Vega/Navi Rumors (Updated)

[Glo.]

The GPU at least in deep learning can be competitive(extremely) against GP100 and GV100 while having smaller die size, and consuming less power.

 

[GoodRevrnd]

I wonder if they were ready to launch Vega and had to do some additional tweaking and binning to compete with Ti and possibly Volta expectations.

 

[Crumpet]

What's up with Frontier Edition? Is it some play off of Founder's Edition or a jab at NVidia?

The benchmark results are nice I guess, but who knows how representative that result is.

Frontier Edition is for machine learning.

It's called frontier edition because it's their first proper step forward in machine learning.

 

[Glo.]

That's how it looks... And that don't look good

Deep Learning is way more important considering the margins than consumer market. Hence the announcement. Consumer GPUs are on track for Q2 launch, IMO.

 

[PhonakV30]

So Nvidia brought Volta because of Vega's deep learning ?

 

[tviceman]

The GPU at least in deep learning can be competitive(extremely) against GP100 and GV100 while having smaller die size, and consuming less power.

Did they say what the die size and power consumption of Vega is?

 

[zinfamous]

I am certain that Raja's shirt + jacket coordinating with the color scheme on that card is no mistake.

 

[Anarchist Mae]

Basically some Vega GPU averaged the same framerate in Sniper Elite 4, as stock GTX 1080 Ti. At least according to TPU scores.

And it was limited to 2GB of ram...

 

[krumme]

Amd using 8 engineers vs nvidia investing 3b
....ehh. i will believe it when it happens.

The apu was interesting though. Vs prior gen we get a 50% cpu uplift and 40% gpu uplift at half the power.
Thats massive. Bodes well.

 

[Crumpet]

And it was limited to 2GB of ram...

No, it wasn't.

Previous demos during the show were, that vega SE4 demo was running 8gb.

 

[Samwell]

Deep Learning is way more important considering the margins than consumer market. Hence the announcement. Consumer GPUs are on track for Q2 launch, IMO.

They never mentioned anything about timeline for consumer product, whereas they said FE comes late june. I have the feeling consumer is Q3.

 

[krumme]

Raja is so cool and funny. He is a brain. A galizilion times more reflective than jim or even papermaster.
A savior for radeon. Much needed capacity. But i think they need 3 or 4 years more to get competitive. Nv is on a roll.

 

[PhonakV30]

No, it wasn't.

Previous demos during the show were, that vega SE4 demo was running 8gb.

It was on Card with 8GB but consumes only 2 GB , right?

 

[T1beriu]

And it was limited to 2GB of ram...

It was on 8 GB but consumes only 2 GB , right?

No. The 2GB story was when they did a demo with HBCC running on just 2GB to show the improvement that you don't need tons of GB to run a game. But this needs developer support, it's not plug & play.

Amd using 8 engineers vs nvidia investing 3b
....ehh.

You're joking, right? He was talking about 8 engineers needed to optimize the software stack for tensor flow, if I'm not mistaking. Radeon group has 3000+ e̶n̶g̶i̶n̶e̶e̶r̶s̶ (LE: employees).

 

[OatisCampbell]

I wouldn't bet too hard on that. I've spent quite some time on these forums trying to convince people that replacing their 1080 with a 1080Ti while they're still on a 1080p60 panel is idiotic. Believe me, not everyone knows their own good.
What I said was that "competing" at the 1080Ti level or above with a dual GPU card in 2017 is suicide. AMD already has a reputation for being the hot-running, buggy, problematic, OC-less, in-need-of-an-AIO GPU brand. A dual GPU flagship card would just cement that reputation, and would hurt AMD far more than a high-end Vega chip that is "just" competitive with the 1080Ti. "Competing" with a dual GPU card is not competing at all, but rather a clear-cut admittance that they can't compete at the high end. It would be the equivalent to Raja Koduri going on stage at the Volta consumer launch and telling Jensen Huang "you know what, you've beat us. Fair and square." Now, something tells me this wouldn't play out well for AMD.

1. I don't share your contempt for tech forum readers and posters. I believe most people reading this are smart enough to know they don't need $500 cards for 1080p or the lower 1440p at 60Hz.
2. I also don't share your contempt for AMD. My last two cards were a 290 and a 7970 and they both served me very well. Probably the only reason I don't have a Vega is there's no Vega to have. Would it be admitting they can't produce a single chip as good as the new Titan Xp? Sure. I don't think most of us Vega will compete with new Titan Xp. However; that card is $1200.. Would a Vega CF card at $900 that beats it in games that scale be welcome in the market? I would think so. I also think AMD has a history of doing this, and believe the card would sell. (at $799 it might sell very well and steal some 1080Ti sales)

 

[Snarf Snarf]

I'm so bummed, I was streaming the event happily at work and then our tech decided we needed to fix a problem with our PXE server and took our network down for the exact amount of time Raja was on stage

 

[Krteq]

No. The 2GB story was when they did a demo with HBCC running on just 2GB to show the improvement that you don't need tons of GB to run a game. But this needs developer support, it's not plug & play.

Well, I don't think HBCC needs any SW modification (driver only)

We had a demonstration of HBCC on DE:MD at February and now on RotTR. Have been those games modified? I quess not.

 

[Crumpet]

I'm so bummed, I was streaming the event happily at work and then our tech decided we needed to fix a problem with our PXE server and took our network down for the exact amount of time Raja was on stage

Nvidia bias confirmed.

Hahahha, hard luck man, don't worry we'll keep you informed.

 

[Malogeek]

They never mentioned anything about timeline for consumer product, whereas they said FE comes late june. I have the feeling consumer is Q3.

Agreed. If consumer card was coming in June they would have said so.

 

[Bouowmx]

2×8-pin power, up to 375 W, for 13 TFLOPS: not sure whether over-specced like after-market products, or actual power requirements.

 

[Malogeek]

You're joking, right? He was talking about 8 engineers needed to optimize the software stack for tensor flow, if I'm not mistaking. Radeon group has 3000+ engineers.

They don't have 3000 engineers, they have 3000+ employees.

 

[T1beriu]

Well, I don't think HBCC needs any SW modification (driver only)

We had a demonstration of HBCC on DE:MD at February and now on RotTR. Have been those games modified? I quess not.

That's what Raja said in this interview.

https://www.youtube.com/watch?v=FwcUMZLvjYw

 

[Karnak]

http://pro.radeon.com/en-us/vega-frontier-edition/

 

[Valantar]

1. I don't share your contempt for tech forum readers and posters. I believe most people reading this are smart enough to know they don't need $500 cards for 1080p or the lower 1440p at 60Hz.
2. I also don't share your contempt for AMD. My last two cards were a 290 and a 7970 and they both served me very well. Probably the only reason I don't have a Vega is there's no Vega to have. Would it be admitting they can't produce a single chip as good as the new Titan Xp? Sure. I don't think most of us Vega will compete with new Titan Xp. However; that card is $1200.. Would a Vega CF card at $900 that beats it in games that scale be welcome in the market? I would think so. I also think AMD has a history of doing this, and believe the card would sell. (at $799 it might sell very well and steal some 1080Ti sales)

Wow, dude, chill out.

I don't have an ounce of contempt for the general public here (or on any other forum/site I frequent - if I did, I wouldn't stay), but I've been here long enough to know that there are quite a few screwballs around, and of many different kinds. That's just humanity for you. I do have some odd ideas of my own, when compared to the general public here, and don't consider that a problem at all. It's what makes for interesting discussion. Still, I (and many/most others here, from my experience) spend a bit of our time trying to guide others a bit. That's what forums are for, after all.

Also, contempt for AMD? What on earth are you on about? Are you seriously denying that AMD has (and has had) that kind of reputation? If so, how do you explain Nvidia's history of dominating marketshare even in times when they have had clearly inferior products? I mean, I could have added "(undeserved)" into that sentence, but for some reason I expect the people reading my posts to be able to tell the difference between commenting on a phenomenon that exists in society and expressing the same as an opinion. A clue: I was doing the first, not the latter. Now please shut up, and leave your straw men lying where they are. Or just work on your reading comprehension.

 

[Bacon1]

2×8-pin power, up to 375 W, for 13 TFLOPS: not sure whether over-specced like after-market products, or actual power requirements.

Fury X has 2x8pin and used 220w: http://www.tomshardware.com/reviews/amd-radeon-r9-fury-x,4196-7.html