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

Rootax said:
Why does hardware canuck said competition is back ?
Click to expand...
Multiple possibilities:
- AMD has found the right knobs in the drivers and Vega outperforms 1080 Ti
- AMD had an ace up their sleeve and RX Vega gets revised silicon that enables problematic functions from 1st revision in FE and/or allows them to substiantially boost clocks
- AMD has shown CLC-version with sustained 1600+ MHz core clock which is enough
- HWC's expectations are low
- HWC wants to generate interest for their upcoming articles
- Vegas pricing is lowered enough to shine against similarly priced competition, whichever card this turns out to be
- a combination of the above.
 
Last edited:
Let's just call it for what it is. Hype. Nothing more, nothing less.

Show me price and performance numbers. Give me a deep dive of the architecture and changes and spice it up with some "silicon porn" showing the bare chip in all its glory.

Frankly I don't care how pretty the fan shroud is or if it has RGB leds. That is simply out of the scope of interest for a site like Beyond3D.
 
CarstenS said:
Multiple possibilities:
- AMD has found the right knobs in the drivers and Vega outperforms 1080 Ti
- AMD had an ace up their sleeve and RX Vega gets revised silicon that enables problematic functions from 1st revision in FE and/or allows them to substiantially boost clocks
- AMD has shown CLC-version with sustained 1600+ MHz core clock which is enoug
- HWC's expectations are low
- HWC want's to generate interest for their upcoming articles
- Vegas pricing is lowered enough to shine against similarly priced competition, whichever card this turns out to be
- a combination of the above.
Click to expand...
Why would Vega need to outperform 1080 Ti or Titan X to be competitive? These are niche products. GTX 1070 and GTX 1080 are Nvidia's main products. Vega FE with newest drivers is already able to compete with GTX 1080. If RX is slightly faster that's a bonus. Current AMD's flagship RX 580 is competitive only at low end. It barely beats GTX 1060 (+5%). At minimum, Vega allows AMD to reach GTX 1070 and GTX 1080 market segments. That's a huge improvement over last gen.

Also you need to consider that Vega's drivers will improve further during its life time. It has more new tech inside than any GCN iteration.

For comparison, this is what tech spot said about RX 480 vs GTX 1060:
"It's interesting to look back at the RX 480 vs. GTX 1060 battle and see how things have evolved. Our day one GTX 1060 coverage back in July 2016 found the green team dominating, beating the RX 480 by a convincing 12% margin. About five months later, AMD's Crimson ReLive drivers drastically reduced those margins, leaving the GTX 1060 just 1% faster than the RX 480."

Update: If that $499 price above it legit and RX performs few percents better than FE, it should be highly competitive against GTX 1080. If it performs halfway between 1080 and 1080 Ti at that price point, it will be a huge success.
 
DavidGraham said:
Or it simply performs on par with the 1080.
Click to expand...

The way they hyped Vega, the power consumption, the fact that they are very late, it better be better than 1080. From what we know, it's not. And it's trouble given all the improvments they talked about (while in fact, Vega FE performs like a Fiji clock for clock, or a little worse).

Or yes, very low price, be I doubt amd can do that with hbm2 and the chip size.

I can't wait for real reviews so we know what's up...
 
If vega turns out to be competition for 1080 (not ti) it is pretty much not great for amd. 1080 is almost 200mm2 smaller chip with cheaper memory. One of the products is going to make bigger profit and the other one is going to struggle to make money?
 
sebbbi said:
Vega FE with newest drivers is already able to compete with GTX 1080.
Click to expand...
Only the water cooled version though. Air still lags behind alot. There is also the power consumption problem, it's too high for a 1080 level of performance.

Edit: clocks

  • AMD Radeon RX Vega 64 Liquid: base 1406 MHz / boost 1677 MHz
  • AMD Radeon RX Vega 64 Limited: base 1247 MHz / boost 1546 MHz
  • AMD Radeon RX Vega 64: base 1247 MHz / boost 1546 MHz
Liquid is slightly faster than FE (77 MHz), this means it will definitely be faster than 1080.
 
Last edited:
sebbbi said:
Why would Vega need to outperform 1080 Ti or Titan X to be competitive? These are niche products. GTX 1070 and GTX 1080 are Nvidia's main products. Vega FE with newest drivers is already able to compete with GTX 1080. If RX is slightly faster that's a bonus. Current AMD's flagship RX 580 is competitive only at low end. It barely beats GTX 1060 (+5%). At minimum, Vega allows AMD to reach GTX 1070 and GTX 1080 market segments. That's a huge improvement over last gen.
Click to expand...

Because if your ~484mm² product with HBM2 memory and all the new architectural bells and whistles enabled barely able to outperform last year's mid-range models at +50% power consumption does not mean you're back, but that your back's against the wall in every conceivable manner. IMHO.
 
Last edited:
CarstenS said:
Because if your ~484mm² product with HBM2 memory and all the new architectural bells and a whistles enabled barely able to outperform last year's mid-range models at +50% power consumption does not mean you're back, but that your back's against the wall in every conceivable manner. IMHO.
Click to expand...
Well, regardles of that fact, 484 mm² is still less than 596 mm², two HBM stacks are less than four, interposer is also smaller - and expected performance is at least 50 % higher. If the manufacturing of Fury X with current selling price close to $599 is profitable for AMD, then top Vega (which is also $599), should increase margins (because of lower manufacturing costs) and increase sales (because >50 % higher performance for the same price). Comparing to competition, those specifications don't look very well, but it seems to be quite good replacement of Fiji/Fury.
 
Surely there's something to be said for the additional functionality and flexibility introduced. It can't all be about performance. Right?
 
CarstenS said:
Because if your ~484mm² product with HBM2 memory and all the new architectural bells and a whistles enabled barely able to outperform last year's mid-range models at +50% power consumption does not mean you're back, but that your back's against the wall in every conceivable manner. IMHO.
Click to expand...
That's not a fair comment. Architectural "bells and whistles" does not come for free. Vega support several things Pascal does not (Primitive shadres, full tier 3 DX12 support, some of them could give it an edge in the future) and it is a chip meant to not only support gaming, but compute first. So size is a down for AMD, but it is not unjustified. Power draw is high, and that's to me the main issue to be addressed in the newer iterations.
 
AMD teases Radeon RX Vega exclusively with Linus – press unhappy

We have noticed that many members of the press aren’t that happy about the exclusive given to Linus with journalists such as Cameron Wilmot from Tweaktown posting ‘feeling frustrated’ on his own Facebook page about the event. There are many other well known people in the post chipping in to share their views. We noticed key industry people including Scan UK and Nvidia staff members such as Bryan Del Rizzo making some points in Cameron’s post.

Cameron uploaded an image which says ‘AMD, you gave exclusive content coverage on new products to Linus. That’s like a reverse sellout or something, and totally uncool’.

We do see the AMD shift towards social media influencers in the last year or so rather than traditional tech websites, but Cameron makes the point ‘Maybe for children, but many still want detailed content with a mix of short stuff.’

Well known global tech websites such as Tweaktown have had key staff flown half way around the world to these events so its obvious that some ill feeling is likely to creep in. From our side (and I checked with my boss), KitGuru already knew AMD had a very strong relationship with Linus and he has such a huge following on Youtube they obviously feel he needs some special treatment.

Playing Devils Advocate however- its a rather strange move from AMD as key members of the press will likely feel that being left out of this almost negates the purpose of a ‘unified launch’ for some of the most important AMD products in the last decade.
Click to expand...

https://www.kitguru.net/components/...rx-vega-exclusively-with-linus-press-unhappy/
 
no-X said:
Well, regardles of that fact, 484 mm² is still less than 596 mm², two HBM stacks are less than four, interposer is also smaller - and expected performance is at least 50 % higher. If the manufacturing of Fury X with current selling price close to $599 is profitable for AMD, then top Vega (which is also $599), should increase margins (because of lower manufacturing costs) and increase sales (because >50 % higher performance for the same price). Comparing to competition, those specifications don't look very well, but it seems to be quite good replacement of Fiji/Fury.
Click to expand...
50%? FE tests show it in the 20-30% range, apart from a few occasions where the Fury X runs out of VRAM.
14nm design and manufacture costs are a lot higher.
VRM circuitry for a 375W card will be expensive.
And it even loses the board form-factor niche Fury X & R9 Nano had.
 
no-X said:
Well, regardles of that fact, 484 mm² is still less than 596 mm², two HBM stacks are less than four, interposer is also smaller - and expected performance is at least 50 % higher. If the manufacturing of Fury X with current selling price close to $599 is profitable for AMD, then top Vega (which is also $599), should increase margins (because of lower manufacturing costs) and increase sales (because >50 % higher performance for the same price). Comparing to competition, those specifications don't look very well, but it seems to be quite good replacement of Fiji/Fury.
Click to expand...
Fiji is on 28 nm, Vega on 14, correct? And I for one do not know the pricing difference between HBM gen1 and 2. What I do know is, though, there seem to more potential customers for gen 2.

leoneazzurro said:
That's not a fair comment. Architectural "bells and whistles" does not come for free. Vega support several things Pascal does not (Primitive shadres, full tier 3 DX12 support, some of them could give it an edge in the future) and it is a chip meant to not only support gaming, but compute first. So size is a down for AMD, but it is not unjustified. Power draw is high, and that's to me the main issue to be addressed in the newer iterations.
Click to expand...
Is it not? If the the non-free (which is of course true) bells and whistles fail to make an impact in one of the more important markets (gaming), then marketing is going to have a hard time to convince people that it will do so in a non-descript future. If they cannot get them to gain traction, then those bells and whistles are not chosen wisely. Integrated graphics can do most of D3D12 Tier3-collection, so yeah, I'll deduct 5 mm² from the die space for good will. The 300mm wafers surely will be unimpressed by that, though.

Please read with big and bolded IFs.
 
kalelovil said:
50%? FE tests show it in the 20-30% range, apart from a few occasions where the Fury X runs out of VRAM.
14nm design and manufacture costs are a lot higher.
VRM circuitry for a 375W card will be expensive.
And it even loses the board form-factor niche Fury X & R9 Nano had.
Click to expand...
It's expected, that top Vega isn't slower than reference GTX 1080, which is 46 % faster than Fury-X (source):
furyxr3ulz.png


PCB components are one of the cheapest part of graphics card desing. GPU and memory are the most expensive ones. Yes, 14nm process is more expensive than 28nm, but not „a lot“. Please consider, that both AMD and Nvidia replaced their 28nm products by 14/16nm products, so they had to be cheaper to manufacture more than 12 months ago. It would be absurd to expect, that all the 28->14/16nm replacements were viable already in Q2 2016 but Vega shouldn't be even in Q3 2017.
 
CarstenS said:
Fiji is on 28 nm, Vega on 14, correct? And I for one do not know the pricing difference between HBM gen1 and 2. What I do know is, though, there seem to more potential customers for gen 2.


Is it not? If the the non-free (which is of course true) bells and whistles fail to make an impact in one of the more important markets (gaming), then marketing is going to have a hard time to convince people that it will do so in a non-descript future. If they cannot get them to gain traction, then those bells and whistles are not chosen wisely. Integrated graphics can do most of D3D12 Tier3-collection, so yeah, I'll deduct 5 mm² from the die space for good will. The 300mm wafers surely will be unimpressed by that, though.

Please read with big and bolded IFs.
Click to expand...

I agree that features must be used to be useful. And IF they are used and AMD gains traction there, then that could be a wise decision buy their part. So far, it's difficult to judge. But as I told before, this is a chip meant for covering multiple markets (compute too). And this also accounts for space loss, too. About that "5%", neither me or you can be sure about how much die size it is needed for that: integrated graphis has a low ALU count, their absolute cost in die size is probably lower.
 
no-X said:
It's expected, that top Vega isn't slower than reference GTX 1080, which is 46 % faster than Fury-X (source):
furyxr3ulz.png


PCB components are one of the cheapest part of graphics card desing. GPU and memory are the most expensive ones. Yes, 14nm process is more expensive than 28nm, but not „a lot“. Please consider, that both AMD and Nvidia replaced their 28nm products by 14/16nm products, so they had to be cheaper to manufacture more than 12 months ago. It would be absurd to expect, that all the 28->14/16nm replacements were viable already in Q2 2016 but Vega shouldn't be even in Q3 2017.
Click to expand...
Not that it detracts much from your point, but total fixed costs (often somewhat sloppily referred to as design cost) is increasing with finer lithographic nodes. This favours volume parts, so it is no greater matter for the most popular GPUs. For something like Vega though, I don't know enough about either projected or real sales volumes to know how strongly design cost vs. sales volume factor into its viability. Process and tools improve with time, making a later introduction of a product a bit cheaper, but narrows its market window as well.
Did Fiji ever produce a net profit for AMD? Will Vega? Its main value may be as an architectural test vehicle for later products across the scale including x86 APUs, rather than being a particularly viable product in and of itself.
 
You must log in or register to reply here.
Back
Top