• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD VEGA confirmed for 2017 H1

Status
Not open for further replies.
Therer is still this rumour:
https://videocardz.com/65649/sk-hynix-updates-memory-product-catalog-hbm2-available-in-q1-2017
The provider of HBM2 memory for AMD is officially saying only slower 1.6Gbps stacks are available, which would leave Vega with less bandwidth than the 1080ti, with availability forecast for Q1 this year but wasn't available when they produced their catalogue.

https://www.extremetech.com/computing/248267-sk-hynix-will-launch-gddr6-2018-hbm2

IF Vega can only launch with the slower HBM2 it would have 410Gbps bw, 1080ti is already at 480Gbps, if AMD can get the faster HM2 then they still only get 512Gb with 2 stacks. GDDR6 is available early next year giving 768Gbps for big gaming Volta.

AMD may have screwed themselves again waiting for HBM technology to catch up.

It's like they've missed the boat completely, What HBM is Nvidia's work card using? Isn't that fast HBM2?

When you look at the speeds 5x and Gddr6 can do it makes me wonder why AMD took the HBM route for faster memory.
AMD seem to have backed the wrong horse.
I think that at the moment the memory being on the die is why overclockings not too good and I think Nvidia aren't interested in it yet because gpu boost won't play nicely with it.
 
It's like they've missed the boat completely, What HBM is Nvidia's work card using? Isn't that fast HBM2?

When you look at the speeds 5x and Gddr6 can do it makes me wonder why AMD took the HBM route for faster memory.
AMD seem to have backed the wrong horse.
I think that at the moment the memory being on the die is why overclockings not too good and I think Nvidia aren't interested in it yet because gpu boost won't play nicely with it.

The NVIDIA P100 uses HBM2 with 720GB/s for 16GB, and 540GB/s for 12GB; both being the 1.4Gbps stacks.
http://www.anandtech.com/show/10433/nvidia-announces-pci-express-tesla-p100

The Quadro GP100 has 717GB/s with 16GB of HBM2
http://www.pny.com/nvidia-quadro-gp100

If AMD can only use 1.6Gbps, it's still than what NVIDIA are currently using, although they're only using two HBM2 dies, instead of 3-4 like NVIDIA.
 
Therer is still this rumour:
https://videocardz.com/65649/sk-hynix-updates-memory-product-catalog-hbm2-available-in-q1-2017
The provider of HBM2 memory for AMD is officially saying only slower 1.6Gbps stacks are available, which would leave Vega with less bandwidth than the 1080ti, with availability forecast for Q1 this year but wasn't available when they produced their catalogue.

https://www.extremetech.com/computing/248267-sk-hynix-will-launch-gddr6-2018-hbm2

IF Vega can only launch with the slower HBM2 it would have 410Gbps bw, 1080ti is already at 480Gbps, if AMD can get the faster HM2 then they still only get 512Gb with 2 stacks. GDDR6 is available early next year giving 768Gbps for big gaming Volta.

AMD may have screwed themselves again waiting for HBM technology to catch up.
Man, I hope this isn't true. That'd be a monumental disappointment and would have huge implications for not just Vega, but the future of AMD's position in the gaming GPU market.

HBM is apparently quite a bit more expensive than GDDR5, so it's possible Vega wont have some great price point as people are hoping if it's not at least a 1080Ti competitor. It would certainly explain why Nvidia haven't made any drastic moves as this would be information they'd be very well aware of.
 
It's like they've missed the boat completely, What HBM is Nvidia's work card using? Isn't that fast HBM2?

When you look at the speeds 5x and Gddr6 can do it makes me wonder why AMD took the HBM route for faster memory.
AMD seem to have backed the wrong horse.
I think that at the moment the memory being on the die is why overclockings not too good and I think Nvidia aren't interested in it yet because gpu boost won't play nicely with it.


The Tesla GP100 uses lower speed HBM2 but there are 4 stacks giving 720Gbps. It is unlikely AMD can afford to have 3 or 4 stack of memory for Vega, it also would require design changes.
 
The NVIDIA P100 uses HBM2 with 720GB/s for 16GB, and 540GB/s for 12GB; both being the 1.4Gbps stacks.
http://www.anandtech.com/show/10433/nvidia-announces-pci-express-tesla-p100

The Quadro GP100 has 717GB/s with 16GB of HBM2
http://www.pny.com/nvidia-quadro-gp100

If AMD can only use 1.6Gbps, it's still than what NVIDIA are currently using, although they're only using two HBM2 dies, instead of 3-4 like NVIDIA.


But since they are only using 2 and not 4 they halve the bandwidth. Thus is AMD have to use 1.6Gbps Vega will have less bandwidth than Pascal, if they get the 2.0Gbps then they will get a tiny bit more. Doesn't mean HBM2 is a complte waste of time, it does lower power consumption which AMD needs to do.
 
But since they are only using 2 and not 4 they halve the bandwidth. Thus is AMD have to use 1.6Gbps Vega will have less bandwidth than Pascal, if they get the 2.0Gbps then they will get a tiny bit more. Doesn't mean HBM2 is a complte waste of time, it does lower power consumption which AMD needs to do.

It'll be interesting to see what they do, although I still feel Vega is for their Instinct line first and foremost.
They want the big money from enterprise and deep learning, and since their R&D budget was rather meh, it does make sense to go for those areas first, and then taper down for consumer levels.

Sadly they couldn't do what NVIDIA did, and have two separate designs, one for HBM2 + HBCC, and one for GDDR5X.

Well, just another month before we find out the truth really; will certainly be interesting.
 
Man, I hope this isn't true. That'd be a monumental disappointment and would have huge implications for not just Vega, but the future of AMD's position in the gaming GPU market.

HBM is apparently quite a bit more expensive than GDDR5, so it's possible Vega wont have some great price point as people are hoping if it's not at least a 1080Ti competitor. It would certainly explain why Nvidia haven't made any drastic moves as this would be information they'd be very well aware of.


Nvidia have been on the ball with memory technology and timelines recently. they didnt waste time getting HBM in to Maxwell, or even Pascal consumer. They skipped HBM1 entirely and went straight to 2.0 which they used for the Tesla parts where huge bandwidth is a bg benefit, margins are huge and volume is low sp a short supply of expensive HBM2 works fine when you are selling $129,000 tesla boxes. Instead they went for GDDR5X which despite some pessimism was delivered on time and gave enough of a performance boost for pascal gaming GPUs. Now it seems that GDDR6 is going to be available for Volta and offer 768Gbps on a 384bit interface. Meanwhile Vega looks to be between 410Gbps and 512Gbps. Of course GDDR6 may get delayed and thus delay Volta.
 
Therer is still this rumour:
https://videocardz.com/65649/sk-hynix-updates-memory-product-catalog-hbm2-available-in-q1-2017
The provider of HBM2 memory for AMD is officially saying only slower 1.6Gbps stacks are available, which would leave Vega with less bandwidth than the 1080ti, with availability forecast for Q1 this year but wasn't available when they produced their catalogue.

https://www.extremetech.com/computing/248267-sk-hynix-will-launch-gddr6-2018-hbm2

IF Vega can only launch with the slower HBM2 it would have 410Gbps bw, 1080ti is already at 480Gbps, if AMD can get the faster HM2 then they still only get 512Gb with 2 stacks. GDDR6 is available early next year giving 768Gbps for big gaming Volta.

AMD may have screwed themselves again waiting for HBM technology to catch up.

That is THREE MONTHS old speculative article.....

From the already pretty limited leaked info as far as three weeks ago, we know that the Vega is equipped with 2048bit bus and 512GB/s HBM2 ram.
 
Buy a Vega for gaming, its like buying a Tesla from Nvidia, and just gaming on that, but you don't have to, as Nvidia also do gaming cards.
 
Last edited:
That is THREE MONTHS old speculative article.....

From the already pretty limited leaked info as far as three weeks ago, we know that the Vega is equipped with 2048bit bus and 512GB/s HBM2 ram.
Wasn't Fury 4096bit bus? 2048bit will make it cheaper I suppose.

I thought Vega, at least the top one might have been 4096bit bus also making it over 1TB/s. Would look nice on paper anyway :p

512GB/s just sounds so standard these days for top end card. Fury X had it ages ago. Not that I am complaining, I would be happy with 512GB/s. My 1080 has around a poorly 320GB/s :D
 
Wasn't Fury 4096bit bus? 2048bit will make it cheaper I suppose.

I thought Vega, at least the top one might have been 4096bit bus also making it over 1TB/s. Would look nice on paper anyway :p

512GB/s just sounds so standard these days for top end card. Fury X had it ages ago. Not that I am complaining, I would be happy with 512GB/s. My 1080 has around a poorly 320GB/s :D

Unfortunately yes, only 512GB/s but it will do the job especially with a nice overclock.
Hell even the FuryX does 614.4Gb/s @ 600Mhz overclock, and the Nano which officially doesn't support is can do 563.2Gb/s.(550Mhz)



Does the job done, but need more. TW Attila, TW Warhammer are the biggest killers of me :(
 
That is THREE MONTHS old speculative article.....

From the already pretty limited leaked info as far as three weeks ago, we know that the Vega is equipped with 2048bit bus and 512GB/s HBM2 ram.

The 2nd article is 3 days old.

Do you have a link to AMD officially listing the specs?
 
Hundred pages back, you will find the supposed leaked benchmarks.

So no actual confirmation then?

We all know that Vega was supposed to use 2Ghz HBM for 512Gbps. Hopefully that will still eb the case, but that speed HBM2 hasn't been available in the past it is expected to be available now although the most recent information indicates it isn't.
 
I wouldn't be surprised if we saw some kind of weird memory arrangement where vega has 2GB of HBM built directly into the gpu die and then has traditional GDDR memory as well, thus explaining what this high bandwidth cache controller is.
 
Yea just read this.... but Quake champions doesn't look good graphically. :o

Yeah I laughed so hard to read RTG will bundled Quake Champions with Vega. Not very impressed with terrible very old dated Quake 3 Arena type graphics after played beta. :o

Nvidia did very good job decided to bundled Paragon with GTX 1060 cards in recently promotion.

I played Paragon open beta, wow it has very stunning photorealism graphics and very fun to play than very boring Quake Champions.


 
Status
Not open for further replies.
Back
Top Bottom