AMD VEGA confirmed for 2017 H1

CAT-THE-FIFTH · 12 Dec 2016 at 12:26

Nobody is saying the GTX970 is not a great value card,but the moment it is mentioned the R9 290 came out on year earlier,looks to be ahead in newer APIs should give it a bit more longevity,and also cratered the prices of the GTX780 to the point it actually for a period was very good value,no one should mention that.

Or the fact it was so overhyped that people were saying it was the best card ever made. Yet the best card ever made can't even beat an ancient grandfather card which got some make-up put on it. LMAO.

I think a few cards might want to talk to the GTX970 about that.

But,but you MUST HATE NVIDIA,right?? Yeah,in a AMD thread. You can't even make it up. Yes I apparently must hate Nvidia SOOOO much,I keep telling people at every chance the Fury series are VRAM limited and the GTX980TI was a better buy and even now at low prices its a bit hit and miss,or to push people from XFire since I keep reminding people of the issues with multi-GPU in games.

Plus OFC,lets just be more worried about the GTX970 when I actually bothered to post some Vega news in a Vega thread.

Baboonanza · 12 Dec 2016 at 12:36

CAT-THE-FIFTH said:
Plus OFC,lets just be more worried about the GTX970 when I actually bothered to post some Vega news in a Vega thread.

This thread is for Vega news? Where should I go for pointless bickering?

tommybhoy · 12 Dec 2016 at 12:37

290/X AND the 970 were both phenomenal cards personally after gaming on them individually for over a year each.

Ones 'better' than the other in different ways but if I absolutely had to pick a winner, I'd only get grief for my choice.

TheRealDeal · 12 Dec 2016 at 12:40

tommybhoy said:
290/X AND the 970 were both phenomenal cards personally after gaming on them individually for over a year each.

Ones 'better' than the other in different ways but if I absolutely had to pick a winner, I'd only get grief for my choice.

It's obvious what's better. The one that's a year older and competing now with the gtx980 in the form of a 390x

.

humbug · 12 Dec 2016 at 12:40

The 970 deserves a lot of respect, but its not the card i would nominate as the best alround card of all time, its a bit flawd and it has failed the test of time.

CAT-THE-FIFTH · 12 Dec 2016 at 12:41

Baboonanza said:
This thread is for Vega news? Where should I go for pointless bickering?

I will post it for the third time:

http://videocardz.com/64677/amd-ann...erator-radeon-instinct-mi25-for-deep-learning

So that is probably what all that talk was about last week,a new deep learning card called the Radeon INSTINCT MI25(yes its a mouthful).

Edit!!

It has around 50% more FP32 TFLOPs than Fiji but this for the compute cards. I am not sure how this relates to the gaming cards.

humbug · 12 Dec 2016 at 12:47

CAT-THE-FIFTH said:
I will post it for the third time:

http://videocardz.com/64677/amd-ann...erator-radeon-instinct-mi25-for-deep-learning

So that is probably what all that talk was about last week,a new deep learning card called the Radeon INSTINCT MI25(yes its a mouthful).

Edit!!

It has around 50% more FP32 TFLOPs than Fiji but this for the compute cards. I am not sure how this relates to the gaming cards.

At least we know it exists now, shouldn't be long before we see gaming versions of it

CAT-THE-FIFTH · 12 Dec 2016 at 12:50

Looking at the TDP ratings,they don't look inspiring though unless AMD is being intentionally vague about it.

humbug · 12 Dec 2016 at 12:59

CAT-THE-FIFTH said:
Looking at the TDP ratings,they don't look inspiring though unless AMD is being intentionally vague about it.

TDP can mean anything, can't take anything away from that.

On 50% more TFlops, workstation cards, especially passively cooled ones are clocked significantly lower than their gamers counterparts.

They have a chart there of Shaders vs Mhz = performance, i would suggest MI25 is running 6250 Shaders @ 1000Mhz for its 25TFlops 16Bit FP - 12.5TFlops 32Bit FP

The gamers card with a proper cooler should run at 1200Mhz or more, that would make it about 15TFlops 32Bit FP.

To put that into perspective the Fury-X is 8.6TFlops, So big Vega would be 75% more powerful than Fury-X, a colossal amount of muscle, 3x RX 480, He-Man of the GPU world :O

CAT-THE-FIFTH · 12 Dec 2016 at 13:06

It depends whether it using the smaller Vega chip or the larger one.

It could be using the purported smaller die but running at higher clockspeeds like the Polaris equivalents. Videocardz thinks it is 4096 shaders running at upto 1.5GHZ or thereabouts.

If that 50% increase in TFLOPs over the Fury based cards is true in gaming scenarios,surely that is around GTX1080 level,and taking into account the Fury based commerical cards tend to be clocked a bit lower.

Lets say 40% higher,and its the smaller die clocked higher:

https://www.techpowerup.com/reviews/MSI/GTX_1050_Gaming_X/27.html

So probably around GXT1080 level at 1080P,but a bit more faster at higher resolutions.

Edit!!

Having said that the clockspeed looks too high for an AMD card.

I dunno now.

humbug · 12 Dec 2016 at 13:16

CAT-THE-FIFTH said:
It depends whether it using the smaller Vega chip or the larger one.

It could be using the purported smaller die but running at higher clockspeeds like the Polaris equivalents. Videocardz thinks it is 4096 shaders running at upto 1.5GHZ or thereabouts.

If that 50% increase in TFLOPs over the Fury based cards is true in gaming scenarios,surely that is around GTX1080 level,and taking into account the Fury based commerical cards tend to be clocked a bit lower.

Lets say 40% higher,and its the smaller die clocked higher:

https://www.techpowerup.com/reviews/MSI/GTX_1050_Gaming_X/27.html

So probably around GXT1080 level at 1080P,but a bit more faster at higher resolutions.

Edit!!

Having said that the clockspeed looks too high for an AMD card.

I dunno now.

Videocardz are not too bright, to get 1.5Ghz on a large passively cooled card AMD would suddenly have to have found a way to clock higher than Nvidia on lower volts.

Workstation cards need a stability level far higher than that of gamers cards, which is one reason they are always clocked far lower, if Big Vega is running 1.5Ghz in workstation form it will clocked around 2Ghz retail.

Can't see it happening.

This is a bigger card clocked low, yet even then its actually only about 75% the size of Fiji

Gregster · 12 Dec 2016 at 13:23

tommybhoy said:
290/X AND the 970 were both phenomenal cards personally after gaming on them individually for over a year each.

Ones 'better' than the other in different ways but if I absolutely had to pick a winner, I'd only get grief for my choice.

If I had to choose between the 2 and no sync screen involved, I would choose the 290X. Grief or not, they are both great cards but the 290X would edge it for me.

CAT-THE-FIFTH · 12 Dec 2016 at 13:30

humbug said:
Videocardz are not too bright, to get 1.5Ghz on a large passively cooled card AMD would suddenly have to have found a way to clock higher than Nvidia on lower volts.

Workstation cards need a stability level far higher than that of gamers cards, which is one reason they are always clocked far lower, if Big Vega is running 1.5Ghz in workstation form it will clocked around 2Ghz retail.

Can't see it happening.

This is a bigger card clocked low, yet even then its actually only about 75% the size of Fiji

So that would put it a tad faster than a Pascal Titan X?? Remember Nvidia could launch an SKU with the remaining shaders unlocked but that would probably be less than a 10% uplift in performance.

I think the clockspeeds are really going to be the deciding factor again.

humbug · 12 Dec 2016 at 13:35

CAT-THE-FIFTH said:
So that would put it a tad faster than a Pascal Titan X?? Remember Nvidia could launch an SKU with the remaining shades unlocked but that would probably be less than a 10% uplift in performance.

I think the clockspeeds are really going to be deciding factor again.

The GPU is more powerful either way, Pascal TX is 10.8TFlops 32Bit vs Big Vega 12.5TFlops 32Bit.

Weather or not AMD have found a way of making use of all that power is another question, the biggest problem with the 8.6TFlop Fury-X and even to a significant extent the 6TFlop 390X is that they haven't. a lot of that power is bottlenecked by DX11 Drawcall overheads, their call pre-emption with those cards just isn't as efficient as Nvidia. thats how they are able to get more gaming performance from weaker cards.

CAT-THE-FIFTH · 12 Dec 2016 at 13:43

humbug said:
The GPU is more powerful either way, Pascal TX is 10.8TFlops 32Bit vs Big Vega 12.5TFlops 32Bit.

Weather or not AMD have found a way of making use of all that power is another question, the biggest problem with the 8.6TFlop Fury-X and even to a significant extent the 6TFlop 390X is that they haven't. a lot of that power is bottlenecked by DX11 Drawcall overheads, their call pre-emption with those cards just isn't as efficient as Nvidia. thats how they are able to get more gaming performance from weaker cards.

Well Polaris 10 seems to do a bit better with less specs overall when compared to Hawaii,and apparently DX11 overhead is a bit better with the new ReLive drivers.

However,I do hope they work on the cooler and drivers before launch,ie,do a polished launch like Nvidia does.

queamin · 12 Dec 2016 at 13:48

A large passively cooled card would be clocked low for less heat I would think.

humbug · 12 Dec 2016 at 13:49

CAT-THE-FIFTH said:
Well Polaris 10 seems to do a bit better with less specs overall when compared to Hawaii,and apparently DX11 overhead is a bit better with the new ReLive drivers.

However,I do hope they work on the cooler and drivers before launch,ie,do a polished launch like Nvidia does.

Right, RX 480 5TFlops vs R9 390X 6TFlops and yet they are on par bar a couple % give and take

CAT-THE-FIFTH · 12 Dec 2016 at 13:52

humbug said:
Right, RX 480 5TFlops vs R9 390X 6TFlops and yet they are on par bar a couple % give and take

More the case,its a 4 ACE design against an 8 ACE design,with nearly half the memory bandwidth and half the ROPs.

So technically Polaris 10 has more bottlenecks as a design,but the improved handling of tessellation,etc seems to have helped.

humbug · 12 Dec 2016 at 14:07

CAT-THE-FIFTH said:
More the case,its a 4 ACE design against an 8 ACE design,with nearly half the memory bandwidth and half the ROPs.

So technically Polaris 10 has more bottlenecks as a design,but the improved handling of tessellation,etc seems to have helped.

True

CAT-THE-FIFTH · 12 Dec 2016 at 14:24

Here is the official announcement:

http://www.anandtech.com/show/10905/amd-announces-radeon-instinct-deep-learning-2017