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

well I hope navi is early next year , they need something out soon

I don't really know, why people still hope for an early 2019 release of navi, when AMD stated that Navi will come at last after the launch of Epyc and Ryzen3. For me that's a clear statement that Computex 2019 is the earliest you could hope for an release of navi but i strongly doubt we will see it in H1. I expect Ryzen3 around Computex and sometime after that it'll be Navis turn.
 
Yeah, and I believe in "mid range" Navi first. No real high end before the end of 2019...
"Mid Range" is a moving target. And when the mid-range has been stagnant for about 3 years in mid-2019, it's actually a sprinting target.
Navi shouldn't perform any lower than a Vega 56 by the time it releases, even if it's a $200-250 GPU.


2.0/1.6=1.25

The 2080 is 30~40% faster than Vega 64. Would it be worth that amount of HBM2?

I said "north of" 2.0 GHz, and that would be for average clocks (where Vega 64 sits at 1525MHz), not boost. Given the similar number of execution units, I am counting with ~30-40% higher clocks on Vega 20 than on Vega 10, yes.
The 2080 is selling for $800 with a 540mm^2 chip. If AMD can get a card with the 340mm^2 Vega 20 selling for $600, it's already a win for consumers IMO.


Does Vega 20 support combination of deactivated memory channel(s) with fully enabled ROPs?

HBCC connects to ROPs through Infinity Fabric, so I'd say probably yes?

https://www.anandtech.com/show/11717/the-amd-radeon-rx-vega-64-and-56-review/2

Ryan Smith & Nate Oh said:
Connecting the memory controllers to the rest of the GPU – and the various fixed function blocks as well – is AMD’s Infinity Fabric.
 
Last edited by a moderator:
I played PTU 3.3 today a bit, performance is.... kinda great!

Gtx1070@1440p 50-60 in all places except high-detailed stations. I finally played few missions, made a trade run, and after getting accustomed to the controls and the flow of the gameplay, I can see why so many people are purchasing expensive ships. Even I got a sligtht temptation to upgrade to Aegis Titan.

Crashes are still common though, I'll wait until 3.35 [new planets!] to play more, but what I saw today was very promising.

A consumer Vega 20 could be pretty competitive if it clocks north of 2GHz and has 12GB of 2.4 Gbps HBM2.

I don't really know, why people still hope for an early 2019 release of navi, when AMD stated that Navi will come at last after the launch of Epyc and Ryzen3. For me that's a clear statement that Computex 2019 is the earliest you could hope for an release of navi but i strongly doubt we will see it in H1. I expect Ryzen3 around Computex and sometime after that it'll be Navis turn.

For me I want to see them competitive in the $500 price range again. I don't know what that is considered anymore
 
Looks to be the worldwide filing of an automatically generated compute shader used for culling, which has popped up from time to time since earlier in 2018.
http://www.freepatentsonline.com/y2018/0033184.html
The filing dates are from July 2016, to give their relative age to when they were available for public consumption. The described scheme also places the culling shader at the front of the graphics pipeline, where it is more similar to claims in a PS4-related method for using a compute shader to feed the rest of the standard vertex setup pipeline.
In comparison a somewhat newer primitive shader patent that more closely lines up with Vega, or what AMD promised for it, is http://www.freepatentsonline.com/20180082399.pdf.
Here, the culling is embedded in a merged-stage pipeline akin to what Vega's marketing and various driver changes outlined, although some of the internal particulars the patent claims to be possible are not readily observable from the outside.
 
Did they?

All I remember was a statement that there would be 7nm GPUs for gaming—well, duh—but without confirming or denying whether Vega 20 was one of those 7nm gaming GPUs. This sort of hints at Navi being the only 7nm gaming architecture (along with its replacement, of course), but it's not quite the same thing as stating outright that Vega 20 would be restricted to the professional market.

Besides, we could see another semi-pro Frontier Edition card, this time based on Vega 20, that could perform quite well in games, albeit at a prohibitive price. Then again, the Vega FE was released at a lower price than the 2080 Ti, so where's the line between a gaming card and a (semi-)pro one these days?

If you want my opinion, back when AMD first made this statement, they probably didn't know whether Vega 20 would be offered as a gaming SKU, because the competitive situation at the time of its release wasn't known; there were probably a bunch of unknowns about the final product and its production (performance, yields, capacity, etc.), and I'd wager that some of these things still aren't known very precisely. So my guess is that AMD will release Vega 20 to the pro market, and then evaluate whether it also makes sense as a gaming card—it could go either way.


That is what most people here are not doing..... is connecting the AMD dots...! DR Su has mentioned using new infinity fabric released this year... so where, or what would it be? I think it is the Vega21 for 7nm gaming.

I asked earlier why wouldn't AMD capitalize on their early 7nm relationship with TSMC and use some of their lithography from Vega20 (ie vega 2.0), to be respun as a vega21 chip that is much smaller and uses energy efficient GDDR6 memory..? Easy math suggest (if so), a 7nm Vega96 would still be smaller than a V64 and use about 250w. And possibly utilize 2Ghz, efficiently.

Remember...If the new Vega 20 7nm GPU was to be made on a 14nm LPP process, it would have measured at 720mm2, instead of 360mm2. And for reference, Vega10 (FE/56/64) is 484mm2.

So what if AMD released both a 12nm Polaris chip and a 7nm Vega chip late this year..? AMD would have an advantage across the sub-4k gaming market. (ie mainstream)


Read through this:
The industry is at a significant inflection point as the pace of Moore's Law slows while the demand for computing and graphics performance continues to grow. This trend is fueling significant shifts throughout the industry and creating new opportunities for companies that can successfully bring together architectural, packaging, system and software innovations with leading-edge process technologies. That is why at AMD we have invested heavily in our architecture and product roadmaps, while also making the strategic decision to bet big on the 7nm process node. While it is still too early to provide more details on the architectural and product advances we have in store with our next wave of products, it is the right time to provide more detail on the flexible foundry sourcing strategy we put in place several years ago.

AMD's next major milestone is the introduction of our upcoming 7nm product portfolio, including the initial products with our second generation "Zen 2" CPU core and our new "Navi" GPU architecture. We have already taped out multiple 7nm products at TSMC, including our first 7nm GPU planned to launch later this year and our first 7nm server CPU that we plan to launch in 2019. Our work with TSMC on their 7nm node has gone very well and we have seen excellent results from early silicon. To streamline our development and align our investments closely with each of our foundry partner's investments, today we are announcing we intend to focus the breadth of our 7nm product portfolio on TSMC's industry-leading 7nm process.

We also continue to have a broad partnership with GLOBALFOUNDRIES spanning multiple process nodes and technologies. We will leverage the additional investments GLOBALFOUNDRIES is making in their robust 14nm and 12nm technologies at their New York fab to support the ongoing ramp of our AMD Ryzen, AMD Radeon and AMD EPYC processors. We do not expect any changes to our product roadmaps as a result of these changes.
 
AMD PR stated this year (was it to Anand, methinks ? ) that we can be assured they will release at least a new (gaming) GPU (series) per year. They explicitly said this "new" GPU might be actually a refresh though (enter RX590).
This speaks volumes as to their resource allocation in the competitive discrete GPU market.

Also note how Vega was all AMD released for both gaming and professional and deep learning markets (however dimminutive these last 2 are) from mid 2017 up til almost the end of 2018.

The shift from this picture to an AMD that releases Polaris 30, Vega 20, "Vega 21", Navi all in a space of almost 12 months is too much. This would require a ramp up phase to which we've seen no evidence of it happening.
 
Remember...If the new Vega 20 7nm GPU was to be made on a 14nm LPP process, it would have measured at 720mm2, instead of 360mm2. And for reference, Vega10 (FE/56/64) is 484mm2..
* VEGA 10 die size is not the marketed 484mm2 but 510mm2.
* VEGA 10 has 2048b memory bus, VEGA 20 got 4 HBM2 stack - 4096b
* VEGA 10 has no external GMI links, unlike VEGA 20
* VEGA 10 does not support PCIe4
* VEGA 10 has 1/16 FP64 support, VEGA 20 1/2 FP64
* numerous structures doesn't scale linearly...
 
AMD PR stated this year (was it to Anand, methinks ? ) that we can be assured they will release at least a new (gaming) GPU (series) per year. They explicitly said this "new" GPU might be actually a refresh though (enter RX590).
This speaks volumes as to their resource allocation in the competitive discrete GPU market.

Also note how Vega was all AMD released for both gaming and professional and deep learning markets (however dimminutive these last 2 are) from mid 2017 up til almost the end of 2018.

The shift from this picture to an AMD that releases Polaris 30, Vega 20, "Vega 21", Navi all in a space of almost 12 months is too much. This would require a ramp up phase to which we've seen no evidence of it happening.
You make it sound like its so many different things and all requiring the same R&D at the same time but it doesn't look that way to me:

Polaris 30 ( if it exists) , nothing more then a minimal port to GF "12" nm, just like Ryzen2, minimal effort, small perf bump + faster memory. Could be worth it from a Wafer agreement perspective alone.
Vega 20 , has been in the works in one form or another for a very long time, i think most people expected for it to be released a year ago, probably design all done on 14nm and then moved to 7nm So not a from scratch 7nm design.
this Vega 21 , nothing more then a different HBM configuration ?
Navi, has been in the works for a long time the one real effort getting serious R&D at this stage

Also remember AMD R&D has increased significantly. I dont expect a million different 7nm designs at once, but i dont really see it that way either. It doesn't look like 7nm APU is a 2019 product, So thats 2 gpu's and 1 CPU on 7nm, one of those GPU's in 2018 (vega20)
 
Increasing Vega's clocks from the 1500 avg you've mentioned to 2000+ will require redesigning some bits of the chip anyway.
 
Increasing Vega's clocks from the 1500 avg you've mentioned to 2000+ will require redesigning some bits of the chip anyway.
Yes one would expect all the physicals to change but all the function/high level would have been done. That's still a significant reduction in required effort during the initial 7nm ramp up
 
AMD Vega 20 Spotted in Final Fantasy XV benchmark?
A graphics card listed with hardware device ID 66AF:C1 was spotted in the Final Fantasy XV benchmark. That could very well be the long-discussed Vega 20, which is pending a release late this year.
...
it could be fake as heck of course. One thing matches, 66AF:C1 belongs to the Vega 20 device ID of hardware identifiers.
index.php


https://www.guru3d.com/news-story/amd-vega-20-spotted-in-final-fantasy-xv-benchmark.html


 
* VEGA 10 die size is not the marketed 484mm2 but 510mm2.
* VEGA 10 has 2048b memory bus, VEGA 20 got 4 HBM2 stack - 4096b
* VEGA 10 has no external GMI links, unlike VEGA 20
* VEGA 10 does not support PCIe4
* VEGA 10 has 1/16 FP64 support, VEGA 20 1/2 FP64
* numerous structures doesn't scale linearly...

It's 486mm2. It is in AMD's white paper for Vega. https://radeon.com/_downloads/vega-whitepaper-11.6.17.pdf
The “Vega” 10 chip is fabricated using 14-nm LPP FinFET process technology, and it packs 12.5 billion transistors into a 486 mm silicon die.
 
It's maybe just a "testing" board / pre production stuff ? Plus drivers pro application oriented, and not gaming.

I wonder if Vega 20 will have working Primitive Shaders...
 
Increasing Vega's frequency to 2.0GHz level will require significant changes to the chip, otherwise the massive increase in power consumption will offset any power efficiency gains from 7nm. The question now is: is Vega 20 a die shrink, or a rework of Vega to accommodate better features. I am leaning toward the die shrink part. AMD wouldn't waste resources on a chip that will soon be replaced by Navi.
 
Increasing Vega's frequency to 2.0GHz level will require significant changes to the chip, otherwise the massive increase in power consumption will offset any power efficiency gains from 7nm.
According to TSMC, you either get 35% higher transistor performance at iso power or half the power consumption at iso frequency. 35% frequency improvements implies a 1.6/2.1 GHz base / boost frequency at the same power level. Though, I'd expect AMD to reign in the top boost frequency to lower the high power requirement we see in Vega 64, so around 1.6/1.9GHz for base/boost.

Cheers
 
Back
Top