AMD "Vega 10" Bears Core-Config Similarities to "Fiji"

[btarunr]

A Linux patch for AMD's GPU drivers reveals that its upcoming "Vega 10" graphics processor bears numeric core-configuration similarities to the "Fiji" silicon which drives the enthusiast-segment Radeon R9 Fury series graphics cards. The patch bears configuration values which tell the software how to utilize the resources on the GPU, by spelling them out. The entry "gfx.config.max_shader_engines = 4," for example, indicates that "Vega 10" features four shader engines, like "Fiji." Another entry "Adev-> gfx.config.max_cu_per_sh = 16" signifies the number of GCN compute units (CUs) per shader engine. Assuming the number of stream processors per CU hasn't changed from 64 in the "Vega" architecture, we're looking at a total stream processor count of 4,096. This could also put the TMU count at 256.

At earlier reveals of the "Vega 10" package, you notice a large, somewhat square GPU die neighboring two smaller rectangular memory stack dies, which together sit on a shiny structure, which is the silicon interposer. The presence of just two memory stack dies sparked speculation that "Vega 10" features a narrower 2048-bit memory interface compared to the 4096-bit of "Fiji," but since the memory itself is newer-generation HBM2, which ticks at higher clocks, AMD could run them at double the memory clock as "Fiji" to arrive at the same 512 GB/s bandwidth. The 4,096 stream processors of "Vega 10" are two generations ahead of the ones on "Fiji," which together with 14 nm process-level improvements, could run at much higher GPU clocks, making AMD get back into the high-end graphics segment.



View at TechPowerUp Main Site

 

[Nokiron]

The presence of just two memory stack dies sparked speculation that "Vega 10" features a narrower 2048-bit memory interface compared to the 4096-bit of "Fiji," but since the memory itself is newer-generation HBM2, which ticks at higher clocks, AMD could run them at double the memory clock as "Fiji" to arrive at the same bandwidth 512 GB/s, which remains unmatched by any NVIDIA consumer graphics GPU.

It might not be strictly "consumer" graphics GPU, but i'm pretty darn sure Titan Xp has a bandwidth of 547.7 GB/s.

 

[Slizzo]

It might not be strictly "consumer" graphics GPU, but i'm pretty darn sure Titan Xp has a bandwidth of 547.7 GB/s.

Yup, it sure does.

https://www.nvidia.com/en-us/geforce/products/10series/titan-xp/

Go there and click on "Specs" then "Full Specs", it's right there, in reverse black and white. 11.4Gbps on a 384-bit wide bus, equals 547.7GB/s

 

[ShurikN]

Well of course it bears similarities, it's still GCN. The biggest thing about Vega is in the new hardware surrounding the cores.

 

[alucasa]

I am waiting for "leaks" that say Vega is GPU.

 

[ppn]

The full bandwidth can never be used. With Cas latency of 20 and 64 byte bursts Titan has 384 GBS at best. It depends on whatever latency and burst the HBM2 has.

 

[xkm1948]

Hopefully it will be a good GPU. 1080 performance in DX11 and >1080 in DX12/Vulkan is all I am hoping for.

 

[G33k2Fr34k]

Vega10 is rumored to have roughly 50% higher GPU frequency over Fiji. If AMD has managed to make decent improvements in IPC and throughput with Vega, then Vega10 XT should hit 2x the performance of Fiji XT, which is where the Titan XP card is at.

 

[Captain_Tom]

Vega10 is rumored to have roughly 50% higher GPU frequency over Fiji. If AMD has managed to make decent improvements in IPC and throughput with Vega, then Vega10 XT should hit 2x the performance of Fiji XT, which is where the Titan XP card is at.

If it actually doubled the Fury X's framerate, Vega would handily beat Titan XP by 10%+. It all depends on if 1) they can hit 1550 - 1600 MHz, and 2) if things 2xGeometry IPC/HBC/Tiled rendering really post numbers similar to what AMD has advertised.

 

[RejZoR]

Just because it has a same arrangement of shaders, TMU and ROP units it doesn't mean they are the same. It's also not just Fiji with higher clock and smaller node. Vega has a new scheduler, new pixel and vertex engine, tile based rasterizer, new more flexible compute unit, new memory controller...

That's about the same absurd as saying Volkswagen Polo with 1.0 engine is the same as Mitsubishi Lancer Evo X with a 2.0 Turbo engine, just because they both have 4 cylinders.

EDIT: Fixed from Volkswagen Up! to Polo because some people were annoyed

 

[P4-630]

IF and IF....

 

[Captain_Tom]

Just because it has a same arrangement of shaders, TMU and ROP units it doesn't mean they are the same. It's also not just Fiji with higher clock and smaller node. Vega has a new scheduler, new pixel and vertex engine, tile based rasterizer, new more flexible compute unit, new memory controller...

That's about the same absurd as saying Volkswagen Up! with 1.0 engine is the same as Mitsubishi Lancer Evo X with a 2.0 Turbo engine, just because they both have 4 cylinders.

God thank you. A rare voice of reason.

Conservative numbers put Vega as double the 480's stats with at least 20% higher IPC (Or a lot more if tiled rendering pans out for AMD like it did for Maxwell). They just better be able to clock it high enough by June!

 

[Cybrnook2002]

I am waiting for "leaks" that say Vega is GPU.

Some bread crumbs:

http://www.tweaktown.com/news/57323/radeon-rx-vega-barely-beats-gtx-1070-leaked-results/index.html

 

[RejZoR]

@Cybrnook2002
This is based on useless leak Guru3D posted before...

I think the most critical thing is how will AMD address the utilization of wide highway of shaders for lower resolutions. The new scheduler could address this, but isn't necessary. This is one of things that are unknowns and it's quite possible I won't be able to buy RX Vega because of this. If AMD will just rely on fact that most demanding users use 4K anyway, this makes RX Vega less desirable for me where I demand high framerate at 1080p to fully utilize 144Hz monitor. Meaning, if outcome is similar to R9 Fury X, it's likely I'll be forced to wait for Volta. I know, it's a rather rare specific thing, but still.

 

[the54thvoid]

Some bread crumbs:

http://www.tweaktown.com/news/57323/radeon-rx-vega-barely-beats-gtx-1070-leaked-results/index.html

Breadcrumbs taste more like poison. We know nothing about Vegas performance except that it did 60-70fps in Doom at 4k a few months ago.

Won't be long now till we can read reviews from random sources if TPU isn't getting one.

 

[Vayra86]

I could say 'told ya'.

But let's see where Vega lands

One big positive I can extract from this is the focus on higher clocks. I really hope they land at 1500 stock, that could put Vega in a pretty good place.

 

[G33k2Fr34k]

Some bread crumbs:

http://www.tweaktown.com/news/57323/radeon-rx-vega-barely-beats-gtx-1070-leaked-results/index.html

Is that a Vega11 or Vega10 based card? Vega 11 is the second Vega chip. We don't much about Vega11, but it's rumored to be the smaller Vega chip.

 

[TheGuruStud]

The full bandwidth can never be used. With Cas latency of 20 and 64 byte bursts Titan has 384 GBS at best. It depends on whatever latency and burst the HBM2 has.

Nvidia fanboys are still in denial about the supremacy of HBM (due to latency).

Don't worry, they'll praise nvidia for using it and say how it's so good when nvidia finally brings it to the consumer level.

 

[Vayra86]

Nvidia fanboys are still in denial about the supremacy of HBM (due to latency).

Don't worry, they'll praise nvidia for using it and say how it's so good when nvidia finally brings it to the consumer level.

Regardless of what VRAM you use, you're still limited by the capabilities of the GPU core. HBM or not is irrelevant, performance is what matters, and Nvidia's delta compression is a step ahead.

 

[TheGuruStud]

Regardless of what VRAM you use, you're still limited by the capabilities of the GPU core. HBM or not is irrelevant, performance is what matters, and Nvidia's delta compression is a step ahead.

Is that why nvidia released slightly faster ram models? Oops, RAM speed is limitation. They did it to milk, but the cards are faster.

 

[RejZoR]

The reason why AMD needs HBM2 even more than it did on Fury X is the HBC. I think Vega models with GDDR5 will suffer quite a bit.

 

[Nokiron]

Is that why nvidia released slightly faster ram models? Oops, RAM speed is limitation. They did it to milk, but the cards are faster.

Bandwidth is not an issue now, or in the foreseeable future. (for gaming)

Not until the GPU itself gets a lot faster.

 

[renz496]

Nvidia fanboys are still in denial about the supremacy of HBM (due to latency).

Don't worry, they'll praise nvidia for using it and say how it's so good when nvidia finally brings it to the consumer level.

fanboyism aside it still doesn't change the fact that using GDDR5/GDDR5X did not give the card any sort of disadvantage vs cards using HBM when it comes to gaming performance.

 

[TheGuruStud]

Bandwidth is not an issue now, or in the foreseeable future. (for gaming)

Not until the GPU itself gets a lot faster.

False. OCing ram on the 1080 gives framerate increases.

Synthetic bench lololololol

 

[TheGuruStud]

fanboyism aside it still doesn't change the fact that using GDDR5/GDDR5X did not give the card any sort of disadvantage vs cards using HBM when it comes to gaming performance.

You would need a Fury (or Tesla) with gddr5 to test...
And we're going to ignore how fury x perf doesn't tank when you GBs over on VRAM usage, eh?