AMD VEGA confirmed for 2017 H1

Griffildur · 18 Nov 2016 at 19:54

LoadsaMoney said:
http://en.yibada.com/articles/174013/20161117/amd-radeon-rx-490-feature-vega-10-gpu-release-slated.htm

that article feels like rubbish. statements like speed up 1080p gaming to 60 fps and 1.5 TFlops performance ... I mean seriously ???

I wouldn't put any stock on it.

Kaapstad · 18 Nov 2016 at 21:10

Seanspeed said:
No idea what Kaapstad is talking about there.

HBM's advantage is bandwidth, size-on-card and lower power primarily. Three things that will absolutely make an impact. Fiji's 4GB limitation and inclusion on a card that was really only ever great for 1440p at best kind of put it in a position of not being able to show off HBM's capabilities too much. HBM2 plus a far more powerful card(Vega 10/11) should really paint a strong picture. Especially with games really starting to push into the 6GB+ area of vRAM requirements even at 1080p. HBM2-equipped cards should be able to just basically handle that like a fire hose scrubbing off crumbs on a plate.

And of course the board size and power saving benefits that help in terms of form factor and thermal limitations.

Don't forget NVidia already have a HBM2 card out that is identical in specs (Core count) to the Pascal Titan except for the use of GDDR5X and higher clockspeeds on the later.

I don't think cost would be a factor on a Titan card so NVidia must have had other sound reasons for going with GDDR5X on the higher clocking Titan. The HBM2 equipped Tesla cards will not be run at anything like the clockspeeds you can reach with a watercooled Pascal Titan.

And yes I do think latency is more important than huge bandwidth when it comes to gaming cards.

Gerard · 18 Nov 2016 at 21:30

Kaapstad said:
Don't forget NVidia already have a HBM2 card out that is identical in specs (Core count) to the Pascal Titan except for the use of GDDR5X and higher clockspeeds on the later.

I don't think cost would be a factor on a Titan card so NVidia must have had other sound reasons for going with GDDR5X on the higher clocking Titan. The HBM2 equipped Tesla cards will not be run at anything like the clockspeeds you can reach with a watercooled Pascal Titan.

And yes I do think latency is more important than huge bandwidth when it comes to gaming cards.

We don't know what the availability of hbm 2 is so its very possible they went with gddr5x which seemed to be in decent availability at the time. Using essentially the same cooler could also be a reason instead of having to design one from the ground up for what might be a much shorter card.

Kaapstad · 18 Nov 2016 at 21:43

Gerard said:
We don't know what the availability of hbm 2 is so its very possible they went with gddr5x which seemed to be in decent availability at the time. Using essentially the same cooler could also be a reason instead of having to design one from the ground up for what might be a much shorter card.

GPUs are planned a very long time in advance, long before it is known what the availability of HBM2 is likely to be exactly.

No I think the decision to avoid HBM2 on a high clocking gaming card was done for other reasons.

Gerard · 18 Nov 2016 at 21:57

Kaapstad said:
GPUs are planned a very long time in advance, long before it is known what the availability of HBM2 is likely to be exactly.

No I think the decision to avoid HBM2 on a high clocking gaming card was done for other reasons.

I'm well aware they're planned years in advance, bottom line is we don't know the reason, could be price, could be availability, could be they wanted to use hbm 2 on the "pro" cards so they could ratchet the price more on those. They have their reason i suppose.

Kaapstad · 18 Nov 2016 at 22:08

Gerard said:
I'm well aware they're planned years in advance, bottom line is we don't know the reason, could be price, could be availability, could be they wanted to use hbm 2 on the "pro" cards so they could ratchet the price more on those. They have their reason i suppose.

Could be that HBM2 is not good for gaming cards !!!

Don't forget that the difference in clockspeed between HBM1 and HBM2 is less than the performance difference between the Fury X and a watercooled Pascal Titan.

I don't think HBM1 did the Fury X any favours at normal resolutions and I think the same would have been the case for HBM2 on the Pascal Titan.

nashathedog · 18 Nov 2016 at 23:08

Kaapstad said:
Don't forget NVidia already have a HBM2 card out that is identical in specs (Core count) to the Pascal Titan except for the use of GDDR5X and higher clockspeeds on the later.

I don't think cost would be a factor on a Titan card so NVidia must have had other sound reasons for going with GDDR5X on the higher clocking Titan. The HBM2 equipped Tesla cards will not be run at anything like the clockspeeds you can reach with a watercooled Pascal Titan.

And yes I do think latency is more important than huge bandwidth when it comes to gaming cards.

As I've been saying for ages I think using HBM affects how well a card can be overclocked and Nvidia currently rely on good overclocking with there gpu 3.0 tech so them not using it on there Pascal gaming cards comes as no surprise..

AllBodies · 19 Nov 2016 at 00:07

Kaapstad said:
Don't forget NVidia already have a HBM2 card out that is identical in specs (Core count) to the Pascal Titan except for the use of GDDR5X and higher clockspeeds on the later.

I don't think cost would be a factor on a Titan card so NVidia must have had other sound reasons for going with GDDR5X on the higher clocking Titan. The HBM2 equipped Tesla cards will not be run at anything like the clockspeeds you can reach with a watercooled Pascal Titan.

And yes I do think latency is more important than huge bandwidth when it comes to gaming cards.

GP100 and GP102 are actually very different.

They both have the same number of FP32 cores, but the GP100 (Tesla P100 card) also has a bunch of FP64 cores (1792 according to what I can find), a different memory controller, and a few other changes.

GP100 is 610mm2
GP102 is 471mm2

So it's the fact it's a larger and denser chip, with 5376 total cores that its TDP is so high, and it runs at lower clocks. We have no evidence HBM2 is preventing its clockspeed increasing further.

CAT-THE-FIFTH · 19 Nov 2016 at 00:27

Apparently,AMD has managed to increase marketshare!!

http://www.mercuryresearch.com/graphics-pr-2016-q4.pdf

Imagine if they actually had Vega launched by now??

Dave2150 · 19 Nov 2016 at 08:18

CAT-THE-FIFTH said:
Apparently,AMD has managed to increase marketshare!!

http://www.mercuryresearch.com/graphics-pr-2016-q4.pdf

Imagine if they actually had Vega launched by now??

Vega will be more expensive than a 1080 if it's rumoured 16GB HBM2 is correct, so I very much doubt it will get them much market share.

Dave2150 · 19 Nov 2016 at 08:25

Kaapstad said:
GPUs are planned a very long time in advance, long before it is known what the availability of HBM2 is likely to be exactly.

No I think the decision to avoid HBM2 on a high clocking gaming card was done for other reasons.

You may be right, though I think NVIDIA knew they could 'get away' with only using GDDR5X for Pascal and still easily get the performance crown.

This then enables them to use HBM2 to further encourage people to upgrade to a 1180 (Pascal refresh or Volta).

Griffildur · 19 Nov 2016 at 08:52

Dave2150 said:
You may be right, though I think NVIDIA knew they could 'get away' with only using GDDR5X for Pascal and still easily get the performance crown.

This then enables them to use HBM2 to further encourage people to upgrade to a 1180 (Pascal refresh or Volta).

In a way I don't feel like HBM2 is going to be actually useful at all. Imagine HBM2 memory on the Pascal cards and then imagine the price of that combo. Heck we're still complaining about the prices of Generation 10. Just look how much that Fury cost with its HBM memory.

It would make for a great laugh, if Vega does come with HBM2 and a crazy price as a result. Pretty sure it's not going to help AMD at all, even if the card is better than a 1080 ( and I don't believe it will be anywhere near that level of performance, I'd be amazed if it's anywhere near 1070 to be honest ).

TNA · 19 Nov 2016 at 09:02

Griffildur said:
Pretty sure it's not going to help AMD at all, even if the card is better than a 1080 ( and I don't believe it will be anywhere near that level of performance, I'd be amazed if it's anywhere near 1070 to be honest ).

For real? Not like Fury X is a million miles away in performance to the 1070. So a brand new architecture for high end + HBM2 + 14nm = 1070 performance :confused:

Mark my words, it will be Titan XP performance or thereabouts at least. If they cannot beat 1080 easily I will eat my hat

Bonjour · 19 Nov 2016 at 09:03

I'll be very surprised if top-end Vega doesn't beat the 1070.

Jono8 · 19 Nov 2016 at 09:21

Vega will have to match/beat a 1080 or it will be a massive failure.

Jono8 · 19 Nov 2016 at 09:22

Griffildur said:
I'd be amazed if it's anywhere near 1070 to be honest .

Why?!

It is obviously going to be faster than a 1070.

triss · 19 Nov 2016 at 09:28

Jono8 said:
Why?! It is obviously going to be faster than a 1070.

TBH other then faster then say a 480x we have no idea where Vega will be

Bonjour · 19 Nov 2016 at 09:31

triss said:
TBH other then faster then say a 480x we have no idea where Vega will be

Faster than a Fury X, surely?

Griffildur · 19 Nov 2016 at 09:32

Jono8 said:
Why?! It is obviously going to be faster than a 1070.

I am not falling for it this time. We had enough build ups for other cards and major disappointments including the 480. Hopefully we only have a few months until Vega is released and then we'll know what's what.

Griffildur · 19 Nov 2016 at 09:39

@SiDeards73 your post is not really worth responding to, but you might want to consider the fact that many people will have points of view different to yours and that does not give you the right to call them anything.

If anything it actually makes you look bad. We are here to discuss things, it's a forum not an AMD or Nvidia convention.

Points of view which do not match our own ideas should be welcomed and if you're not willing to accept other points of view than yours, well, the loss is yours.