU architecture with so many differences from previous product designs?

RK: Developing drivers for new architecture is one of the most complex and difficult engineering tasks for a GPU company…In fact this is one of the reasons why there are only so few GPU companies.

WallyWest: Is the Frontier Edition a card like a Titan X (a professional/gaming card?), will we have the choice between RX driver and Pro Driver? Is it a Pro Card or a Gaming Card? Or both?

RK: The Frontier Edition was designed for a variety of use-cases like Machine Learning, real-time visualization, and game design. Can you play games on Frontier Edition? Yes, absolutely. It supports the RX driver and will deliver smooth 4K gaming. But because it is optimized for professional use cases (and priced accordingly), if gaming is your primary reason for buying a GPU, I’d suggest waiting just a little while longer for the lower-priced, gaming-optimized Radeon RX Vega graphics card

wickedplayer494: Does Frontier Edition use 4 stacks or 2 stacks of HBM2?

RK: Frontier edition employs 2 stacks of HBM2

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

RK: Both Fiji’s and Vega’s HBM(2) implementations offer plenty of bandwidth for all workloads. (nalasco – need help here)

Can we please get the ability to overclock HBM2?

RK: We’ll see what we can do about that

Can we pretty please get a 16 GB variant of Radeon RX Vega?

RK: We will definitely look at that…

Proxiros: Thank you for this AMA knowing how valuable your time is. I don’t expect that you will reveal much today (NDA) but the only thing that all await is: Will Vega for consumers revealed at Computex ( http://www.amdcomputex.com.tw ) this year? or at least a launch date? Keep up the good work!

RK: 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.

RA2lover: What things does the RX Vega have over the Radeon Vega FE that would make it worth the extra wait?

RK: 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

anihallatorx: I understand that the Vega architecture is focused mainly on increasing/enabling performance on large datasets, like utilizing HBM2, HBCC etc to fuel a vision of high frame rate 4K, VR and photorealistic situations. Where does it stand on the compute side of things? Like a new geometry engine?

RK: On the compute side of things..Vega FE will be the fastest single GPU solution (>12.5 TFlops FP32) when it’s available and our NCU packs several additional optimizations, including Rapid-Packed-Math which delivers >25 TFLops of FP16

KoolNerdz: What does Vega mean for Nvidia Volta? Is there gonna be a different design? Tell me a joke.

RK: One thing for certain is that Vega Instinct is well positioned to deliver dramatically better performance per dollar;, and TCO(total-cost-of-system ownership) is probably the most important metric to our machine learning and hyperscale customers and combined with Epic/Naples – Vega is extremely attractive..
You want a joke
Vega needs some extra Volta(ge) for overclocking

Nicolii: How will HBC effect content creation to 3d artists?
Does it help in any way in 3d programs? How so?

RK: Having 16GB of HBC on board will allow 3d artists to work on larger and even more complex models than ever before. Depending on the workload we have seen scenarios where 16 GB of HBC is effectively same performance as having 32 GB or 64 GB of regular VRAM

 

Some interesting answers there. Vega might be looking good after all.

 

I certainly hope they will deliver... NVIDIA is milking their enthusiast customers for a way too long.

What I actually don't understand is the hype around HBCC. From my point of view it will need app support to be able to stream data straight into the VRAM? Considering how much VEGAs are used in games (none currently), I don't know how many developers will be eager to bother with it.

 

The HBCC controller on Vega doesn't require app support. The app sees a giant pool of memory, the HBCC controller automagically stores it in the best possible location, probably using some kind of heuristics system to determine access frequency, among other things.

You might be thinking of the Raedon Pro SSG - the Polaris GPU with the Terabyte SSD on it. That requires app support - at least it did back when they announced it last year. I don't know if it's using Polaris anymore.

 

Well if that is the case, then it will all depend on how smart the controller actually is. Looking at all of the special cases that might occur from frame to frame... I'm a bit skeptical, but we'll see.

 

Realistically, 99.9% of the time everything is going to fit into HBM. The only time the HBCC management stuff is going to impact games is when it runs out of VRAM. Which is why in all the tests they've been showing for it they artificially cap the VRAM amount to 2GB.

For 8GB+ Vega cards in gaming, it's a cool feature but I don't think it's really going to change much. If they ever do 2/4GB HBM2 cards in the line up - it will probably be better there.

The real reason they have it though is compute/HPC purposes. Nvidia through CUDA has unified memory too but the developer has to manually manage it to some degree. Where as on Vega it can just let the controller do all of it.

HBCC is basically Nvidia's Turbocache in hardware form.

And it can write/read from more than just system memory.

 

Well that was exactly my point. They present HBCC so eagerly on DX12 games like DeusEx or ROTTR, while the real benefit of it should be in super or cloud computing. Unless I'm missing something, gamers will not benefit from it much. Maybe they are trying to justify the existence of future 6GB/4GB cards? I dunno...

 

Well in earlier posts I said I think if AMD had the money they'd split their compute cards off from their gaming cards like Nvidia does. But that requires a ton of developer/driver support, engineering and money to just spin a completely separate chip.

So I think when they were building Vega, they added HBCC because it's a really awesome workstation/compute feature - but they are also selling the same Vega to gamers, so they said "how can we take this tech and market it towards gaming in order to differentiate our product?" and this is what they came up with. Which is pretty cool.

Another feature of Vega, that's somewhat related to the gamer/hpc differentation, I don't hear people talking about much is the packed math support. Nvidia has this on their compute cards but not on their gaming ones. AMD has it on both and intends to build software gaming libraries that utilize it. They already said TressFX is going to use it and I'm sure basically all their vertex/particle sim stuff will eventually be ported to use it. It effectively doubles the performance of those workloads, which can potentially increase overall performance pretty drastically.

So you can have a scenario where you get TressFX running on multiple characters - turning it on impacts performance by 10-15% on Nvidia, but only 5-7% on AMD because Nvidia has no FP16 packed math on it's GTX cards.

The lesser precision shouldn't really impact those type of vertex/particle effects because they're basically somewhat random anyway.

 

Actually due to HBCC taking data in smaller chunks it mainly saves Memory Bandwidth.
So there is plain benefit for APUs. Secondly, Raja mentioned something bit weird where I am not sure how exactly he meant it.

That should not be forgotten in APU based notebooks. In many scenarios it will make AMD's mobile chips even more attractive than they are now to those who know current small difference to intel's offering.

And It would be crazy cool if APU finally had 2 Stacks of HBM2 (8GB) and therefore ~480GB/s access to unified system memory.

 

And maby it can be Turned On for HBM Fiji GPUs :mhp: (Maby as a Tweak or in ReLive)

 

I somehow believe it's really in the hardware this time Onna.

 

Maby i will go for VEGA but - Sept/December not sooner
And we shall see

WattMan is working for almost any GCN

 

https://www.reddit.com/r/Amd/comments/6co1rn/rx_vega_to_be_launched_a_couple_months_after_vega/

Seems like RX Vega isn't coming till August/September now.

 

Should have made a larger Polaris.

 

I think it would have spread amd thin and delayed vega, so far I think the biggest hurdle amd has faced is getting enough hbm2.

while vega may not be ready for mass market adoption, it will be ready for professional use however.

Those new vega based pro ssgs are likely to bring amd alot of much needed cash as it lacks any competition.

7nm vega with 1/2 double precision, will also likely be ahead or meet of what ever nvidia has in store to replace the new v100 accelerators.

I think is far more important for amd to have a real advantage rather than just barely getting by with another barely baked solution (fiji) that ends up not competing very well.

 

omg they're really having trouble with getting them hbm2 out

Maybe I should buy some polaris card while waiting for vega? I'd really like to play the new Prey

 

It's a shame they couldn't do HBM2 on the professional cards and DDR on the consumer version like what Nvidia is doing. At least that way we can have a Vega card before Xmas. The way it's going who knows when we will get consumer cards based on Vega.

Also I very much doubt we will get 7nm Vega by the first quarter of 2018. That is when Volta is supposed to launch from the looks of the rumors.

 

Is there a reason why they chose to partner with SK Hynix, and not Samsung? Are they hoping to build growth (for Hynix) and partnership with Hynix, and thus went with them?

 

No issues with them going HBM for the professional cards. But do consumers really need it.