AMD's Vega-based Cards to Reportedly Launch in May 2017 - Leak

[the54thvoid]

Branch, there's a reason why Volta was pushed back (and Pascal didn't exist on engineering slides until a year and a bit before release). Vulkan\DX12 caught Nvidia with their pants down, so Pascal ended up being a Maxwell+ arch to use as a stepping whilst Volta is rearchitected.

The slide below is from GTC2014

The slide below is from GTC 2015

In 2013, Volta was present but it's HMC design stalled somewhat.

In 2014 (that would be almost 3 years ago, Volta had disappeared from the road map.

In 2015, Volta reappeared and it seems very much for a late 2017 release, though it's far enough to be 2018.

So really, Vega, DX12 etc has got nothing to do with Volta. The memory arrangement affected his position.

Please bear in mind that the only reason Titan X is £1100 is because AMD have NOTHING to touch it with. Nvidia couldn't care less about DX12 and Vulkan for it's GFX cards - their own mid range GP104 (not GP102 and not GP100) is still top dog.

Nvidia by your own definition (a revamped Maxwell masquerading as Pascal) don't even need to try to stay King of Cards.

 

[jabbadap]

Maybe its coming under the RX 5XX name instead and there will not be an RX 490... Seems a bit odd either way as something is not lining up.

Well that depends how they define "Generation" on their devising structure.

Spoiler

Spoiler: Off topic

afaik Pascal only exists because of tsmcs 20nm process was crap. Original plan was Kepler&28nm, Maxwell@20nm and then Volta@16nm.

 

[cdawall]

Vendors always previews the products with the most favorable benchmarks. AMD chose to use DOOM, which is a AMD favoring game. They even said the benchmark of DOOM that this is how Vega 10 will perform, which means it's a GTX 1080 competitor, not a GTX 1080 Ti competitor. Vega 10 is also roughly the size of GP102, but considering Pascal is 80% more efficient than Polaris. Let's give AMD the benefit of the doubt and assume they manage to cut this advantage in half, which would mean the Vega 10 will roughly perform in the range of GP104.

Vega 10 was not at release clocks nor was it using a driver designed for Vega. What is favorable about that situation?

 

[efikkan]

Vega 10 was not at release clocks nor was it using a driver designed for Vega. What is favorable about that situation?

What? There is no way they will run a driver on Vega hardware if it was not designed for it. If you don't understand this then you'll need to learn what a driver is.
DOOM is a highly AMD favoring game, meaning that Vega will look better in DOOM than the average game.

 

[kruk]

What? There is no way they will run a driver on Vega hardware if it was not designed for it. If you don't understand this then you'll need to learn what a driver is.
DOOM is a highly AMD favoring game, meaning that Vega will look better in DOOM than the average game.

PCGamesHardware says otherwise (http://www.pcgameshardware.de/AMD-Radeon-Grafikkarte-255597/Specials/Vega-10-HBM2-GTX-1080-1215734/):

Zu beachten gilt jedoch, dass erstens kein Vega-optimierter Treiber genutzt wurde, sondern einfach ein Fiji-Treiber mit ein wenig zusätzlicher Debugging-Arbeit.

So Fiji driver + some tweaks.

 

[Aquinus]

So Fiji driver + some tweaks.

For those who can't use Google translate and aren't bilingual:

Note, however, that first, no Vega-optimized driver was used, but simply a Fiji driver with a little additional debugging work.

 

[efikkan]

So Fiji driver + some tweaks.

This is just typical journalists failing to use the appropriate technical terms.
It will use a driver derived from the previous drivers, it's not like they are writing one from scratch. There will be continuous tweaks in the coming months, but these benchmarks still represent the "best case" of what we can expect from Vega 10.

 

[theeldest]

This is just typical journalists failing to use the appropriate technical terms.
It will use a driver derived from the previous drivers, it's not like they are writing one from scratch. There will be continuous tweaks in the coming months, but these benchmarks still represent the "best case" of what we can expect from Vega 10.

Current best case. But that is different than the NOW + 4 Months best case.

 

[efikkan]

Current best case. But that is different than the NOW + 4 Months best case.

The driver they have now is the same they'll have in four months minus some tweaks. It will not change a lot, just minor tweaks here and there.

If this was a bad demonstration of Vega, they wouldn't have used it you know.

 

[kruk]

The driver they have now is the same they'll have in four months minus some tweaks. It will not change a lot, just minor tweaks here and there.

AMD usually increases performance in months following the release and you are saying that the card will be only slightly faster? If they are able to exceed 1080 by 10% now they need only 15% to catch up with Titan XP. Why do you think it's impossible? A month or two before the launch, Polaris 10 was rumored to have trouble achieving 850 MHz, to surprise of many it launched with 30+% higher clocks. Why do you think they won't be able to achieve better clocks in 4 months with Vega?

 

[efikkan]

AMD usually increases performance in months following the release and you are saying that the card will be only slightly faster? If they are able to exceed 1080 by 10% now they need only 15% to catch up with Titan XP. Why do you think it's impossible? A month or two before the launch, Polaris 10 was rumored to have trouble achieving 850 MHz, to surprise of many it launched with 30+% higher clocks. Why do you think they won't be able to achieve better clocks in 4 months with Vega?

This is just wild guesses. AMD wanted to showcase the performance of Vega, so this demo is close to the final product. If this was a underclocked sample they would most certainly emphasize that the "production model would be xx % better".

 

[theeldest]

AMD usually increases performance in months following the release and you are saying that the card will be only slightly faster? If they are able to exceed 1080 by 10% now they need only 15% to catch up with Titan XP. Why do you think it's impossible? A month or two before the launch, Polaris 10 was rumored to have trouble achieving 850 MHz, to surprise of many it launched with 30+% higher clocks. Why do you think they won't be able to achieve better clocks in 4 months with Vega?

First Zen demo: September '16 -> 3.0GHz
First Ryzen demo: December '16 -> 3.4GHz
Second Ryzen demo: January '16 -> 3.6GHz

So 20% clock increase in 4-5 months.

Precedent shows that getting the 10% - 15% increase in the next few months isn't unreasonable.

 

[kruk]

This is just wild guesses. AMD wanted to showcase the performance of Vega, so this demo is close to the final product. If this was a underclocked sample they would most certainly emphasize that the "production model would be xx % better".

I use past available data to extrapolate what could happen. AMD already got burned with the FuryX vs 980Ti situation, so if they are smart they will keep their cards hidden until 1080Ti releases.

 

[theeldest]

I use past available data to extrapolate what could happen. AMD already got burned with the FuryX vs 980Ti situation, so if they are smart they will keep their cards hidden until 1080Ti releases.

Agree. It would behoove AMD to discuss Vega, let nVidia release the 1080Ti, then realease VegaX. (though I'm not optimistic that this is actually happening)

 

[efikkan]

I use past available data to extrapolate what could happen. AMD already got burned with the FuryX vs 980Ti situation, so if they are smart they will keep their cards hidden until 1080Ti releases.

Believing the Vega 10 will compete with GP102 is pure fantacy, that would require AMD to regain all of Nvidia's advantage, and all of this in a single iteration.
Vega 10 is a GP104 competitor.

 

[theeldest]

Believing the Vega 10 will compete with GP102 is pure fantacy, that would require AMD to regain all of Nvidia's advantage, and all of this in a single iteration.
Vega 10 is a GP104 competitor.

Given that most of nVidia's advantage is a larger die, it seems pretty realistic that AMD's larger die would be more competitive.

TitanXp = 471mm
1080 = 314mm (33% smaller than TitanX but 21% slower)
1060 = 200mm (58% smaller than 1080 but 41% slower)
480 = 232mm (15% larger than 1060 but 10% slower)
TitanXm = 601mm
FuryX = 596mm (1% smaller but 5% slower)

Estimates for Vega are between 475mm and 525mm. (https://www.reddit.com/r/hardware/comments/5m6tfu/vega_die_size_475mm2/)

So at the same die efficiency Vega should be between 20% slower than the TitanX and 10% faster than the TitanX.

 

[Camm]

I would not be surprised with driver advances, take 480 for example, a 10% deficit against the 1060 in DX11 on release, and now is on an even-keel in DX11, and beats it in DX12. Theoretical tflop levels in AMD cards have always pointed to a card with better legs than what release performance is.

However, neither do I think this will 'destroy' Titan XP, and we'll likely have another case of superior in DX12, slower in DX11.

 

[cdawall]

What? There is no way they will run a driver on Vega hardware if it was not designed for it. If you don't understand this then you'll need to learn what a driver is.
DOOM is a highly AMD favoring game, meaning that Vega will look better in DOOM than the average game.

They actually said in the PR for it that they were using a modified fiji driver that was in no way shape form or fashion optimal.

This is just typical journalists failing to use the appropriate technical terms.
It will use a driver derived from the previous drivers, it's not like they are writing one from scratch. There will be continuous tweaks in the coming months, but these benchmarks still represent the "best case" of what we can expect from Vega 10.

Yes because the original driver for every video card is the best one. They are the most tweaked drivers out there and no one ever sees performance increases from updated drivers.

 

[Evo85]

 

[efikkan]

Yes because the original driver for every video card is the best one. They are the most tweaked drivers out there and no one ever sees performance increases from updated drivers.

That doesn't make any sense at all.
AMD will not make a new driver before the release of Vega. It will be a new release of the same driver, people commonly keep confusing this. The driver implementation for Vega was started a while ago, from now on they will just be doing minor tweaks.

 

[theeldest]

That doesn't make any sense at all.
AMD will not make a new driver before the release of Vega. It will be a new release of the same driver, people commonly keep confusing this. The driver implementation for Vega was started a while ago, from now on they will just be doing minor tweaks.

The point others are trying to make is that the initial testing occurs with basic changes to the existing driver. Just enough changes that the new architecture will run though new features will not be taken advantage of.

As the product is developed, those new features are added in pieces.

So depending on when a demo occurs, there are different hardware optimizations that may or may not be enabled in the driver being used at that point in time.

Yes, the vega driver development began at the same time as the chip development, but they are both being worked on today as well. Seeing another 10+ % increase before official launch is not out of the question and 10% is the difference from a 1080 and 1080Ti.

 

[cdawall]

That doesn't make any sense at all.
AMD will not make a new driver before the release of Vega. It will be a new release of the same driver, people commonly keep confusing this. The driver implementation for Vega was started a while ago, from now on they will just be doing minor tweaks.

Yes they will? There will be an updated AMD driver in 6 months.

 

[efikkan]

Yes they will? There will be an updated AMD driver in 6 months.

Try reading my post again.

 

[cdawall]

Try reading my post again.

What do you mean there will be a different coding for vega than fiji. One that is optimized for vega and not fiji.

 

[Th3pwn3r]

Try reading my post again.

Is there any chance you can stop posting? That'd be great.