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

Seems to me that, just like the Radeon Pro Duo (and APUs with fast GPUs and slow CPUs), the Radeon Vega FE is a product in search of a market.

The only advantage it seems to have is that double rate fp16 which is exclusive to GP100. As far as workstation graphics go, judging by their internal benchmarks, it is irrelevant and loses to the p4000.

I don't understand why some users in this thread are upset over the comparisons to P4000, I struggle how to see it is invalid... They are marketed towards the same audience, if you look at perf/W and perf/$ in that context it is far from impressive.

As for the max board power, this should be 375W for both versions, as they both have dual 8pin power. These are not overclocking oriented cards... This is a workstation product
 
Yes, it is important... for SOME users. For tasks where frame buffer size is not an over-riding consideration, absolute performance or price/performance are the primary consideration. To take a single paramenter that is of overriding importance in a subset of all use cases and declare it as the sole correct and proper classification is ridiculous.

That "single parameter" is the only hardware bullet-point that defines the price bracket and market for the card, with the other being driver and sales/technical support.
If you want to compare "absolute performance or price/performance" then what's ridiculous is to even mention Quadro and Frontier / Radeon Pro models. On those factors, the aforementioned Quadro P4000 pales in comparison to a similarly priced GTX 1080 Ti.



But, let's play your game to its logical conclusion:

"The cherrypicked comparison with Titan Xp nonsense. That's a card with 12GB VRAM. Whoever builds a system with the Vega Frontier is either an enthusiast with the money to spare or someone who actually needs 16GB VRAM. The market for these two cards doesn't overlap. When 12GB Vega Pro cards come out, then maybe we can make price/performance comparisons with the Titan Px."

That conclusion is anything but logical because there won't be any 12GB Vega. HBM2 is either 2-Hi, 4-Hi or 8-Hi, meaning a 2 stack solution like Vega 10 will only get 4, 8 or 16GB VRAM.
The Titan Xp is the closest Pascal card there is in comparison to Vega FE. The GP102 has a 384bit bus meaning the card can only have 6, 12 or 24GB of VRAM. nvidia would need to cut a third of the memory bus and ROPs to reach 16GB, but that would essentially end up in a GP104 with slightly higher compute throughput but lower fillrate.


Seems to me that, just like the Radeon Pro Duo (and APUs with fast GPUs and slow CPUs), the Radeon Vega FE is a product in search of a market.

Really? Some very knowledgeable people who probably know a lot more than you seem convinced the card will sell out for sure.

Ryan Smith said:
To date, AMD has not yet said anything further about the launch since last month’s Computex unveil, however it appears that either AMD is opting to quietly release the sure to sell out cards, or some of their retailers have jumped the gun, as listings for both models have begun to show up.

I guess we'll see if sales for the Vega FE are so weak that AMD will have to lower their price during summer, since it's a product in search of a market.




I don't understand why some users in this thread are upset over the comparisons to P4000, I struggle how to see it is invalid... They are marketed towards the same audience,

No one's upset, don't worry ;)
But if at this point you can't differentiate the audience between 8GB and 16GB workstation cards, it's only natural that you would struggle to see how it's an invalid comparison.

But at the same time, I have to wonder why you aren't preaching about how bad the Quadro P5000 is in perf/$ in comparison with the P4000. The difference in the SPECviewperf results is 15-50% but the difference in price is 300%.
Wow, what a terrible card that Quadro P5000 is...
 
Last edited by a moderator:
That "single parameter" is the only hardware bullet-point that defines the price bracket and market for the card, with the other being driver and sales/technical support.
If you want to compare "absolute performance or price/performance" then what's ridiculous is to even mention Quadro and Frontier / Radeon Pro models. On those factors, the aforementioned Quadro P4000 pales in comparison to a similarly priced GTX 1080 Ti.





That conclusion is anything but logical because there won't be any 12GB Vega. HBM2 is either 2-Hi, 4-Hi or 8-Hi, meaning a 2 stack solution like Vega 10 will only get 4, 8 or 16GB VRAM.
The Titan Xp is the closest Pascal card there is in comparison to Vega FE. The GP102 has a 384bit bus meaning the card can only have 6, 12 or 24GB of VRAM. nvidia would need to cut a third of the memory bus and ROPs to reach 16GB, but that would essentially end up in a GP104 with slightly higher compute throughput but lower fillrate.




Really? Some very knowledgeable people who probably know a lot more than you seem convinced the card will sell out for sure.



I guess we'll see if sales for the Vega FE are so weak that AMD will have to lower their price during summer, since it's a product in search of a market.






No one's upset, don't worry ;)
But if at this point you can't differentiate the audience between 8GB and 16GB workstation cards, it's only natural that you would struggle to see how it's an invalid comparison.

But at the same time, I have to wonder why you aren't preaching about how bad the Quadro P5000 is in perf/$ in comparison with the P4000. The difference in the SPECviewperf results is 15-50% but the difference in price is 300%.
Wow, what a terrible card that Quadro P5000 is...

300% difference in price eh?
tmp_1390-Screenshot_20170616-181009632609527.jpg
Looks to me like they're priced identically.


15% more performance?
quadro-spec9.png


Same price, far better performance, same FB size, less than 60% the power draw.

Its only natural that you struggle to see the validity in this comparison, you are making up figures to support a conclusion you reached well before looking at any actual data. It's only natural.
 
Last edited:
But at the same time, I have to wonder why you aren't preaching about how bad the Quadro P5000 is in perf/$ in comparison with the P4000.
Looks to me like they're priced identically.

Please do share two links (and not unreasonably large spammy screenshots if I may) that show an identical price between the P4000 and the P5000.
Because at this point I'm starting to think you're just trolling.


15% more performance?
You're right, according to that graphic the difference doesn't even start at 15%. 168.11/148.68 = 1,13
It's a 13% difference. Wow, what a terrible card the P5000 is in comparison to the P4000, right?
If only there was some critical difference in the hardware between those two cards. Like, I don't know, maybe VRAM amount or something.
 
I guess we'll see if sales for the Vega FE are so weak that AMD will have to lower their price during summer, since it's a product in search of a market.
It's definitely positioned very weird.

Raja was emphatic about not buying for a gaming. They publish SpecViewPerf results, which really only matter for people who care about Pro features like HW antialiased lines etc, but then compare it against a Titan Xp, which has all these features disabled. It's going to be awesome for machine learning, but that's all predicated on the availability of a sufficiently debugged library.

Of course, anything will sell out if volumes are low enough.
 
It's definitely positioned very weird.

Raja was emphatic about not buying for a gaming. They publish SpecViewPerf results, which really only matter for people who care about Pro features like HW antialiased lines etc, but then compare it against a Titan Xp, which has all these features disabled. It's going to be awesome for machine learning, but that's all predicated on the availability of a sufficiently debugged library.

Of course, anything will sell out if volumes are low enough.

Agree wholeheartedly, only makes sense for machine learning, at which point it is significantly more cost efficient than a Quadro GP100, but we are talking about raw FP16 throughput here and the stability/usability of AMD's libraries vs the CUDA family will play a determining role here
 
It's definitely positioned very weird.

Raja was emphatic about not buying for a gaming. They publish SpecViewPerf results, which really only matter for people who care about Pro features like HW antialiased lines etc, but then compare it against a Titan Xp, which has all these features disabled. It's going to be awesome for machine learning, but that's all predicated on the availability of a sufficiently debugged library.

Of course, anything will sell out if volumes are low enough.


Oh I'll agree the marketing and general info passed around Vega has been generally crappy and all-around disconnected. Just like telling consumers back in April to "wait for Computex" only to show a super vague and short Prey playthrough and say "wait for Siggraph".
It doesn't mean the anticipation for Vega is pretty high nonetheless and the Frontier Edition will sell out, regardless if it's for graphics workstations or machine learning.
IMO, the main problem with the message around Frontier Edition isn't that it's a product in search of a market, it's the other way around. It's a product seemingly catering to too many markets, it's too flexible..
Which in turn actually made me surprised the base version is coming well under $2000. AMD is putting it on so many fronts, I thought they would charge extra for its flexibility. Turns out it's simply going after Titan Xp.
 
Last edited by a moderator:
It's definitely positioned very weird.

Raja was emphatic about not buying for a gaming. They publish SpecViewPerf results, which really only matter for people who care about Pro features like HW antialiased lines etc, but then compare it against a Titan Xp, which has all these features disabled. It's going to be awesome for machine learning, but that's all predicated on the availability of a sufficiently debugged library.

Of course, anything will sell out if volumes are low enough.
They are targeting the same market as the Titan Xp. I don't know what market that is but that is why they are comparing the 2. The card is for people who need high compute performance but also some pro features, no idea who is in that market. They also want the card to be used for development of new applications that makes use of FP16 and other features that Vega offers. What makes you think the Vega FE card has all the pro features that specviewperf uses? Why would you compared vs a Quadro then? Does comparing a Titan to a Quadro make sense?
 
Oh I'll agree the marketing and general info passed around Vega has been generally crappy and all-around disconnected. Just like telling consumers back in April to "wait for Computex" only to show a super vague and short Prey playthrough and say "wait for Siggraph".
I'm tempted to make a negative inference from all this.
Is it that the data points marketing has brought out are disconnected, or that the positive data points they had available are isolated?
I don't think I'd blame marketing when they ask people to wait for date X, and then have nothing to communicate but a desire to wait longer when the deadline is reached. They wouldn't do that if they weren't given the expectation there would be something to offer by that time.

IMO, the main problem with the message around Frontier Edition isn't that it's a product in search of a market, it's the other way around. It's a product seemingly catering to too many markets, it's too flexible..
Which in turn actually made me surprised the base version is coming well under $2000. AMD is putting it on so many fronts, I thought they would charge extra for its flexibility.

Too flexible in the "jack of all trades, master of none" sense? It seems like there may be some strong elements to the product, but some significant demerits and inconsistencies that leave it less compelling once the product has to apply itself to a specific use case.

Turns out it's simply going after Titan Xp.
Perhaps it's going after the "AMD has been telling investors it's done screwing up and it is barely managing to avoid missing its timeline on Vega entirely" market.
 
What makes you think the Vega FE card has all the pro features that specviewperf uses?
It's marketed for pro graphics ... Titan Xp is marketed for games, and double as a cheap inferencing card for machine learning with a big 12GB memory. Honestly it feels like these are just excuses being made for the card. This ***is*** the AMD equivalent of the Quadro, and it performs like a GTX 1070 in quadro form for more than double the price.

I don't understand how this kind of marketing is even beneficial. Only an idiot can be willing to spend 2k$ on a workstation card and be fooled into thinking this performs well based on a comparison to a GP102 based card running on consumer drivers.
 
Regarding those numbers and positioning against Titan Xp - do we actually know towards what Vega Frontier Edition drivers are optimized for?
Are they the standard Radeon Pro drivers and if so, why isn't it Radeon Pro Vega? Or is it using gaming drivers? Instinct drivers?
For what I recall from how it's presented, they're targetting "Prosumers" who might go for Titans and those considering Radeon Instinct to test the platform on single or couple cards before ordering racks full of them, neither of which are supposed to actually compete against Quadros, which is left for Radeon Pro -models?
 
Regarding those numbers and positioning against Titan Xp - do we actually know towards what Vega Frontier Edition drivers are optimized for?
Are they the standard Radeon Pro drivers and if so, why isn't it Radeon Pro Vega? Or is it using gaming drivers? Instinct drivers?
For what I recall from how it's presented, they're targetting "Prosumers" who might go for Titans and those considering Radeon Instinct to test the platform on single or couple cards before ordering racks full of them, neither of which are supposed to actually compete against Quadros, which is left for Radeon Pro -models?

AMD 100-506061 Radeon Pro Vega Frontier Edition Graphics Card (Blower Model)

https://exxactcorp.com/index.php/product/prod_detail/2565

Can't get any clearer than this, conversely you could argue that it makes no damned sense to market a card for pro graphics and have it ship with gaming drivers.

Furthermore, if this were the case then we would have gaming benchmarks of Vega by the end of the month, and I am certain people will be claiming it is invalid because it is not using gaming drivers
 
It's marketed for pro graphics ... Titan Xp is marketed for games, and double as a cheap inferencing card for machine learning with a big 12GB memory. Honestly it feels like these are just excuses being made for the card. This ***is*** the AMD equivalent of the Quadro, and it performs like a GTX 1070 in quadro form for more than double the price.

I don't understand how this kind of marketing is even beneficial. Only an idiot can be willing to spend 2k$ on a workstation card and be fooled into thinking this performs well based on a comparison to a GP102 based card running on consumer drivers.
This is not the AMD Quadro, where did you get that idea? AMD pro card are still under the FirePro branding.
 
Regarding those numbers and positioning against Titan Xp - do we actually know towards what Vega Frontier Edition drivers are optimized for?
Are they the standard Radeon Pro drivers and if so, why isn't it Radeon Pro Vega? Or is it using gaming drivers? Instinct drivers?
For what I recall from how it's presented, they're targetting "Prosumers" who might go for Titans and those considering Radeon Instinct to test the platform on single or couple cards before ordering racks full of them, neither of which are supposed to actually compete against Quadros, which is left for Radeon Pro -models?
From the comparison between the Quadros and the Radeon Pro Duo, it does look like the Frontier Edition is a very large improvement. I think it would be very compelling if Vega could get Quadro-level performance without professional features enabled or optimized for, but the scenario of a Titan to Quadro gap with the current Frontier Edition's baseline seems optimistic. I thought the Pro cards had a subset of the workstation applications optimized for.
 
Actually, the numbers you are discussing, are the same given by AMD during the FE presentation: nothing more..
Actually there are more, SPECAPC Siemens and Cinebench, however these two are just random number generators, for example In SPECAPC Siemens 10, even an M2000(GM108?) is like 20% faster than Vega FE. Cinebench just tests OpenGL performance and all NV GPUs older and newer score exactly the same in this test. AMD is trying hard here.
Are they the standard Radeon Pro drivers and if so, why isn't it Radeon Pro Vega? Or is it using gaming drivers? Instinct drivers?
This is Raja's statement:
RK: The Frontier Edition was designed for a variety of use-cases like Machine Learning, real-time visualization, and game design. Can you play games on Frontier Edition? Yes, absolutely. It supports the RX driver and will deliver smooth 4K gaming. But because it is optimized for professional use cases (and priced accordingly), if gaming is your primary reason for buying a GPU, I’d suggest waiting just a little while longer for the lower-priced, gaming-optimized Radeon RX Vega graphics card
https://videocardz.com/69662/raja-koduri-explains-where-radeon-rx-vega-in-reddit-ama
 
People keep repeating this fallacy.

http://www.pcworld.com/article/3099...-radeon-pro-wx-series-to-replace-firepro.html

AMD is set to replace its venerable FirePro lineup with Radeon Pro and Radeon Pro WX GPUs.

Where did you get the idea that pro cards are still under the FirePro branding ?
I always thought they had both lines since they only recently replace some of the FirePros with Radeon Pros. Anyways, it looks like WX series are what they are actually marketing as replacement for the FirePros. So the non WX series are not comparable to Quadros.
 
Reply ban(s) have been issued in this thread for those who respond with insults. Insults do not contribute positively to the discussion and have no place on Beyond3d.
 
I'm tempted to make a negative inference from all this.
Is it that the data points marketing has brought out are disconnected, or that the positive data points they had available are isolated?
(...)
Too flexible in the "jack of all trades, master of none" sense? It seems like there may be some strong elements to the product, but some significant demerits and inconsistencies that leave it less compelling once the product has to apply itself to a specific use case.
(...)
Perhaps it's going after the "AMD has been telling investors it's done screwing up and it is barely managing to avoid missing its timeline on Vega entirely" market.

Perfectly possible, all of them. Not imminent IMO, but possible.
I'm in no way refusing the possibility that there are/were serious problems with Vega's execution.
There have been both negative (delays, vague demos) and positive (8-Hi stacks coming out of nowhere) surprises around Vega. How the products themselves are actually going to behave is still unknown to the public, as far as I can see.


I don't think I'd blame marketing when they ask people to wait for date X, and then have nothing to communicate but a desire to wait longer when the deadline is reached. They wouldn't do that if they weren't given the expectation there would be something to offer by that time.
Isn't it even worse if these consecutive wait another 2 months statements did not get somehow vetted by the marketing team? Then the blame is 100% on the management, not the marketing team.




It certainly is, if one actually reads what's in that link:

Developers, and creators working with design tools who want the superior performance and are fine with not having application certification. Those who wish to have certified app support can turn to the “Vega” versions of Radeon™ Pro WX coming Q3, 2017
 
Back
Top