Vega/Navi Rumors (Updated)

[Crumpet]

A random "leak" has appeared:

https://www.reddit.com/r/Amd/comments/6aeat3/full_vega_lineup_and_release_date_revealed/

The random leak has since been deleted and the poster removed.

Seems unusual considering most random rumours get completely ignored.

Genuine leak perhaps?

 

[Glo.]

The random leak has since been deleted and the poster removed.

Seems unusual considering most random rumours get completely ignored.

Genuine leak perhaps?

All we can do is wait and see.

 

[Crumpet]

All we can do is wait and see.

Well if the rumoured speeds and prices of that post were true, i'd be a buyer for sure.

 

[Glo.]

Well if the rumoured speeds and prices of that post were true, i'd be a buyer for sure.

I don't think the performance levels were correct. But this is just my personal opinion.

 

[Crumpet]

I don't think the performance levels were correct. But this is just my personal opinion.

I personally see them as a bit high, for my own personal expectations.

But I certainly wouldn't complain if it went the other way.

 

[davide445]

Didn't read the long thread so maybe the topic was already discussed.
I'm wondering why AMD did delayed so much Vega delivery on the market.
If as is written somewhere there is a problem with HBM memory supply I'm wondering why AMD is relying only on that and didn't add i.e. a GDDR5X version, considering the same problem was with latest Fury GPU.
I understood also AMD might be sticking with HBM since GCN arch like big bus instead of high frequencies, but with a bigger production cost, so interested if anyone have idea about the bus/frequency tradeoff AMD was focused on.

 

[T1beriu]

Because they are poor, not having the R&D budget to keep up with the competition is the short answer.

 

[Veradun]

Didn't read the long thread so maybe the topic was already discussed.
I'm wondering why AMD did delayed so much Vega delivery on the market.
If as is written somewhere there is a problem with HBM memory supply I'm wondering why AMD is relying only on that and didn't add i.e. a GDDR5X version, considering the same problem was with latest Fury GPU.
I understood also AMD might be sticking with HBM since GCN arch like big bus instead of high frequencies, but with a bigger production cost, so interested if anyone have idea about the bus/frequency tradeoff AMD was focused on.

They are not on the market yet because they will be when the product is ready (obviously).

As for redesigning with GDDR5x in mind you have to take into account a major delay in that case, as you would need a new tapeout and so on.

I'm not even sure GDDR5x has enough volume to feed AMD also, and btw I have no clue as to costs (there is a lot of chat about HBM high costs, but nobody compared the costs scientifically and might just be FUD, expecially considering GDDR5x surely costs more than GDDR5).

 

[Tweak155]

I don't think the performance levels were correct. But this is just my personal opinion.

Do you think they were short or high? If short, I'd be super glad I waited. If higher than what they truly are, it's a bit disappointing but at least the prices seem right.

 

[Glo.]

Do you think they were short or high? If short, I'd be super glad I waited. If higher than what they truly are, it's a bit disappointing but at least the prices seem right.

It depends how big improvement in IPC is Vega over previous generations of AMD GPUs.

 

[Tweak155]

It depends how big improvement in IPC is Vega over previous generations of AMD GPUs.

Thank you for your non-answer

Here's for hoping

 

[Glo.]

Thank you for your non-answer

Here's for hoping

I do not want to spread false(potentially) information.

 

[SpaceBeer]

I really doubt Vega delay* is related to HBM2. It's been in production for more than a year (P100). So for sure there are more than ~30.000 stacks available. It is just improvement over HBM1, not completely new technology, so it's not so harder or more expensive to make. HBM has been in consumer products for 2 years now. It is not new technology any more, from manufacturing viewpoint

I also don't think it's a bad decesion to go with HBM2 instead of GDRR5X or GDDR6. If we go back to this comparation
http://electroiq.com/insights-from-leading-edge/wp-content/uploads/sites/4/2016/03/Sys-plus-1.jpg
it is only ~40$ more expensive to have it instead of GDDR5. Maybe even less at this moment. Which is totaly ok for $400+ cards. So yes, earnings will be little bit lower, but you can have better performance with smaller amount of memory (with HBCC), lower board TDP (10-20W), smaller cards (SFF), new technology (useful for marketing). And of course if/when HBM becomes as cheap as GDDR (and there will be some low-cost stacks soon), you could use it in mid-range products.

*In August 2016 AMD anounnced Vega will be launched in H1/2017. Which means Q2/2017, otherwise, they would say Q1

 

[Tweak155]

I do not want to spread false(potentially) information.

Just asking for an opinion, no one holding you to it... imagine if that were the case, how many people would be banned? Haha. Everything is speculation until we get the reviews if you ask me / word from official sources. We're just posters on a forum.

 

[Valantar]

Didn't read the long thread so maybe the topic was already discussed.
I'm wondering why AMD did delayed so much Vega delivery on the market.
If as is written somewhere there is a problem with HBM memory supply I'm wondering why AMD is relying only on that and didn't add i.e. a GDDR5X version, considering the same problem was with latest Fury GPU.
I understood also AMD might be sticking with HBM since GCN arch like big bus instead of high frequencies, but with a bigger production cost, so interested if anyone have idea about the bus/frequency tradeoff AMD was focused on.

Including a GDDR5 memory interface with sufficient bandwidth on the die 'just in case' would very significantly affect die size. Which would again make the chip far more expensive to produce, no matter the cost of the memory. Remember, even a 256-bit GDDR5(X) interface takes up 8-10 times the die area of a 2-stack (2048-bit) HBM2 interface. Enough to seriously lower chips per wafer, and thus significantly increase chop production costs and complexity.

 

[nad-]

Could someone fill me in why AMD hasn't released a FP64 GPU since Hawaii/Tahiti? I thought compared to Nvidia they didn't have dedicated FP64 units so it's only a matter of firmware, or do their FP64 units need bigger registers? Fiji at 1/2 would be pretty potent, it would still be good even at at 1/4 rate.

 

[Veradun]

Could someone fill me in why AMD hasn't released a FP64 GPU since Hawaii/Tahiti? I thought compared to Nvidia they didn't have dedicated FP64 units so it's only a matter of firmware, or do their FP64 units need bigger registers? Fiji at 1/2 would be pretty potent, it would still be good even at at 1/4 rate.

No HBM volumes at the time to make Fiji a candidate for datacenters?

 

[exquisitechar]

I personally see them as a bit high, for my own personal expectations.

But I certainly wouldn't complain if it went the other way.

GTX 1070 performance for 400$...doesn't seem very high performance to me.

If the leak is true, then the only great Vega card will be the flagship with GTX 1080ti or higher performance.

 

[Qwertilot]

Could someone fill me in why AMD hasn't released a FP64 GPU since Hawaii/Tahiti? I thought compared to Nvidia they didn't have dedicated FP64 units so it's only a matter of firmware, or do their FP64 units need bigger registers? Fiji at 1/2 would be pretty potent, it would still be good even at at 1/4 rate.

Likely the same reason they released half a product range with polaris/roughly the same with Fiji or even going back why Tonga was such a strange card in terms of availability/SKUs etc. They quite clearly haven't had the money required to do even all of the basics they'd like to, let alone anything vaguely speculative.

 

[Krteq]

Let's hope

Computex 2017 is fast approaching so we wanted to share a save-the-date for the AMD press conference, scheduled for May 31st from 10 a.m. – 11 a.m.

Hosted by AMD CEO, Dr. Lisa Su and other key AMD executives, you will have the opportunity to hear more about the latest products and leading-edge technologies coming from AMD in 2017. The past year has seen AMD bringing innovation and competition back to the high-performance desktop market with the release of Ryzen™ processors and we look forward to providing new details on 2017 products and the ecosystems, both OEM and channel, that will support them.

VideoCardz - AMD to hold a press conference on May 31st

 

[Zstream]

M

as a happily whipped man I may use the word committee in the following way.

"I'd like to request the committee to approve my purchase of a Vega 10.

Thank you dear. "

What was the leak?

 

[Glo.]

So the leak was after all correct, at least on dates.

 

[Krteq]

Let's wait for 16th May, if there will be some Vega teaser, that guy on reddit was right.

 

[Gatecrasher3]

But really, lets say Vega is a flop (no faster then 1080 for around the same cost), what should AMD do next in regards to their GPUs? Ask for a bit more R&D money from the tentative success of Ryzen?

 

[CatMerc]

Could someone fill me in why AMD hasn't released a FP64 GPU since Hawaii/Tahiti? I thought compared to Nvidia they didn't have dedicated FP64 units so it's only a matter of firmware, or do their FP64 units need bigger registers? Fiji at 1/2 would be pretty potent, it would still be good even at at 1/4 rate.

They need to pack extra hardware for FP64. It's not separate cores, but it's not simply a matter of FP32 + FP32 = FP64.