AMD to Launch First "Polaris" Graphics Cards by Late May

[medi01]

. The Polaris 10 card that was demoed back then was the pro card, not the XT, because GDDR5X started mass production in mid March, so it wasn't available then.

So, according to your theory, which card was it:

1) 480
2) 480x
3) 490
4) 490x

?

 

[TheinsanegamerN]

Well, possibly.
If you are talking about 480/480x.
But if they plan to release 490/490x too (which, afaik, they do) staying on 390x levels would be a huge disappointment (and fuck lower power consumption)

You may not care about power consumption, but a lot of people do. The 980 was not significantly faster than the 780ti, but pulled less power, and it sold like gangbusters. AMD did not lower power consumption, and they lost a ton of market share. Getting the picture?

390x/980 levels of performance are perfectly good for 1080p, which more than half the market is at or under, and quite good for 1440p. very little of the market is above that rez.

Mid range chips hitting flagship performance levels several years later is pretty common. Nvidia does it too. Considering polaris 10 is a mid range chip, they SHOULD be targeting the mainstream resolution. And if the general market wants lower TDP chips, then AMD should make them.

Not to mention, why on earth are you concerned with amd targeting TDP on a mainstream chip? You are the market for cards like Vega, not polaris.

 

[bug]

I think AMD has done an good job with Polaris and have fixed many of the bottlenecks that were present in the original GCN micro-architecture.

Despite the fact that you, like everyone else, don't have a clue about Polaris. All we have atm is some sparse info from AMD (cherry-picked, most likely) and a statement Polaris 10 is aimed at the mid-range.
Seriously, is actually having someone review a product no longer a requirement for someone else to decide whether it's good or bad?

 

[JMccovery]

It's unfortunate for AMD that they have such stupid management that decided to not release Fury/FuryX replacements this year and let Nvidia dominate the high-end GFX market. What they didn't account for is Nvidia dominating the mid-range GFX market too.

Ever think the reason could be that HBM2 isn't available in massive quantities yet?

Would there be any point in releasing Vega with HBM1?

 

[G33k2Fr34k]

Ever think the reason could be that HBM2 isn't available in massive quantities yet?

Would there be any point in releasing Vega with HBM1?

What happened to GDDR5X?

 

[medi01]

You may not care about power consumption, but a lot of people do. The 980 was not significantly faster than the 780ti, but pulled less power, and it sold like gangbusters. AMD did not lower power consumption, and they lost a ton of market share. Getting the picture?

390x/980 levels of performance are perfectly good for 1080p, which more than half the market is at or under, and quite good for 1440p. very little of the market is above that rez.

Mid range chips hitting flagship performance levels several years later is pretty common. Nvidia does it too. Considering polaris 10 is a mid range chip, they SHOULD be targeting the mainstream resolution. And if the general market wants lower TDP chips, then AMD should make them.

Not to mention, why on earth are you concerned with amd targeting TDP on a mainstream chip? You are the market for cards like Vega, not polaris.

I do care about power consumption.
But I don't see "same performance as last gen, but much lower power consumption" as a viable strategy today for AMD. They must up performance too, or they are toast on PC market. 480 > 390, 490 > 1070 would be enough.

 

[rruff]

They must up performance too, or they are toast on PC market. 480 > 390, 490 > 1070 would be enough.

I'd put equal signs on those, and frankly it wouldn't be a disaster if they missed those targets. They'd just have to price accordingly.

I doubt the 490 is coming out soon but I guess we will find out next week.

 

[Kanan]

Back in Early March, AMD demoed a Polaris 10 GFX running the new Hitman game in DX12 mode at 2k on "highest settings". The performance was consistently above 60 fps. The Polaris 10 card that was demoed back then was the pro card, not the XT, because GDDR5X started mass production in mid March, so it wasn't available then.

The only AMD cards that score above 60fps are the 390X, Fury and FuryX. This means that Polaris 10 Pro is at least as fast as the 390X. That is an excellent performance level for a ~230mm^2 chip running at ~1GHz.
I think AMD has done an good job with Polaris and have fixed many of the bottlenecks that were present in the original GCN micro-architecture. We're talking about a card that has 80% (36/44) of the SP resources of the 390X and similar clocks speeds. That's at least a 25% increase in shader efficiency going from Hawaii to Polaris. I expect the Polaris 10 XT to get dangerously close to the FuryX.

Yes, I don't really have a different opinion here. The 480/480X must be at least as fast as 390X, that means, they can nearly reach Fury/X speeds, or even be faster. I'm just saying, these small chips are not the succesor of the 390/390X - the succesor for 390/390X will be a equally sized, highend chip, which will easily be faster than 980 Ti (comparable at least to 1080 in speeds), with a lot more shaders on it too. You have to keep sizes of the GPUs in mind and shader amounts. 2560 shaders is a good replacement to a 390/390X but not to a 4096 shader card (with 596mm² chip) like Fiji (Fury X). Also Vega should be a lot faster than Fury X, not only 20% or so. I'm thinking about 50-100%, or even more.

Again, I'm just speculating here but I have my reasons.

Ever think the reason could be that HBM2 isn't available in massive quantities yet?

Would there be any point in releasing Vega with HBM1?

No, HBM1 is limited to 4GB Vram which is clearly not enough for the upcoming cards, wasn't even really good for the Fury X. I think HBM2 is available in enough quantities, but it's simply only needed on the fastest GPUs which aren't done yet (Vega). See the new GP100 Quadro card of Nvidia, which uses HBM2 memory (cost: 10.000$). Big 16nm chips are extremely expensive now, because they are simply hard to produce, that's why both companys are starting with smaller chips for the consumer market.

I do care about power consumption.
But I don't see "same performance as last gen, but much lower power consumption" as a viable strategy today for AMD. They must up performance too, or they are toast on PC market. 480 > 390, 490 > 1070 would be enough.

The performance/semi highend section is very important for sales, afaik even more important than highend sales. So if AMD goes in that direction first, and then does highend cards, it's a viable strategy. As I see it, they are exactly doing that.

 

[medi01]

The performance/semi highend section is very important for sales, afaik even more important than highend sales. So if AMD goes in that direction first, and then does highend cards, it's a viable strategy. As I see it, they are exactly doing that.

Indeed. but there is the problem, where does that segment end? I'd say 970 (inclusive). And that's pretty close to 1070. And if what AMD rolls out is slower than 1070, they'd need to price it SIGNIFICANTLY cheaper, for it to sell well. (because FuckYouWorks, DriverFUD et al)

 

[arbiter]

No, HBM1 is limited to 4GB Vram which is clearly not enough for the upcoming cards, wasn't even really good for the Fury X. I think HBM2 is available in enough quantities, but it's simply only needed on the fastest GPUs which aren't done yet (Vega). See the new GP100 Quadro card of Nvidia, which uses HBM2 memory (cost: 10.000$). Big 16nm chips are extremely expensive now, because they are simply hard to produce, that's why both companys are starting with smaller chips for the consumer market.

If they were making a mid range card maybe HBM1 would work ok if it was for 1080p graphic's but if gpu targets 1440p or even 4k, need least 8gb .

 

[Kanan]

Indeed. but there is the problem, where does that segment end? I'd say 970 (inclusive). And that's pretty close to 1070. And if what AMD rolls out is slower than 1070, they'd need to price it SIGNIFICANTLY cheaper, for it to sell well. (because FuckYouWorks, DriverFUD et al)

If the 1070 is "faster than Titan X" (just quoting) it's a lot faster than GTX 970, not so close then. Well, let's wait and see what the AMD chips can do, speculating about them is over - for me at least.

If they were making a mid range card maybe HBM1 would work ok if it was for 1080p graphic's but if gpu targets 1440p or even 4k, need least 8gb .

I think the Fury X / Fury / Nano are okay for 1440p, but 4K they are 1) too slow, 2) on the long run the Ram will not be enough to drive that high resolution. So I think they are decent 1440p cards, nothing more. The reason to use HBM1 was only for the power savings, because Fury X with GDDR5 would have used over 325-350W - that's simply too much. They wanted a card that is modern by any means, because of PR and prestige. They knew it would be expensive (for themselves, meaning less $ earned per card) and a hard sell compared to other cards, but they did it nonetheless.

 

[arbiter]

The reason to use HBM1 was only for the power savings, because Fury X with GDDR5 would have used over 325-350W

Power draw wouldn't be that high don't pull amd marketing exaggeration.

 

[medi01]

If the 1070 is "faster than Titan X" (just quoting)

In original quote, it's more than 2 times faster than Titan X at certain "VR" things.

So I think they are decent 1440p cards, nothing more.

So is 1080. Heck, that Doom demo was run at 1080p.
We aren't there yet. Maybe in 2017.

 

[arbiter]

So is 1080. Heck, that Doom demo was run at 1080p.
We aren't there yet. Maybe in 2017.

If doom was at all the highest settings possible, that would mean probably least 4x AA or even 8x AA. Turn that off 4k should be possible on it at pretty good fps.

 

[Kanan]

Power draw wouldn't be that high don't pull amd marketing exaggeration.

I don't pull, and hadn't to do with any marketing, I made that up myself, it's a good educated guess, nothing more. Fury X now has a TDP of about 250-275W, with GDDR5 it would be a lot higher, so what I said before holds true for me.

In original quote, it's more than 2 times faster than Titan X at certain "VR" things.

I don't care about VR. I think GTX 1070 will be slower than 980 Ti... custom. Maybe it's a tad faster than 980 Ti ref models, but you never know, that's only Nvidia marketing that I don't care much about. Let's wait and see.

 

[arbiter]

I don't pull, and hadn't to do with any marketing, I made that up myself, it's a good educated guess, nothing more. Fury X now has a TDP of about 250-275W, with GDDR5 it would be a lot higher, so what I said before holds true for me.

Educated guess, yea aka based off of nothing. HBM could have same draw as gddr5 for all you know so don't claim something less you have facts and proof to back that claim.

 

[Kanan]

Educated guess, yea aka based off of nothing. HBM could have same draw as gddr5 for all you know so don't claim something less you have facts and proof to back that claim.

Go read up HBM informations, it's a fact and easily understood because HBM runs on much lower frequencies. You need proof for the easiest things on earth? Well ... I don't have the time, sorry. What I say is a good guess, and it's NOT based on nothing, lol. Again, go read up HBM informations, I'm not your teacher.

PS. Just go read up a Fury X review, for that matter. The "facts" you seek are included there.

PPS. I'm even so nice as to find one for you. "Facts":
http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-review/5

 

[john_]

Well, I guess AMD missed a nice opportunity AGAIN. With Polaris targeting, in theory, the market that is around $300, they would be playing ball alone. Now they are losing this opportunity and they are giving Nvidia the chance, not only to sell more 1070's, but also to have enough time to prepare 1060 for the market.

AMD Polaris Tech Day: NDA ends on June 29th | VideoCardz.com

 

[Fluffmeister]

Another month? Ouch.

Wonder if something is brewing, Kyle over @ HardOCP has posted a pretty scathing article today about AMD:

http://hardocp.com/article/2016/05/27/from_ati_to_amd_back_journey_in_futility#.V0huzEwzqCg

People will say it's just becuase he wasn't invited by i wonder if at least most of what he says is true, guess we still have time to ponder either way.

 

[john_]

Another month? Ouch.

Wonder if something is brewing, Kyle over @ HardOCP has posted a pretty scathing article today about AMD:

http://hardocp.com/article/2016/05/27/from_ati_to_amd_back_journey_in_futility#.V0huzEwzqCg

People will say it's just becuase he wasn't invited by i wonder if at least most of what he says is true, guess we still have time to ponder either way.

Hardocp is considered a very very friendly to Nvidia site. In a way, he almost admits it in the end of the article, not directly of course.
I guess when you are going to offer cards one month latter than what most people believe, you don't call to the event someone that you think is closer to the competition. And of course we all know that AMD changed it's attitude toward the press from last year.

Anyway, the article reminds me many articles about politics. Authors supporting political party A, always find backstabbing behaviors between the top leaders of competing political party B. That's what he is doing in there.

 

[Fluffmeister]

Hardocp is considered a very very friendly to Nvidia site. In a way, he almost admits it in the end of the article, not directly of course.
I guess when you are going to offer cards one month latter than what most people believe, you don't call to the event someone that you think is closer to the competition. And of course we all know that AMD changed it's attitude toward the press from last year.

Anyway, the article reminds me many articles about politics. Authors supporting political party A, always find backstabbing behaviors between the top leaders of competing political party B. That's what he is doing in there.

Well it will be pretty easy to say he has egg on his face when AMD release an undeniable winner and everything he claimed was just plain wrong.

Exciting times ahead either way.

 

[john_]

Well it will be pretty easy to say he has egg on his face when AMD release an undeniable winner and everything he claimed was just plain wrong.

Exciting times ahead either way.

Most say that Polaris will be a 230mm chip. So, if AMD didn't just miraculously fixed an under performing for 3 generation GCN architecture, turning it in version 4.0 into something with much higher performance at the same clock speeds, it's an easy bet that Polaris was never going to compete with GP104. To say the truth that's where he lost me in his article, much earlier that the theories about Raja stealing RTG from Su and running to Intel with the spoils.
I do agree with one part in his article. AMD wasn't expecting a 2GHz Maxwell. They where surprised. I think we all did, no matter what company we prefer, Nvidia and TSMC delivered more than anyone was expecting..

 

[cdawall]

ermm arent you two agreeing now?
He said 50/60% better performance then the GTX980, lets look at the Witcher 3 there...yep about 60 - 70% better.
The VR info in the lower left is some extra information regarding the first "Virtual Reality" performance comparison.

He is, just arguing his agreement I guess. That happens a lot in these threads...

 

[Fluffmeister]

Most say that Polaris will be a 230mm chip. So, if AMD didn't just miraculously fixed an under performing for 3 generation GCN architecture, turning it in version 4.0 into something with much higher performance at the same clock speeds, it's an easy bet that Polaris was never going to compete with GP104. To say the truth that's where he lost me in his article, much earlier that the theories about Raja stealing RTG from Su and running to Intel with the spoils.
I do agree with one part in his article. AMD wasn't expecting a 2GHz Maxwell. They where surprised. I think we all did, no matter what company we prefer, Nvidia and TSMC delivered more than anyone was expecting..

Well yeah I know, Roy Taylor has been banging the TAM drum, saying they are targeting a completely different market and that's fine.

If people are happy to have similar performance they can get today but in a cheap but fast and efficient package then that's great too.

And i agree with that one part of his article too, with the apparently expensive FE (HardOCP berated them for that as well) selling out everywhere nVidia's more traditional high end market is covered and already happily gobbling up supply. But then as you rightfully pointed out, Nvidia won't be ignoring the mainstream market forever either.

 

[john_]

Nvidia will keep 1060 as long as it needs to sell enough 1070s and 1080s to secure a high enough profit margin and high enough income. With AMD losing a month that means faster sales of 1070s and 1080s at the prices Nvidia wants. That also means that Nvidia can bring 1060 in the market faster without having seconds thoughts about that. AMD is shooting itself in both feet.