AMD vega 10 information leak

[Lawliet93]

Some new information regarding the Vega 10 . THE performance should be up to 12TFLOPS of FP32.

 

The core should feature 4096 stream processors and 16GB of HBM2 Memory. 

 

The TDP should be about 225W.

 

http://www.overclock3d.net/news/gpu_displays/amd_vega_10_and_vega_20_information_leak/1

 

For what I know... the GPU clock should be around 1,460GHz which would indicate, that VEGA 10 is using an upgraded 14nm FF+ or even TSMCs 16nm  process

[TheRandomness]

Now make an R9 nano sized one and have the fury X revamp have over 5000 steam processors. 

[patrickjp93]

1.2*10^13 = clock rate x SP count * 2 (FMA)

1.2*10^13 / 8192 = 1.465*10^9 or 1465MHz. That's a solid improvement for close to the same TDP as the Fiji Fury X.

[Misanthrope]

First half of 2017 means Nvidia will  be ready with Volta anyway so color me unimpressed.

[Carclis]

It's hard to tell if they're actually talking about Vega 10 or 11 here since they are suggesting that Vega 11 is the smaller chip in the article.

[patrickjp93]

27 minutes ago, Carclis said:

It's hard to tell if they're actually talking about Vega 10 or 11 here since they are suggesting that Vega 11 is the smaller chip in the article.

There was a rumor that big Vega was 6144 SPs, so it would be small Vega IF this was true.

[Vode]

Can't wait. Will be upgrading to this for sure.

[Ashaira]

What was the titan xp at in terms of tflops? 10.8 or something?

[patrickjp93]

Just now, Ashaira said:

What was the titan xp at in terms of tflops? 10.8 or something?

3584 * 2 * 1.531*10^9 = 10.974*10^12, or ~11 TFlops.

[Ashaira]

Just now, patrickjp93 said:

3584 * 2 * 1.531*10^9 = 10.974*10^12, or ~11 TFlops.

so not bad. if this hold true i may finally be able to get 2 of them and finally finish my build.

unless volta comes out in like a month or 2 after these.......

godamit tech industry!!!!!!!

[patrickjp93]

2 minutes ago, Ashaira said:

so not bad. if this hold true i may finally be able to get 2 of them and finally finish my build.

unless volta comes out in like a month or 2 after these.......

godamit tech industry!!!!!!!

Don't forget Nvidia's liquid-cooled dual-Titan card at 2GHz per core. It will crush AMD. AMDead confirmed. 

[Ashaira]

2 minutes ago, patrickjp93 said:

Don't forget Nvidia's liquid-cooled dual-Titan card at 2GHz per core. It will crush AMD. AMDead confirmed. 

if it doesn't have hbm2 not interested.

[patrickjp93]

Just now, Ashaira said:

if it doesn't have hbm2 not interested.

HBM2 is pointless for the time being. GPUs aren't even bandwidth-starved hilariously enough. Even they're starting to become more choked by memory latency because they don't have a big enough cache to make up the difference in timings.

[Bouzoo]

Straight from the source:

Quote

It is also rumoured that AMD is also working on a 7nm version of Vega 10, codenamed Vega 20, which will support PCIe 4.0 and also have 64 compute units. 

Now this is something new, if it's true.

[Ashaira]

15 minutes ago, Bouzoo said:

Straight from the source:

Now this is something new, if it's true.

not really it was bound to happen ever since we heard about pcie 4.0

[Ashaira]

25 minutes ago, patrickjp93 said:

HBM2 is pointless for the time being. GPUs aren't even bandwidth-starved hilariously enough. Even they're starting to become more choked by memory latency because they don't have a big enough cache to make up the difference in timings.

i know but for my setup it should work better (3x4k + 4k aux)

[Carclis]

38 minutes ago, patrickjp93 said:

There was a rumor that big Vega was 6144 SPs, so it would be small Vega IF this was true.

That would certainly uphold the longstanding tradition of moar cores. 6144 SPs sounds a little too crazy to be true though, especially at 1460Mhz.

[Bouzoo]

10 minutes ago, Ashaira said:

not really it was bound to happen ever since we heard about pcie 4.0

I don't remember a single rumor/source saying we'll be getting 7nm with Vega. 

[Ashaira]

Just now, Bouzoo said:

I don't remember a single rumor/source saying we'll be getting 7nm with Vega. 

well considering that
1) pcie 4.0 is a ways off
2) even further from becoming mainstream

it's fair to say 7nm would not be far off from when pcie 4.0 becomes somewhat adopted.

[Phoenix721]

Vega 11 was supposed to be a BIG chip with 6144 I think? The successor to Fury? Anyone?
SMH, I really hope both Vega 10 and 11 (with the 6144) are the real deal. I don't see Vega 10 outperforming the 1080 with -just- 4096 cores.

[Carclis]

8 minutes ago, Phoenix721 said:

I don't see Vega 10 outperforming the 1080 with -just- 4096 cores.

It's very possible with a clock speed around 1400Mhz and especially so since the CUs will be at least 15% faster like they were on Polaris.

[The Benjamins]

The fact it will have HBM means we get to see more small PCB graphics cards, and I like that idea.

[Bouzoo]

3 hours ago, Ashaira said:

well considering that
1) pcie 4.0 is a ways off
2) even further from becoming mainstream

it's fair to say 7nm would not be far off from when pcie 4.0 becomes somewhat adopted.

I don't see a connection between those two and how it was bound to happen since we've heard about 4.0. Pcie revision has nothing to do with 7nm becoming available.

Considering 4.0 should be coming out in early 2017, it won't even have the time to become somewhat adopted, since 7nm cards will according to this start coming out few months after this, if they even take that long. 

[Bouzoo]

3 minutes ago, The Benjamins said:

The fact it will have HBM means we get to see more small PCB graphics cards, and I like that idea.

More smaller cards, yes, but I wouldn't hold my breath on too many of them. Fury X and Nano were small, but one had water cooling and the other one throttled like crazy, and the Fury (air cooled, your "normal card") was not any smaller than your regular non-HBM cards. 

[The Benjamins]

2 minutes ago, Bouzoo said:

More smaller cards, yes, but I wouldn't hold my breath on too many of them. Fury X and Nano were small, but one had water cooling and the other one throttled like crazy, and the Fury (air cooled, your "normal card") was not any smaller than your regular non-HBM cards. 

well I only plan on using GPU's in a custom loop so I dont get any of those draw backs. but I am still waiting on the come back for mini DTX MOBO so i can have 2 single slot watercooled GPU's in a "mITX" case.