Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Fun fact 2.5 perf/watt is the same as 250% perf/w, or 150% extra performance/watt.
#percentagesdontlie
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
RUMOR: Radeon 480 to be priced to replace 380, similar output as 390.
- Thread starter Zion Halcyon
- Start date
Fun fact 2.5 perf/watt is the same as 250% perf/w, or 150% extra performance/watt.
#percentagesdontlie
TaintedSquirrel
[H]F Junkie
- Joined
- Aug 5, 2013
- Messages
- 12,707
300/2.5 = 120W.
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
300/2.5=(300x2)/5
TaintedSquirrel
[H]F Junkie
- Joined
- Aug 5, 2013
- Messages
- 12,707
Three hundred divided by two point five equals one-hundred and twenty watts.
Coincidentally this is the same reported TDP of the new Polaris chips. And using AMD's own estimated efficiency improvement, matches up with Hawaii/Grenada performance. Which also matches up to what people in this thread have been saying for weeks.
CONSPIRACY! Wake up sheeple.
Coincidentally this is the same reported TDP of the new Polaris chips. And using AMD's own estimated efficiency improvement, matches up with Hawaii/Grenada performance. Which also matches up to what people in this thread have been saying for weeks.
CONSPIRACY! Wake up sheeple.
I don't know. We haven't seen many leaks for GP106 yet. But if we extrapolate from Maxwell then we can see that GM204 is around 70% faster than GM206. With GP104 rumored to be as fast as the 980Ti if not faster, that would place extrapolated GP206 performance right at 390x levels or higher, at $200 or less (current price for GTX 960).
That'll put it squarely on Polaris 10 territory. And with NVIDIA pretty much having the whole performance/enthusiast segments for themselves, they can be very flexible with pricing in the mainstream/low end. Great news for us, but not so much for AMD I think. They desperately need cash.
That'll put it squarely on Polaris 10 territory. And with NVIDIA pretty much having the whole performance/enthusiast segments for themselves, they can be very flexible with pricing in the mainstream/low end. Great news for us, but not so much for AMD I think. They desperately need cash.
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
They have 3 custom design wins, and should be launching a very competitive midrange. Should be good for amd I think
It will be great for AMD's bottom-line. Just not particularly interesting for enthusiasts on forums like this one.
From what AMD actually said earlier, they will be positioning multi-GPU at the mid-range. Multiple tiny cheap power-efficient GPUs with high yields on the same card, fits perfectly with Polaris, doesn't it?
They will be pushing multi-GPU cards, and since Polaris 10 won't compete with GP104, it seems logical they will begin with this coming generation.
Then again, he also hints at a much larger enthusiast-focused "Polaris 12" later on, presumably coming before Vega. So maybe it just hasn't leaked yet. Or maybe he's blowing smoke, AMD has certainly done that before. Remember what they said about Fury overclocking.
He also strongly implied that Polaris would come with HBM1. So likely capped at 4GB per GPU.
AMD's Raja Koduri talks moving past CrossFire, smaller GPU dies, HBM2 and more. | PC Perspective
Also don't forget that multi-GPUs and even heterogenous multi-GPUs are directly supported in DirectX12, bypassing crossfire or SLI, with Explicit Multi-Adapter. Still very early, and nobody knows how it will actually work, but it is promising. And memory is pooled in EMA, so two GPUs with 4GB of HBM1 would give 8GB of usable memory in DX12 games.
Ashes of the Singularity Revisited: A Beta Look at DirectX 12 & Asynchronous Shading
From what AMD actually said earlier, they will be positioning multi-GPU at the mid-range. Multiple tiny cheap power-efficient GPUs with high yields on the same card, fits perfectly with Polaris, doesn't it?
They will be pushing multi-GPU cards, and since Polaris 10 won't compete with GP104, it seems logical they will begin with this coming generation.
Raja Koduri said:
Then again, he also hints at a much larger enthusiast-focused "Polaris 12" later on, presumably coming before Vega. So maybe it just hasn't leaked yet. Or maybe he's blowing smoke, AMD has certainly done that before. Remember what they said about Fury overclocking.
Raja Koduri said:
He also strongly implied that Polaris would come with HBM1. So likely capped at 4GB per GPU.
AMD's Raja Koduri talks moving past CrossFire, smaller GPU dies, HBM2 and more. | PC Perspective
Also don't forget that multi-GPUs and even heterogenous multi-GPUs are directly supported in DirectX12, bypassing crossfire or SLI, with Explicit Multi-Adapter. Still very early, and nobody knows how it will actually work, but it is promising. And memory is pooled in EMA, so two GPUs with 4GB of HBM1 would give 8GB of usable memory in DX12 games.
Ashes of the Singularity Revisited: A Beta Look at DirectX 12 & Asynchronous Shading
Last edited:
ah this is from Capsaiscin. Raja was talking about Polaris 10, the chip bigger than Polaris 11 (this is why the numbers don't mean anything)
Also stated in the interview, they mention Polaris 10 and 11 will scale across many segments I am presuming that he was focused on mobile and desktop.
He stated it again, first priority is VR that is their focus first *this is Polaris* for as many people as possible *mid range, volume sales*, then the enthusiast segment is their second focus *Vega.
Raja even stated, connect the dots, Polaris won't have HBM2. New tech is expensive, they will only bring it when the cost benefit is ready for mainstream.
Also stated in the interview, they mention Polaris 10 and 11 will scale across many segments I am presuming that he was focused on mobile and desktop.
He stated it again, first priority is VR that is their focus first *this is Polaris* for as many people as possible *mid range, volume sales*, then the enthusiast segment is their second focus *Vega.
Raja even stated, connect the dots, Polaris won't have HBM2. New tech is expensive, they will only bring it when the cost benefit is ready for mainstream.
Last edited:
Well, watch the whole thing. It seems like most people missed this interview. Lots of interesting stuff mentioned there.
You're right, he definitively stated that VR was their primary focus (and the minimum VR spec is 390/970-ish). I didn't catch him saying "the entire performance stack" in the interview; that might have been the PCper guy reading into what he was saying, or perhaps incorrectly paraphrasing.
HBM2 definitely isn't coming to Polaris, but he left HBM1 open, talking about their investment in the HBM pipeline.
You're right, he definitively stated that VR was their primary focus (and the minimum VR spec is 390/970-ish). I didn't catch him saying "the entire performance stack" in the interview; that might have been the PCper guy reading into what he was saying, or perhaps incorrectly paraphrasing.
HBM2 definitely isn't coming to Polaris, but he left HBM1 open, talking about their investment in the HBM pipeline.
Last edited:
Yeah, Raja was very clear on what their focus was at Capsaiscin, and the CFO stated pretty much the same thing he did. Unless they want to redfine the midrange performance metrics like what they did with the 4xxx line, which btw is highly unlikely as nV has a new architecture unlike the last legs of Tesla, I don't see that happening, nor are they will to cut into their margins to go into a price war with nV, Su wants the margins to stay healthy.
Last edited:
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Well timed edit, you had me confused
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,580
Hmmm Well it is WCCFTECH, but it looks like Polaris 10 is possibly at 980ti performance?
The Polaris 10 GPU is said to have 3DMark Firestrike Ultra performance around 4000 points which is about what a Radeon R9 Fury X and GeForce GTX 980 Ti score. By 4000 points, we don’t mean exactly 4000 but it’s actually quite a bit less but that’s the number we were told.
Read more: AMD Confirms Polaris GPU Positioning - Polaris 10 Aims The Mainstream Desktop/Notebook Market, Polaris 11 Targets Notebook Market
If this is the cut down version...imagine a full ALU polaris with GDDR5x.
The Polaris 10 GPU is said to have 3DMark Firestrike Ultra performance around 4000 points which is about what a Radeon R9 Fury X and GeForce GTX 980 Ti score. By 4000 points, we don’t mean exactly 4000 but it’s actually quite a bit less but that’s the number we were told.
Read more: AMD Confirms Polaris GPU Positioning - Polaris 10 Aims The Mainstream Desktop/Notebook Market, Polaris 11 Targets Notebook Market
If this is the cut down version...imagine a full ALU polaris with GDDR5x.
Last edited:
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,580
All I have to say is. 90% of the people were wrong on the 970 GTX performance. We just never know.
Of course Firestrike is not a game.
*cough*
Like I have been saying... changes to architecture + new process = power gains AND performance gains... 2560 ALUs performing better than 2816 of the 390x and almost as good as the 4096 of the Fury X.
Im not claiming to have any idea on how GPUs work, but some of the logic that was being spout off here was just baffling... ("die size is half of Hawaii, 2 confirmed cards shows an ALU count, they said "mainstream", so we can deduce what their highest performer will be like" wtf?)
Anyways, I expect Polaris 10 to see par-for-par performance with the 300/Fury series, with much better power and a shift down in price points (top card being $500-$600 instead of $650+)... Dual Fiji will remain #1 until Vega comes out, which will beat it with a single GPU... Unlikely a Dual Polaris, to ensure the Dual Fury sells their existing Fiji/hbm investment.
Still waiting for a 490 announcement, not sure if that will be Vega or Polaris, but I would still expect Polaris which would mean there is more to come
Like I have been saying... changes to architecture + new process = power gains AND performance gains... 2560 ALUs performing better than 2816 of the 390x and almost as good as the 4096 of the Fury X.
Im not claiming to have any idea on how GPUs work, but some of the logic that was being spout off here was just baffling... ("die size is half of Hawaii, 2 confirmed cards shows an ALU count, they said "mainstream", so we can deduce what their highest performer will be like" wtf?)
Anyways, I expect Polaris 10 to see par-for-par performance with the 300/Fury series, with much better power and a shift down in price points (top card being $500-$600 instead of $650+)... Dual Fiji will remain #1 until Vega comes out, which will beat it with a single GPU... Unlikely a Dual Polaris, to ensure the Dual Fury sells their existing Fiji/hbm investment.
Still waiting for a 490 announcement, not sure if that will be Vega or Polaris, but I would still expect Polaris which would mean there is more to come
Did you understand when Raja stated they are keeping HBM 1 on the market to recoup their investment? What does that mean
Two possibilities
A) Fury line is staying a bit longer
B) Polaris will come with HBM 1
B doesn't make sense because it will be quickly replaced by HBM2 Vega because the need for 8gb's of vram that would be a waste of time, and money to do such a thing.
So we are left with A
now this is what it comes down to, how can they Keep Fury X, nano around if Polaris is getting up to those performance numbers and spanking them in power consumption?
Two possibilities
A) Fury line is staying a bit longer
B) Polaris will come with HBM 1
B doesn't make sense because it will be quickly replaced by HBM2 Vega because the need for 8gb's of vram that would be a waste of time, and money to do such a thing.
So we are left with A
now this is what it comes down to, how can they Keep Fury X, nano around if Polaris is getting up to those performance numbers and spanking them in power consumption?
Finally, some real reasoning for your conclusions...
How about C) They keep Dual Fiji as their #1 card to sell HBM1, and then also include HBM1 in the lower end Vega chips (possible to use HBM1 or HBM2 on the same imposer, just different sized stacks? 4gb HBM1 lower end, 8gb HBM2 higher end).
It was initially why I thought HBM1 was going to be in the top end Polaris, but the conversion to GDDR5x 'because it is less expensive', dispite having developed Polaris for HBM1 by their own accounts, leads me to believe that they are either; A) saving their HBM1 investment for Vega, or B) not as overstocked/invested in HBM1 as may have been thought/implied. Otherwise adding new inventory/investment of a different memory type based on 'costs' wouldnt make much sense at all, if you developed and prepared to use the existing inventory/investment that you already have.
How about C) They keep Dual Fiji as their #1 card to sell HBM1, and then also include HBM1 in the lower end Vega chips (possible to use HBM1 or HBM2 on the same imposer, just different sized stacks? 4gb HBM1 lower end, 8gb HBM2 higher end).
It was initially why I thought HBM1 was going to be in the top end Polaris, but the conversion to GDDR5x 'because it is less expensive', dispite having developed Polaris for HBM1 by their own accounts, leads me to believe that they are either; A) saving their HBM1 investment for Vega, or B) not as overstocked/invested in HBM1 as may have been thought/implied. Otherwise adding new inventory/investment of a different memory type based on 'costs' wouldnt make much sense at all, if you developed and prepared to use the existing inventory/investment that you already have.
Last edited:
One slide that VideoCardz has shows that AMD defines "Mainstream Core" up to Performance, and I think this excluded in the past the 290x/390x/Fury models.
The top model then we should expect is an improved performance replacement for the 390.
I would be surprised if they are saying a replacement to the 390 is hitting 980ti performance if it also is focused on power efficiency, I can see it getting close-ish; same could be expected from 1070 at standard resolutions (probably key to both) IMO.
http://cdn.videocardz.com/1/2016/04/AMD-Mainstream-segment.jpg
Cheers
The top model then we should expect is an improved performance replacement for the 390.
I would be surprised if they are saying a replacement to the 390 is hitting 980ti performance if it also is focused on power efficiency, I can see it getting close-ish; same could be expected from 1070 at standard resolutions (probably key to both) IMO.
http://cdn.videocardz.com/1/2016/04/AMD-Mainstream-segment.jpg
Cheers
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Sure.
Buy this 1600 euro dual gpu Fiji today! Get beaten by a 500 euro dual Polaris 10 setup tomorrow.
Brilliant plan.
One slide that VideoCardz has shows that AMD defines "Mainstream Core" up to Performance, and I think this excluded in the past the 290x/390x/Fury models.
The top model then we should expect is an improved performance replacement for the 390.
I would be surprised if they are saying a replacement to the 390 is hitting 980ti performance if it also is focused on power efficiency, I can see it getting close-ish; same could be expected from 1070 at standard resolutions (probably key to both) IMO.
http://cdn.videocardz.com/1/2016/04/AMD-Mainstream-segment.jpg
Cheers
AMD briefly dropped the word 'mainstream' in a conference.. They never made an official slide or Polaris roadmap that shows it will 'Only be a mainstream card', they didnt even say 'only mainstream' when they said it briefly during the conference. Even if they did there is no saying they are using some consistent AMD definintion of what 'mainstream' applies to in all contexts...
People are REALLY reaching here in an attempt to show that AMD has implied they wont be making a high end card, and I dont know why.
I dont think there will be a Dual Polaris. Where has there been intel on a dual Polaris coming out?
Dual Fiji will beat Polaris and no Dual Polaris will come out, ensuring the selling of Fiji and HBM chips until Vega,
Last edited:
They will not recoup costs by doing dual GPU cards. That market is so small it probably doesn't even give a blimp on their bottom lines. Wasn't SLI and Xfire users less than 1% of the enthusiast market let along the entire market?
My conclusions weren't even based off of Raja's statements
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Nobody is slamming amd, we just disagree with your overly enthusiastic predictions
Are we not allowed to? Do we have to accept your conjecture as fact?
Two Polaris 10 cards will smoke dual Fiji if they're as fast as you claim.
All information points to 390/x performance at lower wattage and lower price bracket
no one is slamming anyone, its just not made to be in the same bracket, its not like AMD is going to put Polaris at 500 bucks and try to sell us coal. They are making a good product just not something that is going to be in the same bracket as the initial Pascal. And its not based on assumptions its based on everything AMD has been saying thus far. If its a classic misdirect yeah I can see that, but when you start looking at what EVERYONE has been saying and what AMD's tactics are for Polaris, anything outside of that just doesn't fit.
I would say you have overly pessimistic predictions, but either way I am not using the predictions to judge failure/success or compare in a positive/negative light with their competitor... It just seems people are so intent of labeling a line as a failure or success before it has been released, and you see AMD always being labeled as a failure and nVidia as a success before any benchmarks or real intel is gathered... Its pretty crazy to watch/see people get defensive if their prediction of an AMD failure is questioned.
What has AMD said so far that leads you to conlcude that it will not be in the same bracket as Pascal. Serious question.
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
See...
Pascal is an architecure, not a gpu. Nobody said Polaris is worse than pascal.
We're saying Polaris 10 and gp104 aren't in the same bracket
Look at the financial conference call last week, look all the slides that were released at GDC, look at all the marketing material AMD has released, look at all the interviews done about Polaris since end of last year.
Then go over to 3dcenter journalists and ask them what they think, then go to Hexus and ask the tech journalists (any of them) of that site things, then go over to TPU and ask wizzard what he thinks. They were pretty understanding in their comments in their articles.
You are assuming we are saying its a bad architecture, i don't think anyone stated that. I certainly didn't.
Marketing material shows Polaris 10 hitting the Fury level stuff...
The ONLY thing that would lead to believe that they wont hit the high-end, is the use of 'mainstream' in the financial conference, with no context or determined definition of the word... People reading into that are simply reading into something that isnt there.
What am I missing?
And what is out there that supports this assumption?
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Pretty much everything we know about Polaris 10 and gp104
It never hit fury levels, at least nothing conclusively, because we don't know the settings used. That should have raised your eyebrows. How many times has AMD played the benchmark game where they made sure settings were used to show an intended purpose? And that intended purpose was the total opposite of reality?
Nothing known about either chip would be able to lead to these conclusions. Share what you think does, please.
All slides and marketing material show Polaris 10 at Fury levels.... What else is there to go on?
Benchmarks? We have the "FireStrike" score and a "passively cooled 4k card" demo, but im not sure what else there is... Either way, neither point to AMD not producing a high end card... What am I missing?
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
It talks about only 2 Polaris 10 cards existing, and puts them at the 480 level.... 1) Do you think only 2 Polaris 10 cards will exist? 2) do you think there will be no 490? And if you think it will exist as Dual Polaris, what makes you think that?
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
I mean, if die size, power consumption, and memory configuration mean nothing to him, what kind of information does he expect us to provide?
I don't know looks like he is looking for the holy trinity or something.
Jasmesgalb
We know more about pascal's inner workings than we know about Polaris (GTC pascal white papers),
We know Pascal will have better through put than Maxwell 2 did close to 20% more through put and in compute based situations its going to be even higher than that
We know they will have higher clocks too. We also know, 16nm tends to give a bit more advantage over 14nm with power consumption, look at Iphone's A9 chip it was built on both 14nm Samsung and 16nm TSMC.
We know nV has more experience in extracting performance on nodes by using custom libraries. Should I go on?
Jasmesgalb
We know more about pascal's inner workings than we know about Polaris (GTC pascal white papers),
We know Pascal will have better through put than Maxwell 2 did close to 20% more through put and in compute based situations its going to be even higher than that
We know they will have higher clocks too. We also know, 16nm tends to give a bit more advantage over 14nm with power consumption, look at Iphone's A9 chip it was built on both 14nm Samsung and 16nm TSMC.
We know nV has more experience in extracting performance on nodes by using custom libraries. Should I go on?