AMD: Speculation, Rumors, and Discussion (Archive)

No surprises there.

The only question now is how a Polaris 11 will stack up against a GP104.

 

Apparently Polaris 10 is the one that had its first ever demonstration, going by the press about the event.
The marketing's use of a decently old 28nm baseline may cap expectations somewhat before Vega. Although if a card gets 2-2.5x Tonga's efficiency and also gets the same 190W envelope, it does put some interesting numbers in the relatively few benchmarks that have both the 285 and the 390/Fury boards.

 

Apparently Polaris 10 had been shown before too, but behind closed doors (was it CES or where they showed Polaris 11 for the first time? same event for Polaris 10 in secret)

 

I have specifically been told that Polaris 11 is the small GPU, and consequently is what we saw in December. Polaris 10 is the larger of the two, and was being used to render some of tonight's demos, which were running at 2560x1440.

 

So Polaris 10 is the Hawaii replacement and Polaris 11 is the mobile gpu we saw being compared against a GTX 950?

 

Then Polaris will have HBM1 or GDDR5/X?. If Vega is the big chip I supposse Polaris could be only GDRR5/X like GP104.

And what´s NAVI´s next gen memory?. What could be more next gen than HBM2?. Playstation 5 chip? .

 

why stream a GDC conference?
Bad PR for the customers.

No big flagship coming anytime soon.
while the Polaris is aimed for a VR enabled experience thats a though sell.
As far I can tell only the die shrink with power envelope is actually making Polaris interesting.

People buy speed when it comes to GPU.
AMD is doing nothing to change that....and while VR needs low latency I guess the buyer base seems not to go for that when they buy as far...yet.

if no one buys your stuff then soon your out of a job.
Drunk engineers screaming an overclockers dream dont help AMD either.
and besides who wanna hire people who sank a company?

 

Isn't Radeon Pro Duo, which was introduced at the show, a "big flagship"?

 

Seriously, how do people expect GDDR5X memories from products launching before GDDR5X mass production?

 

If it´s GDDR5 then meh. Will have to wait for Vega for the real next gen architecture.

Anyway, didn´t Fury launch with HBM1 when HBM1 wasn´t even in mass production?. That takes me to another question, do you think we can consider HBM1 sales in AMD products as a "mass production" volume?.

 

Because we've been "stuck" in 28nm for quite some time now and even people who don't understand how typical production works when transitioning to a new node want new GPUs, basically believing any sort of rumor that's thrown their way. I'll be surprised if we see the flagship Pascal/Polaris this year, but then again i haven't been following the news closely

 

It launched just after HBM1 was announced to be in mass production. HBM1 had also been sampling for a lot longer than GDDR5X, if I remember right.
Having "new" or "old" memory type doesn't affect how "real next gen" some architecture is.

Well AMD did confirm Vega for late 2016/early 2017

 

It seems Polaris could use HBM1 after all:

http://www.pcper.com/news/Graphics-...past-CrossFire-smaller-GPU-dies-HBM2-and-more

 

Vega is probably just Polaris+HBM2. Serious architectural changes between it and Polaris would be odd considering the mere 6 month gap and AMD's tight financial situation.
Polaris does include substantial architectural advancements upon current-gen GCN, including significantly improved memory bandwidth compression which should reduce any bottleneck imposed by GDDR5. Considering the performance GM200 can offer with a 384bit GDDR5 interface I'm not too worried.

 

It was a kinda cack-handed 'launch' though:
'VR blah blah VR, VR authoring is hard, here's a card for pro VR authoring, VR, blah, 16TF, VR pro authoring, oh its also the most powerful gaming card, VR, blah, pro authoring, VR'

 

its a 6 month gap to our eyes but who knows how long amd has been sitting on the move to 14nm . The fury stuff was most likely a cheap way for them to try and catch up performance wise with NVidia

 

The way I understand Raja, Polaris 10 and 11 will use GDDR5. The whole HBM 'pipeline' is still too expensive.

And his comments about perf/$ and how pleased and surprised we will be, makes me believe, launch will be something similar to HD4000 series.

 

Reading this I understand the contrary, they want to recover part of the investment in HBM1:

Why is Polaris going to use HBM1? Raja pointed towards the extreme cost and expense of building the HBM ecosystem prepping the pipeline for the new memory technology as the culprit and AMD obviously wants to recoup some of that cost with another generation of GPU usage.

 

Lol, kinda reminds you of the Xbone launch doesn't it? TV, TV, TV, Call of Duty, Call of Duty, Call of Duty, TV, TV, halo, TV TV, TV.

I can only assume they de-emphasize the gaming aspect because they're supply limited (and possibly selling for poor profit margin.) $1.5k is a fat chunk of change, but Nvidia has sold single-GPU cards for that much. This is dual-GPU, watercooled, stacked memory... NV pricing for the same hardware would be like, $3.5k.

 

You have to watch the video. These are all Ryan's interpretations. HBM1 'pipeline' is no less expensive than HBM2.

It would also make 0 sense for AMD to limit itself to 4GB VRAM in 2016.