Vega/Navi Rumors (Updated)

[n0x1ous]

Apparently optimization for "only" 4GB RAM has been a large part of Fury driver/software since it's release. I'm talking about memory management and pre-loading memory over low speed/latency interfaces like PCIs. Obviously AMD has used (or designed) this as a learning lesson.

Right and it holds up very well in a lot of cases, but there are some cases @ 4k where it crushes the minimums or the performance falls off a cliff.

 

[richaron]

Right and it holds up very well in a lot of cases, but there are some cases @ 4k where it crushes the minimums or the performance falls off a cliff.

No arguments here. You are correct and I try to remain in reality even if I appear to defend AMD.

 

[SpaceBeer]

That's why they made hardware to do that, instead of software (drivers)

 

[sm625]

Wow this has to be the third time I've posted this. I'm not sure you'll pick it up but ...

bla bla bla?

It's also the third time AMD has come up with smoke and mirrors tricks to hide the fact that their cards burn 30-60% more power per frame vs their competition. This sounds like smoke and mirrors. If it looks like a duck... I mean come on seriously how can you actually buy this? Because of some demo contrived by AMD? Like that demo where polaris was as efficient as a GTX1080? Haha look how that turned out. AMD's marketing jargon doesnt mean a thing until we actually see it produce real gains.

 

[n0x1ous]

If it has reference AIO I will probably be a day 1 buyer

 

[tential]

If it has reference AIO I will probably be a day 1 buyer

If it's better than the 980ti vs fury x definitely. If not... I'll probably whine for a couple weeks before buying it since I have no other option for 4k.

 

[richaron]

bla bla bla?

It's also the third time AMD has come up with smoke and mirrors tricks to hide the fact that their cards burn 30-60% more power per frame vs their competition. This sounds like smoke and mirrors. If it looks like a duck... I mean come on seriously how can you actually buy this? Because of some demo contrived by AMD? Like that demo where polaris was as efficient as a GTX1080? Haha look how that turned out. AMD's marketing jargon doesnt mean a thing until we actually see it produce real gains.

Lol yeah I'll bite. We were talking about HBCC and PCIe data.

HBCC and PCIe data: There we go I've said it 5 times now including this last redundant one. And your argument has moved onto some other tripe.

But I'm talking about HBCC and PCIe data and what I've said is true: Vega is superior.

 

[Glo.]

bla bla bla?

It's also the third time AMD has come up with smoke and mirrors tricks to hide the fact that their cards burn 30-60% more power per frame vs their competition. This sounds like smoke and mirrors. If it looks like a duck... I mean come on seriously how can you actually buy this? Because of some demo contrived by AMD? Like that demo where polaris was as efficient as a GTX1080? Haha look how that turned out. AMD's marketing jargon doesnt mean a thing until we actually see it produce real gains.

I genuinely suggest reading people which are in higher understanding of GPU architectures, than you are, because you are spawning BS, that makes you look like an "...", attached only to your view of the world. There is enough analysis done on Beyond3D forum, by people from game development. Do your research, then post.

 

[CatMerc]

This is so far off the mark that not even wccftech would run it as a story.

 

[zinfamous]

People still buy Nvidia when Radeon has more vram.

However, we can only hope that the succesful launch of Ryzen might change some peoples minds about their other products too. And I will be spreading the good word if the product delivers.

hah! that successful launch has actually been labeled a titanic disaster by the very same minds that would need to be changed.

 

[vissarix]

I genuinely suggest reading people which are in higher understanding of GPU architectures, than you are, because you are spawning BS, that makes you look like an "...", attached only to your view of the world. There is enough analysis done on Beyond3D forum, by people from game development. Do your research, then post.

LMAO, please tell us more since you're one of those guys that has ''higher understanding of GPU architectures''...
After reading your second quote it seems like you're the one spawning Epic BS...Without a single proof....

3072 GCN core chip with 4 GB of HBM2 with 512 GB/s bandwidth will be enough to play in 4K@60Hz in epic/very high settings.

 

[Malogeek]

bla bla bla?

It's also the third time AMD has come up with smoke and mirrors tricks to hide the fact that their cards burn 30-60% more power per frame vs their competition. This sounds like smoke and mirrors. If it looks like a duck... I mean come on seriously how can you actually buy this? Because of some demo contrived by AMD? Like that demo where polaris was as efficient as a GTX1080? Haha look how that turned out. AMD's marketing jargon doesnt mean a thing until we actually see it produce real gains.

Not buying into anything since it hasn't been released yet so there's little practical information to go on. This is about what they've designed and dedicated silicon to.

So don't tell us to not buy into it when at the same time you're completely dismissing it with no actual product to show us its capabilities either way. It's called keeping an open mind.

 

[IEC]

I'm skeptical of a 4GB + HBCC combination working well at 4K60. While it's true I can use a mix of ultra/high settings currently in almost all games in 4K60 on my Fury, it would be nice to have 8GB local HBM2.

The proof is in the pudding of course, so I'll reserve judgment until I see actual independent benchmarks.

 

[Glo.]

LMAO, please tell us more since you're one of those guys that has ''higher understanding of GPU architectures''...
After reading your second quote it seems like you're the one spawning Epic BS...Without a single proof....

Its not me who has higher understanding of GPU architectures, but game developers. I suggest reading their posts.

 

[ElFenix]

The level of discourse seems to be on a downward trend here. I suggest you all try to raise it a bit.

AT Moderator ElFenix

 

[Arachnotronic]

With reviewers seemingly having samples of Vega, it shouldn't be too long before we know everything we want to know about this card.

Should be interesting.

 

[Dribble]

If you don't have enough memory and the additional stuff is on the other side of a low bandwidth interface (PCIE) then there's nothing you can do. No hardware on the gpu can magically increase the PCIE speed, it is what it is. The hardware required is enough gpu memory, make sure you have that and everything works.

Personally I believe the rumours: performance = stock 1080 + 10%, 8GB HBM2, very large die. That's the top end part and it must be too expensive to make already (being as competes with a 1080 which can gain +20% performance with o/c and the new 11Ghz memory yet is a much smaller chip and uses cheap DDR5X).

There's no way a lower end 4GB HBM2 card can be financially viable as AMD hardly save anything on the hardware bill (you still need the HBM2 and 2 stacks).

 

[Head1985]

With reviewers seemingly having samples of Vega, it shouldn't be too long before we know everything we want to know about this card.

Should be interesting.

how do you know they have vega samples?

 

[Arachnotronic]

how do you know they have vega samples?

https://twitter.com/RyanSmithAT/status/854584751948038144

 

[Glo.]

I'm skeptical of a 4GB + HBCC combination working well at 4K60. While it's true I can use a mix of ultra/high settings currently in almost all games in 4K60 on my Fury, it would be nice to have 8GB local HBM2.

The proof is in the pudding of course, so I'll reserve judgment until I see actual independent benchmarks.

Let me get back on track:
https://forum.beyond3d.com/threads/...ssion-soc-tegra-x1.59968/page-14#post-1973875
Sebbbi from Beyond3D:
Just wanted to clarify that I meant AMD GCN2 (consoles) vs Nvidia's latest (Maxwell/Pascal). AMD PC GPUs have also improved since GCN2.

Improvements for general performance:
- GCN3 introduced delta color compression. Including ability to sample/load compressed textures without decompress step.
- GCN3 improved geometry tessellation performance
- GCN4 improved geometry performance in general (including fast strips, primitive discard, etc).
- GCN4 improved delta color compression.
- GCN4 added instruction prefetch (reduces pipeline latency, again helps with geom bottleneck).
- GCN4 improved async compute scheduling (GPU side)

GCN5 (Vega) adds these general performance improvements:
- L2 cache includes L2 ROP cache (L1 ROP caches under L2). Don't need to flush caches between pixel shader passes.
- Tiled rasterizer. Reduces overdraw, bandwidth and makes ROPs more efficient in general.
- Improved geometry pipeline (including proper load balancing, up to 2x higher peak throughput)
- General purpose memory paging system

(I didn't list features that don't bring performance improvements without programmer intervention)

All of these improvements mean that GCN5 should run general purpose pixel/vertex shader code much better than GCN2. GCN5 has most of the same tricks that are seen in modern Nvidia GPUs. There are nice compute improvements as well, but they need special programmer support (DPP, SDWA, FP16). We will see the real impact of these improvements when DX12 SM 6.0 becomes available. Doom is already using these features with Vulkan, resulting in nice gains.


So in order to get performance gains from FP16 you need to update the code of your application, same thing for Primitive Shaders(Programmable Geometry Pipeline). As for General Purpose Memory Paging System:
https://forums.macrumors.com/thread...her-apu-polaris-announcements.1975249/page-58

Part about The memory system:
General Purpose Memory Paging system works in a way that is pretty much revolutionary. You have 512 TB indexing available done on hardware level, for full HSA 2.0 Unified Memory compatibility without any software level abstraction. Lets take a theoretical approach, and lets say we have 3072 GCN core GPU, with 4 GB of HBM2 with 512 GB/s of bandwidth.

Current models of Memory System store both used, and unused data in memory of the GPU, because in general GPUs do not have enough horsepower to handle all of it in particular time. Vega changes this approach. Tile Based Rasterization, next generation Pixel Engine, that is connected to L2 cache, and massively improved geometry performance increase throughput of the GPU. What is important is feeding this GPU with data. GDDR5 memory cannot give enough bandwidth to feed those cores, with reasonable amounts of power consumed. Neither does GDDR5X. Titan Xp memory system consumes around 50W of power alone, and memory subsystem consumes at peak 75W of power, due to amount of memory controllers, but averages are lower due to both Memory Compression, and Tile Based Rasterization, and ROPs connected to L2 cache. HBM2 memory cubes - 8W, and whole memory subsystem will consume at peak 15W of power, and you still get the benefit of Tile Based Rasterization, ROPs connected to L2 cache, rather than memory controller, ETC.

What actually does this Memory format? Framebuffer compared to large amount of GDDR5/X memory is smaller, but the data are available immediately to the GPU, and larger portions can be executed in particular time. Think about this like Non-Volatile data stored in memory, and indexed data in the system(because the memory controller has access to data in the System RAM, SSD's, HDD's in your computer and even network storage) is volatile. You save even more memory power consumption because of unused data. The Framebuffer is small enough to not exceed PCIe bandwidth, so the data can be delivered where its needed, when it is needed. Its all done on hardware, without any software abstraction.

Im sure people can explain all of this much more clearly and in more detail.


In essence this works not only for games, but compute, rendering, professional applications, everything that can benefit from it.

 

[Head1985]

https://twitter.com/RyanSmithAT/status/854584751948038144

He didnt say anything about vega.

 

[Arachnotronic]

He didnt say anything about vega.

It's pretty obvious what he's working on.

 

[richaron]

No hardware on the gpu can magically increase the PCIE speed, it is what it is.

Yeah this is where many people have problems. I myself have problems. In fact I posted about this before.

In fact the hardware does increase "PCIE speed"

there is a significant software overhead for a GPU to access RAM over PCIe, which is part of what the HBCC is there to address; So your claims of [speed over PCIe] equality are off target as well.

Edit: Yeah pls ignore odd terminology. By "increase PCIE speed" I'm talking about HBCC PCIe data transfer vs conventional.

 

[Glo.]

Yeah this is where many people have problems. I myself have problems. In fact I posted about this before.

In fact the hardware does increase "PCIE speed"

The framebuffer of memory available to GPU at particular time is smaller, than you have bandwidth of PCIe 3.0 x16. With PCIe you get 16 GB/s of bandwidth, whereas amount of data, stored in GPU VRAM is limited to its capacity. So there is absolutely no problem if the size is smaller, than the bandwidth of the PCIe lane.

 

[Head1985]

It's pretty obvious what he's working on.

not really.If he and others reviewers have vega then launch should be like 2-7 days from now.Not gonna happen.