[News] AMD Vega 10, Vega 20, and Vega 11 GPUs Detailed

[StyM]

https://www.techpowerup.com/226012/a...-gpus-detailed

AMD CTO, speaking at an investors event organized by Deutsche Bank, recently announced that the company's next-generation "Vega" GPUs, its first high-end parts in close to two years, will be launched in the first half of 2017. AMD is said to have made significant performance/Watt refinements with Vega, over its current "Polaris" architecture. VideoCardz posted probable specs of three parts based on the architecture.

AMD will begin the "Vega" architecture lineup with the Vega 10, an upper-performance segment part designed to disrupt NVIDIA's high-end lineup, with a performance positioning somewhere between the GP104 and GP102. This chip is expected to be endowed with 4,096 stream processors, with up to 24 TFLOP/s 16-bit (half-precision) floating point performance. It will feature 8-16 GB of HBM2 memory with up to 512 GB/s memory bandwidth. AMD is looking at typical board power (TBP) ratings around 225W.

Next up, is "Vega 20." This is one part we've never heard of today, and it's likely scheduled for much later. "Vega 20" is a die-shrink of Vega 10 to the 7 nm GF9 process being developed by GlobalFoundries. It will feature 4,096 stream processors, too, but likely at higher clocks, up to 32 GB of HBM2 memory running full-cylinders at 1 TB/s, PCI-Express gen 4.0 bus support, and a typical board power of 150W.

The "Vega 11" part is a mid-range chip designed to replace "Polaris 10" from the product-stack, and offer slightly higher performance at vastly better performance/Watt. AMD is expecting to roll out the "Navi" architecture some time in 2019, and so AMD will hold out for the next two years with "Vega." There's even talk of a dual-GPU "Vega" product featuring a pair of Vega 10 ASICs.

[AliG]

Honestly, given the timing, I think fitting the 1080 and Titan X is pretty disappointing. Especially given that P100 will be coming out around that same time...

I'm all for competition; I just wish AMD would have really swung for the fences with Vega.

[StAndrew]

HBM2 is probably holding them up? Should release Vega 10 now, Vega 11 with HBM2 later.

[AliG]

HBM2 is probably holding them up? Should release Vega 10 now, Vega 11 with HBM2 later.

That's my thought. Why waste valuable time to market waiting for a technology that honestly won't make a difference? Or even better yet, do Vega 10 Lite with GDDR5X and then update to HBM2.

I just don't understand the decision making here. Waiting this long inherently let Nvidia saturate 2016's high end market, and their answer doesn't even compete with Nvidia's flagship.

[StAndrew]

That's my thought. Why waste valuable time to market waiting for a technology that honestly won't make a difference? Or even better yet, do Vega 10 Lite with GDDR5X and then update to HBM2.

I just don't understand the decision making here. Waiting this long inherently let Nvidia saturate 2016's high end market, and their answer doesn't even compete with Nvidia's flagship.

I'm not to familiar with the scaling of games and GPU memory bandwidth, specifically IRT 4K resolution. Will HBM2 really make that much of a difference with 4K gaming? My thoughts (very speculator) is the Vega 10 doesn't compete with Nvidia's GP104/102 and they are hoping HBM2 will help prop up performance (or prop up sales with an "exclusive" technology).

Its also kind of odd they would mention Vega 11 before Vega 10 is even released; they don't seem to have much faith in Vega 10...

One more thought; maybe they don't have much choice. I'm going to assume silicon will have to be revised to accept GDDR5 vice HBM2. Maybe they gambled on HBM2 yields being available sooner.

[Tim]

First half of 2017? Unless I'm missing something, that's really not going to be good enough. Q1, maybe, H1, not so much as Volta is now getting there pretty quickly and nvidia has been sitting pretty with a mid range GP102 chip and probably has the aces still firmly tucked up their sleeves. At this rate AMD could lag one architecture behind. Unless nvidia would make a catastrophic mistake with Volta, this could be a tricky one for AMD.

[AliG]

First half of 2017? Unless I'm missing something, that's really not going to be good enough. Q1, maybe, H1, not so much as Volta is now getting there pretty quickly and nvidia has been sitting pretty with a mid range GP102 chip and probably has the aces still firmly tucked up their sleeves. At this rate AMD could lag one architecture behind. Unless nvidia would make a catastrophic mistake with Volta, this could be a tricky one for AMD.

That's pretty much my point. I just don't see how AMD's GPU team thought this timing was acceptable.

[zanzabar]

That's pretty much my point. I just don't see how AMD's GPU team thought this timing was acceptable.

hmb2 large shipments from hynix, samsung, and micron are scheduled for h1. amd messed up and put all of their eggs in the hmb2 being ahead of schedule basket.

[vario]

So, the first vega card will have same shader amount and same bandwith as fury x ? At almost the same power consumption.Yes i know, there will be architectural differences so it will be faster, clocks probably will be higher also.But still, somehow doesnt seem right to me, 2 years after fury.
So, it will compete with 1080.While nvidia will release volta ...

[zanzabar]

So, the first vega card will have same shader amount and same bandwith as fury x ? At almost the same power consumption.Yes i know, there will be architectural differences so it will be faster, clocks probably will be higher also.But still, somehow doesnt seem right to me, 2 years after fury.
So, it will compete with 1080.While nvidia will release volta ...

fury ran really hot, was around 300W if you wanted it to keep its clocks under load, and did not clock very well. nvidia did not add much for shader count but they have much higher clock speeds so i would not count it out, but it does not look promising if it is like pascal.

[vario]

fury ran really hot, was around 300W if you wanted it to keep its clocks under load, and did not clock very well. nvidia did not add much for shader count but they have much higher clock speeds so i would not count it out, but it does not look promising if it is like pascal.

Yeah, i just browsed some fury X reviews, power consumption at load is all over the place friom 250 to 300.However if AMD states 275W, then we can assume same metrics apply to vega ,so it could also consume more power, thats 50W diff.
Fury X GCN 1.2 1050 Core clock, polaris on the same process as vega has 1120,Mem bandwith will be the same but with addition of compression.I guess nobody estimated gains on polaris due to the newer architecture, but they certainly arent huge.That gives 7% for clocks, and maybe another 5-10% on top of that for architecture.OF course, maybe vega will be some dramatic departure from even polaris.
But on paper it looks like this card would still lose to a 1080 on DX11 titles.Would win DOOM, barely.Maybe somewhat on par with DX12 titles that were written properly.All that, at higher power consumption.
Hope they have some serious magic dust to be competetive.

[zanzabar]

Yeah, i just browsed some fury X reviews, power consumption at load is all over the place friom 250 to 300.However if AMD states 275W, then we can assume same metrics apply to vega ,so it could also consume more power, thats 50W diff.
Fury X GCN 1.2 1050 Core clock, polaris on the same process as vega has 1120,Mem bandwith will be the same but with addition of compression.I guess nobody estimated gains on polaris due to the newer architecture, but they certainly arent huge.That gives 7% for clocks, and maybe another 5-10% on top of that for architecture.OF course, maybe vega will be some dramatic departure from even polaris.
But on paper it looks like this card would still lose to a 1080 on DX11 titles.Would win DOOM, barely.Maybe somewhat on par with DX12 titles that were written properly.All that, at higher power consumption.
Hope they have some serious magic dust to be competetive.

the fury pro still trades blows with the titan x pascal in dx11, if they can get to 1600mhz or so on the core of a single gpu card it should keep up fine against the p100. i honestly dont even see the point of HMB for consumer workloads like gaming. there is almost difference in the 1080 if you over or under clock the ram.

[vario]

the fury pro still trades blows with the titan x pascal in dx11, if they can get to 1600mhz or so on the core of a single gpu card it should keep up fine against the p100. i honestly dont even see the point of HMB for consumer workloads like gaming. there is almost difference in the 1080 if you over or under clock the ram.

Comparing a dual GPU watercooled 600W card to a normal single GPU and TDP card is somewhat pointless i think .1600Mhz, yea, if they could.But polaris OC`s to 1300 ,and power consumption spikes if you overvolt to achieve 1400-1500.Will they fix these problems in 8 months ? Maybe.But i HIGHLY doubt vega will be clocked more than 1200-1300.Hope im wrong tho.1500Mhz with room to spare and yea , these cards would rock.Im hoping to buy one, but if launch will be as underwhelming as polaris, im gonna wait more.

[AliG]

Comparing a dual GPU watercooled 600W card to a normal single GPU and TDP card is somewhat pointless i think .1600Mhz, yea, if they could.But polaris OC`s to 1300 ,and power consumption spikes if you overvolt to achieve 1400-1500.Will they fix these problems in 8 months ? Maybe.But i HIGHLY doubt vega will be clocked more than 1200-1300.Hope im wrong tho.1500Mhz with room to spare and yea , these cards would rock.Im hoping to buy one, but if launch will be as underwhelming as polaris, im gonna wait more.

Regarding clocks, AMD is claiming they made significant improvements in performance:watt. As far as we know, that could be purely to HBM2.

Having said that, they did pull off quite a feat with their 28nm APUs. Not promising they'll hit 1600 MHz (I'm with you that 1200 is more realistic), but I wouldn't be surprised if the cards have more headroom due to the design efficiency improvements.

[zanzabar]

Comparing a dual GPU watercooled 600W card to a normal single GPU and TDP card is somewhat pointless i think .1600Mhz, yea, if they could.But polaris OC`s to 1300 ,and power consumption spikes if you overvolt to achieve 1400-1500.Will they fix these problems in 8 months ? Maybe.But i HIGHLY doubt vega will be clocked more than 1200-1300.Hope im wrong tho.1500Mhz with room to spare and yea , these cards would rock.Im hoping to buy one, but if launch will be as underwhelming as polaris, im gonna wait more.

i am not saying that the pro is viable card, it just gives us an idea that 2 fury X will scale about 60% and that the pascal titian is also about the same speed. i really hope it will clock high, but you and AliG are most likely on the right track of 1200-1300. but then again if it does clock i dont the p100 will completely outclass it.

[AliG]

I just don't understand why AMD's design is so unbalanced. The raw numbers suggest the RX 480 has way more TFLOPs than the GTX 1060, and is even within striking distance of the 1070. Especially given the power consumption levels, you would think the real world performance would scale much better.

While the GCN cards are fine in DX12, it's clear the lack of ROPs is killing them everywhere else. It baffles me why they continued to force the issue with Polaris given how long they had to make design updates (due to lack of 20nm).

[AliG]

I guess the only thing I'll mention is if they price Vega 10 at same $600, they'll steal a huge chunk of market share.

The only good thing about the RX 480 was its price point enabled AMD to get big volume out. That allowed them to dig deep into the midrange market and is why they expect to be profitable in Q3. At the right price, the RX 490 can at least be a big boon to gamers who want high end performance but can't afford the ridiculous P102 and P104 pricing.