AMD releases VEGA die shot

[Coaxialgamer]

I'm honestly hoping that hbm2 overclocks well . Would be nice if you could get 1ghz out of those modules.

[rn8686]

6 minutes ago, Taf the Ghost said:

1080 was almost exactly a year ago.  GPU generations are normally closer to 18 months. So Vega isn't late, that's just the consumer base being annoyed that AMD took this generation off in the high-end.

Still going to be another 2-3 months until it is released, even then it may be a few a months until it is widely available (If it isnt then immediately bought up to cryto mine with). Very doubtful its in a position to compete with Volta, it seems pretty late to me. 

[Taf the Ghost]

2 minutes ago, Agost said:

According to SK Hynix HBM2 size, a rough calculation gives me about 470 mm^2 for the core. However those might be Samsung HBM2, and therefore dimensions can differ a bit.

 

https://www.hardocp.com/news/2017/06/05/amd_rx_vega_radeon_frontier_die_shot_faked_by_marketing

 

It's a marketing shot, btw, not a real die picture.

[Agosto]

2 minutes ago, Taf the Ghost said:

https://www.hardocp.com/news/2017/06/05/amd_rx_vega_radeon_frontier_die_shot_faked_by_marketing

 

It's a marketing shot, btw, not a real die picture.

That's obvious that those aren't the actual parts of the die, they just slapped an image on an actual photo of the MCM though

[Princess Luna]

I do think that by the time Vega finally release to compete with the GTX 1080 the Volta GTX 2080 will be so close to release for higher performance and same pricing that it'll render Vega pointless to any one who's not mining.

[WereCat]

17 minutes ago, Coaxialgamer said:

I'm honestly hoping that hbm2 overclocks well . Would be nice if you could get 1ghz out of those modules.

If it will make any difference at all

[AlwaysFSX]

21 minutes ago, Taf the Ghost said:

1080 was almost exactly a year ago.  GPU generations are normally closer to 18 months. So Vega isn't late, that's just the consumer base being annoyed that AMD took this generation off in the high-end.

So you mean they're behind by a generation?

[Coaxialgamer]

Just now, WereCat said:

If it will make any difference at all

considering the amount of ROP's , more bandwith always helps.

[Taf the Ghost]

6 minutes ago, AlwaysFSX said:

So you mean they're behind by a generation?

There was no Big Polaris. That was a clear choice and not a lack of technology.  Otherwise, the 580 and 1060 wouldn't be trading blows just fine.

[AlwaysFSX]

1 minute ago, Taf the Ghost said:

There was no Big Polaris. That was a clear choice and not a lack of technology.  Otherwise, the 580 and 1060 wouldn't be trading blows just fine.

Vega is either competing with a generation old GPU or they decided (for no good reason) to ignore an entire market segment and skip to competing in a new generation.

 

Neither of those are good choices.

[Valentyn]

2 hours ago, AlwaysFSX said:

Vega is either competing with a generation old GPU or they decided (for no good reason) to ignore an entire market segment and skip to competing in a new generation.

 

Neither of those are good choices.

 

The "no good" reason is most likely down to money. AMD as a whole is worth less than NVIDIA's R&D budget, never mind Intels.

If Vega is there to actually compete with Volta, it might have been the best decision they could have made given their limited resources. Especially after the previous management by 2015 considered Discrete Consumer graphics to be dead. They firmly believed APUs and Consoles were the future.

Hence further cuts to the GPU R&D, although at least putting it into the CPU section paid off.

I really doubt Vega will blow away Pascal and match Volta, but I can see the reason they have this massive gap since Polaris being their money issues. They're still in debt by over a Billion after all.

 

Hopefully with their Zen architecture they'll finally break even, or get out of debt entire by the end of the year; and we can finally get the AMD of 2013 and prior back where they were actually competing on all GPU fronts.

[Humbug]

 

[Taf the Ghost]

Raja mentioned at the Financial Analyst day that they weren't even on the map in parts of the compute segment.  That's likely where they put the resources.  There's also the issue that Nvidia pretty much cuts down their computer card design to make consumer graphics.  It's clearly the more cost effective way to go.

[Doobeedoo]

Looking nice

[App4that]

I just made a mess of my keyboard. 

[cj09beira]

3 hours ago, Taf the Ghost said:

Raja mentioned at the Financial Analyst day that they weren't even on the map in parts of the compute segment.  That's likely where they put the resources.  There's also the issue that Nvidia pretty much cuts down their computer card design to make consumer graphics.  It's clearly the more cost effective way to go.

like always they did both, they remade their compute unit to be able to do both fp32 and double fp16, that alone makes them have twice the perf for AI and ML, for gaming they increased the geometry output, improved the rasterizer and doubled the hardware threads to allow better compute unit utilization.

it seems to me like they focused on both 

[Red Hardware]

6 hours ago, Taf the Ghost said:

Well, if Apple signed up to buy a lot of units for their products, which is a high-end SKU and not a massive seller, those supply issues would have gotten a lot worse.

Well , we should wait and see how things will go.

I'm happy that apple is choosing AMD because that helps them with $ and it will lead to more R&D so AMD will be more competitive

[TeoIzAwezome]

12 hours ago, Curufinwe_wins said:

Confirmed not an actual die shot.... Just marketing BS, which explains why the HBM marker isn't either known manufacturer.

 

https://www.hardocp.com/news/2017/06/05/amd_rx_vega_radeon_frontier_die_shot_faked_by_marketing

 

Would it matter? Still gives us tons of info, and it has to be based on Vega's actual architecture or AMD might have gotten themselves into a bit of gaps advertising trouble.

[cj09beira]

i am sure that it is a vega die shot with an high level architecture block diagram on top of it