AMD: Speculation, Rumors, and Discussion (Archive)

Status
Not open for further replies.
Actually, when Lisa Su was asked at the technology leadership conference not quite a week ago (which I assume you're referring to), if we can expect high end cards over the next two years, she said 'more like within the next two quarters'.

Dual Polaris 490 more than likely.
 
I think quarters are almost always used to mean January–March, April–June, July–September, October–December, or the specific company's fiscal quarters. So we can probably expect a (possibly papery) launch in December.
Judging from Polaris, we might see a reveal at the end of October, a presentation the week before Black Friday, a dribble in the beginning of December, and depending on price/performance relative to nVidia, reviews and preorders either ten days before Xmas or a week into January, followed by retail availability God knows when. :D
 
It's disappointing for sure, but I suspect they just don't have the engineering manpower to pull off a more aggressive development schedule than this - and even so, Polaris is mediocre at best. Money may be a concern as well. And their console commitments will be stealing resources as well.

One example is the Xbox One S as another Polaris 10-class piece of silicon (edit: in area), with AMD also doing the work to hop foundries to make silicon with last-gen's CPU and GPU tech.

If the Playstation Neo is a little more successful at shrinking than the Xbox One S, it stands to be in the same size range as GP104. It's something of a question mark if this investment in resources was put into a Polaris-derived architecture, or if this went into a variant of the PS4's IP level.


With the star theme, and Vega also being a rocket, a cutesy idea for a launch venue would be a literal launch venue.
I would feel it would take a bit of time to get something like that to not look that dilapidated, although if Vega is not a significant architectural revamp AMD's relying an aging and unimproved framework would be equally on the nose.
 
Last edited:
Judging from Polaris, we might see a reveal at the end of October, a presentation the week before Black Friday, a dribble in the beginning of December, and depending on price/performance relative to nVidia, reviews and preorders either ten days before Xmas or a week into January, followed by retail availability God knows when. :D
You must have inside info on this fast Vega rollout plan! Polaris's events started with the first proclamation in November 2015. Unveiled in December 2015. Demoed January 2016. Specced in February. Announced in March 2016. Revealed in May 26 2016. Heralded on May 31, 2016. Released June 29.

So Entropy's new info of a mere 4 month Vega launch announcement cycle shows a powerful doubling of AMD's PR cadence.
 
Last edited:
Not much a problem. We had GCN first information way before the release ( around august 2011, when 7970 have been released in early 2012 )..

First "presentation " we had the name and roadmap", sorry but i find AMD really slow to give name of new sku, features and roadmap compared to Nvidia ..

March 2014 ... Volta... Stacked Vram, Nvlink etc...
http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

March 2015 " Pascal enter, push volta in the roadmap : features shown and announced mixed precision, HBM, NvLink, etc etc. 10x performance over Maxwell ( in mixed precision ofc ). ooh he even show a prototype.

http://www.extremetech.com/gaming/2...mance-gains-from-upcoming-pascal-architecture

I dont say i like the way it is made, but its bit how things work for this industry.
 
Last edited:
The only thing I care about Vega is the rasterizer. GCN Gen 1 rasterizer is too old now.
In my layman POV the GCN itself is "too old"... The shader core and other crucial parts haven't enjoyed a major update since it's launch in Jan 2012. Although nVidia keeps rolling out one change after another. The only visible updates to GCN are: stupid 3rd party IPs (yes, im looking at you AUDIO), compute related stuff (hyped dedicated HW for dynamic scheduling, queueing, etc.) and minor geometry improvements.

I know that AMD is still close to bankruptcy but still, this is more and more looking like another Via/Maxtor/SiS story...

Let's hope that GFX IP 9.0 aka Vega will be a major departure from rusty GCN.
 
The only thing I care about Vega is the rasterizer. GCN Gen 1 rasterizer is too old now.
Yeah. The compute units are still pretty good and the improved geometry pipeline + DCC have certainly helped. AMD drivers are no longer a problem in DX12 and Vulkan. Async compute also helps with utilization. Rasterizer should be top priority right now.
 
16nm is limited in that TSMC is *not* fully rampped up yet, they expect full ramp up by end of this year. 14nm Gloflo not sure, haven't heard any news on their side.

TSMC has already ramped up 16nm production ahead of the iPhone 7 launch..but overall demand is still higher than supply. In years past, after FPGAs, GPUs were usually the next chips out on a new process node and SoCs were never on the leading process. This started changing around 45/40nm and ever since the 28nm node, both AMD and Nvidia have faced significant competition from Qualcomm and Apple to secure wafer capacity. The foundries are laughing all the way to the bank.
Only in my wildest dreams.

I'd love to see a high-end GPU from the red team, tomorrow if at all possible, but moving a big, complex GPU forwards by months - possibly half a year - when has that ever happened?

RV770? :smile:
 
Yeah. The compute units are still pretty good and the improved geometry pipeline + DCC have certainly helped. AMD drivers are no longer a problem in DX12 and Vulkan. Async compute also helps with utilization. Rasterizer should be top priority right now.

Lets hope it is a priority for vega and not for the following architecture. Should not be so hard for them to introduct thing as tiled rasterization looking at all the papers they have made about it. ( and patents they surely own ( Intel have some patents, but i have not check for AMD ).

When reading about the work made by Nvidia on Pascal related to clock speed ( trying to find all the way possible for maximize it ), i hope AMD can do similar work on Vega.

In my opinion, this is surely what give them a big advantage over the 14FF of GF ( im not quite sure the problem is the process itself, but more that Nvidia have extremely well optimized each pieces of their chips for attain higher clock ).

Not increasing much the SP, but who can work at 500+mhz that your "competitor" do a big difference. Its an interessant move from Nvidia ( even if it have allready started with Maxwell ).

Both architectures can be realy close on "base performance", but if you have one who is able to hit 1.6x more clock speed, specially on today games, this will push clearly the difference.
Its where it will be interessant to take a TitanX (maxwell ) and see if OC to 1.8ghz what relative performance it obtain vs Titan Pascal
 
Last edited:

Call me crazy, but a picture of run-down industrial building is the last thing that comes to mind when someone says "imminent launch". But then again, launching a solid competitor to card "X " ~22 months after the fact when it already saturated the market AND alongside the successor to card "X" seemed like a strange strategy and yet here we are.
 
Call me crazy, but a picture of run-down industrial building is the last thing that comes to mind when someone says "imminent launch". But then again, launching a solid competitor to card "X " ~22 months after the fact when it already saturated the market AND alongside the successor to card "X" seemed like a strange strategy and yet here we are.

Yes it sure, this will maybe not be the image i will choose ( this said, i dont know where it have been taken ), Im not quite sure about what GPU you are talking about anyway.
 
I knew the spin was coming but it still shocks me that anyone could see this as anything other than a complete and total failure.
I though your post was a little rough till I got the power consumption figure for the card, imho AMD pushed the clocks too hard. Leaving Nvidia upcoming line of product aside, the card should have beat the GT 750 Ti at its game (low power consumption), it is an old card that Nv have not refreshed since launch. It does not as it burns significantly more power. I'm sure it could be tweaked to operate within the constraint Nvidia set for the GT 750 Ti, its performances might be greater but by how much?
I think that one again AMD did not got the specs of its card right. They needed to replace the Cap Verde/Bonaire cards and land a clean win on Nvidia line of GPU which can operate out of PCI power and are good entry level gaming laptop GPU. It was important strike the good balance as the aforementioned parts are to be replaced. In that context it was not necessary to try to get that close fro Nvidia higher GPU, neither to try to achieve for example higher geometry throughput (I suspect the GT 750 replacement will hardly be a monster on that account (1 GPC most likely) and it will be perfectly fine for its usage case and power usage).
What AMD needed was a cheap card and stick to the PCI power requirement: they failed on those accounts.
They are selling a good card to budget gamers though I wonder how much one card makes them to the bank compare to older generation Pitcairn derivatives for example

For me they failed the most important part aka reach a segment in which they do not exist.
 
Last edited:
There is not any third-party benchmark could proof the “aggressive primitive culling” that AMD said introduced in Polaris so far.
 
Not much a problem. We had GCN first information way before the release ( around august 2011, when 7970 have been released in early 2012 )..

First "presentation " we had the name and roadmap", sorry but i find AMD really slow to give name of new sku, features and roadmap compared to Nvidia ..

March 2014 ... Volta... Stacked Vram, Nvlink etc...
http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

March 2015 " Pascal enter, push volta in the roadmap : features shown and announced mixed precision, HBM, NvLink, etc etc. 10x performance over Maxwell ( in mixed precision ofc ). ooh he even show a prototype.

http://www.extremetech.com/gaming/2...mance-gains-from-upcoming-pascal-architecture

I dont say i like the way it is made, but its bit how things work for this industry.


March 2013 - Volta and stacked DRAM:

http://www.anandtech.com/show/6846/nvidia-updates-gpu-roadmap-announces-volta-family-for-beyond-2014

 
Status
Not open for further replies.
Back
Top