AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

[Nebuchadnezzar]

They _are_ mass produced. It's not like 16/14 nm FinFET is an unkown variable at this point. Nor are the circuits on the Vega dies hand carved by specifically employed virgins-from-venus or something. We're talking about a selling product here, not some super-early pre-alpha stuff (which might have been valid for the prototype shown back in december in Sonoma).

You don't know when the silicon came out of the fab, and even in mass production there can be large variations in power. You need proper high volume to be able to have a proper distribution and a median that you can draw conclusions from. How many FE's is AMD selling? 1000's? 10000's? That's too low to be representative. You only have to look at Ryzen to see the actual variations that come out of a factory. This is why binning even exists.

 

[chavvdarrr]

The card is selling with 2*8-pin power connectors, so the power is as expected

 

[seahawk]

You don't know when the silicon came out of the fab, and even in mass production there can be large variations in power. You need proper high volume to be able to have a proper distribution and a median that you can draw conclusions from. How many FE's is AMD selling? 1000's? 10000's? That's too low to be representative. You only have to look at Ryzen to see the actual variations that come out of a factory. This is why binning even exists.

Remember the RX480 done right story, after one reviewer measured some low power consumption on one card and how the respin would close the gap in Perf/W to the 1060? Never happened.

I remember AMD saying it works without application support. And what should be that "hidden bottleneck"?.

I do not not know, maybe it requires a certain API to work without application support. Maybe the card is limited by some other factor, maybe the tiled rasterizer is much less effective as the NV version?

 

[Rootax]

Or it just doesn't work. The "true" reviews can't come soon enough.

 

[CarstenS]

How many FE's is AMD selling? 1000's? 10000's? That's too low to be representative.

10000's too low ...* ymmd*. You believe whatever it is you want to believe. Fine by me.

 

[jacozz]

Well, guess we just have to wait a month to for sure know the true gaming performance. Fingers crossed

 

[Gipsel]

I do not not know, maybe it requires a certain API to work without application support.

That sentence appears to be somewhat self-contradictory.
But of course, we don't yet know the efficiency of AMD's tiling approach.

 

[fehu]

You don't understand!
It's all a sophisticated ruse apt to make feel nVidia comfortable and then crush them while sleeping with acid firing Vegas! For the glory of the Amd empire!

 

[CarstenS]

Timespy graphics score from our early adopter klaudiusz:
7126 (that's a bit more than the YT guy)

In my (older!) testing, I've had 1080 at 7011, Fury X at 5067 and RX 470 (0,5× Vega) at 3436.

 

[DavidGraham]

Maybe @Rys could shine a light on the matter, just one question: is the Vega FE gaming driver so far gimped compared to the Vega RX driver coming a month later?

 

[Clukos]

Timespy graphics score from our early adopter klaudiusz:
7126 (that's a bit more than the YT guy)

In my (older!) testing, I've had 1080 at 7011, Fury X at 5067 and RX 470 (0,5× Vega) at 3436.

He forced more stable clocks via Wattman that's why it looks like that

 

[Rys]

It's not gimped (that would be completely ridiculous), but it is older.

 

[Ike Turner]

Timespy graphics score from our early adopter klaudiusz:
7126 (that's a bit more than the YT guy)

In my (older!) testing, I've had 1080 at 7011, Fury X at 5067 and RX 470 (0,5× Vega) at 3436.

My Fury X score:
#1329

 

[entity279]

10000's too low ...* ymmd*. You believe whatever it is you want to believe. Fine by me.

Obviously for a high end video card that addresses only a niche of the already small market, 10000 should not be considered 'low' volume

But the Vega chip overall will see much higher numbers, and I believe this was what Nebuchadnezzar was refering to

 

[seahawk]

That sentence appears to be somewhat self-contradictory.
But of course, we don't yet know the efficiency of AMD's tiling approach.

I do agree with you, but AMD is always could for a "you do not need to adjust the application to use this, if you are already using Vulkan, DX12 or whatever"

We all remember the converting the software from CUDA is easy....... *if you were following the guidelines to have a easy conversion process when coding in CUDA...."

 

[DavidGraham]

It's not gimped (that would be completely ridiculous), but it is older.

Naturally, of course. Last question I promise and I completely understand if you ignore it: From where we stand right now with current leaks, Should we expect any major change in the position of products when RX Vega arrives?

 

[Nebuchadnezzar]

10000's too low ...* ymmd*. You believe whatever it is you want to believe. Fine by me.

I've been doing bin tracking for a whole lot longer than anybody here so don't take me for a fool.
Vega is supposed to be 475mm² and at 100% yield that is 122 chips per wafer. Actual factory yields on ideal dies are probably in the 75% range. On a chip of this size the yields are probably closer to 50-60%. Even at those low yields all you need for 10000 units is a little over 20 wafers. That's not even a day in fab line.

There are two options:

• AMD is voltage binning the chips to get better yields, this means that power is suffering and to get a proper median distribution and representative data on power we need a chip that has been in line production for a couple of weeks already. The proof of this is Polaris as you've clearly seen the power difference between a day 1 card and a card produced a year later. It doesn't matter if 14nm is "known", the yield curve for a new chip is something that only goes up in time, irrespective of process maturity. You can safely bet that the FE cards have the earliest silicon out of the fab.
• AMD is eating the yields and throwing away bad chips or keeping them for an even worse SKU, this means the power we see would actually be representative and the architecture is just disappointing.

It's not gimped (that would be completely ridiculous), but it is older.

Thank you, so again, this would be option #1.

 

[Malo]

For the YouTube guy, he had the card in "workstation" mode for all the tests and only found out and switched to "gaming" mode near the end and ran one more test in Witcher 3 with vsync on. I have no idea what the difference is between modes but I don't think the difference in performance will be much.

 

[Gipsel]

I've been doing bin tracking for a whole lot longer than anybody here so don't take me for a fool.
Vega is supposed to be 475mm² and at 100% yield that is 24 chips per wafer.

On 5" wafers maybe. The common size is 12" (300mm) nowadays. That equals about 70,000mm² usable space per wafer .

 

[SpaceBeer]

I've been doing bin tracking for a whole lot longer than anybody here so don't take me for a fool.
Vega is supposed to be 475mm² and at 100% yield that is 24 chips per wafer. Actual factory yields on ideal dies are probably in the 75% range. On a chip of this size the yields are probably closer to 50-60%. Even at those low yields all you need for 10000 units is a little over 800 wafers. That is just a couple of days on a fab line.
.

Aren't those produced on 300mm wafers, which should give 100-110 chips (50-60 good ones, depending on yield)