[Rumor]R9 300 series will be manufactured in 20nm!

[Cloudfire777]

Latest news/leak from Korea (was posted yesterday - Cant give out source):

R9 300 series from AMD is manufactured in 20nm


It actually falls in line with the news in October last year that AMD have started making the Xbox One APU in 20nm.
That APU draw about 120-130W so it seems they can make high power in 20nm now. I`m guessing 20nm from GlobalFoundries!

Take it with some salt but the leaker have given people a warning that they will be banned if they post the news elsewhere.

Amazing if true! :ninja:




Edit #2:
Anyone remembers the R9 370 card specs that was leaked months ago by Videocardz with power draw from 110W to 130W?

We later found out it was a R9 270X rebrand through driver listing.
Except 270X is at 180W.

Guess thats another sign for 20nm....

 

[Cloudfire777]

Suddenly R9 395X2 makes sense!
Power and thermals over R9 200 series is probably much better. Die size will be smaller too than if they made it on 28nm.

 

[TemjinGold]

Hahaha... 20nm... can't give out source... hahaha... Good one, I needed a laugh today.

 

[burninatortech4]

What a joke.... Valid sources please.

 

[Cloudfire777]

Not sure whats so funny. I said he will ban people if people post sources. Maybe he will not post more

I already said if AMD can manufacture Xbox One APU in 20nm they can do it on GPUs. Meaning someone is giving them HP 20nm chips.
And it explains why AMD rebrands almost the entire 200 series.
Why R9 395X2 is coming.
Why GPU-Z reads 1024 shaders on the R9 M375 screenshot. Its either nothing (blank squares) or correct shader count I think.

 

[Madpacket]

Yeah no...

 

[HurleyBird]

There are several problems here:
We know its a Cape Verde chip because the 3DMark11 score is identical with 7870M (I have compared), another Cape Verde chip. But the shader amount is wrong, it says 1024 shaders but it is 640 shaders (Cape Verde XT). And the VRAM amount is also grossly miscalculated.
If it was a simple rebrand with nothing new, GPU-Z would have no issues reporting right specs for the chip. Atleast not miscalculating it like this.
Another issue is the Texture fillrate. 65 GTexels/s? That is also much higher than Cape Verde

But if it is 20nm, its a different chip and GPU-z doesnt support it yet.

If it has so many more functional shader units, why is the 3DMark score the same as Cape Verde? I'd say it's more likely that the GPU-z is messed up because it's a new stepping with support for Freesync, True Audio, and maybe the latest GCN enhancements. Is this mysterious source basing 20nm on this and the rumoured new XBOX One APU and extrapolating, or does he actually have real evidence as opposed to conjecture?

 

[Head1985]

No way its 20nm
Maybe some low-ends cards but not Fiji.

 

[Techhog]

Not sure whats so funny. I said he will ban people if people post sources. Maybe he will not post more

I already said if AMD can manufacture Xbox One APU in 20nm they can do it on GPUs. Meaning someone is giving them HP 20nm chips.
And it explains why AMD rebrands almost the entire 200 series.
Why R9 395X2 is coming.
Why GPU-Z reads 1024 shaders on the R9 M375 screenshot. Its either nothing (blank squares) or correct shader count I think.

Weren't you adamant just a few days ago that Fiji itself is two Tonga GPUs?

 

[Cloudfire777]

If it has so many more functional shader units, why is the 3DMark score the same as Cape Verde? I'd say it's more likely that the GPU-z is messed up because it's a new stepping with support for Freesync and True Audio. And is this mysterious source basing 20nm on this and the new XBOX One APU and extrapolating, or does he actually have real evidence as opposed to conjecture?

It doesnt have 1024 shaders, thats the thing.
It reads its a Cape Verde, but think it got more shaders than it do. I never seen that before on an existing old chip.

The source is a korean with insiders in the industry. Thats all I can tell.
He already knows the specs about 960Ti which I havent bothered to post.

He havent mentioned anything about Xbox or anything. That is me making sense of it

 

[Head1985]

The source is a korean with insiders in the industry. Thats all I can tell.
He already knows the specs about 960Ti which I havent bothered to post.

Let me guess:
1280SP
80TMU
48Rops
192bit
3Gb Vram
btw AMD says they skip 20nm
http://www.techpowerup.com/211973/a...-to-14-nm-with-arctic-islands-gpu-family.html

 

[Cloudfire777]

Let me guess:
1280SP
80TMU
48Rops
192bit
3Gb Vram
btw AMD says they skip 20nm
http://www.techpowerup.com/211973/a...-to-14-nm-with-arctic-islands-gpu-family.html

Again, 14nm will not happen on AMD GPUs. The process there is for LOW POWER. That is fudzilla, wccftech that is the original "source".
And they should know 14nm is for LPP and LPE.

 

[Head1985]

Now its low power but that gonna change in end of 2016
new process is always Low power first and then HP

 

[JDG1980]

Suffice to say that I'm very skeptical of this 20nm rumor. It contradicts most of what we've heard previously; all indications are that 20nm is suitable only for low-power portable devices, not high-performance CPUs or GPUs. Is GloFo even doing 20nm at all? I thought they were going straight to 14nm FinFET licensed from Samsung.

That said, if the XB1 and PS4 APUs can be manufactured on 20nm, then there wouldn't be any reason I can think of why a discrete GPU couldn't be. The PS4 APU basically contains a Pitcairn GPU with two CUs disabled, only slightly downclocked (800 MHz). But there hasn't been any official confirmation of APUs on 20nm, either; this could have been an experiment that didn't pan out. On the other hand, if AMD is contractually obligated to do a 20nm APU for Microsoft and Sony, then using the same process for discrete GPUs to close the perf/watt gap with Nvidia starts to look very tempting. In for a penny, in for a pound.

IF AMD does somehow manage to pull off 20nm GPUs, it would explain a lot, including the delays and the company's silence in the face of rumors. GCN is still a viable architecture - a node shrink would match if not beat Maxwell in perf/watt even if no other improvements have been made since Tonga, and this would buy AMD some time to catch up on the design side before Nvidia executes their own die-shrink.

All that said, I think the balance of the evidence still is that AMD's 300 series GPUs will be manufactured on 28nm - probably GloFo, because it was designed specifically with GPUs and APUs in mind and thus may be more efficient for these applications than TSMC, and because AMD has to buy their wafers anyway. Still, the GPU-Z screenshot in the original post is encouraging because it contradicts the "everything but Fiji will be a rebrand" rumor that's been floated. It looks like we might actually get a full new lineup from AMD, and that's great news. Hopefully, these cards will include hardware H.265 decoding, as Carrizo's iGPU has been confirmed to do. At the moment, the only shipping product that does full H.265 in hardware is the GTX 960.

 

[Cloudfire777]

Now its low power but that gonna change in end of 2016
new process is always Low power first and then HP

Who said Samsung were in the business or even have the capacity to retool and make wafers for high power?

 

[Silverforce11]

Who said Samaung were in the business or even have the capacity to retool and make wafers for high power?

Samsung. LPE & LPP are separate designs from the start.

I find it hilarious that 20nm planar which was designed for mobile SoCs can now make an Xbone SOC at high power... but a new 14nm node specifically designed for performance cannot, according to you.

Apparently it can't.. but then it can.

 

[waffleironhead]

Re m375 shows 1024 shaders. My m275 shows that as well when not under load. Once its under load it shows the proper shader count.

 

[Cloudfire777]

Samsung. LPE & LPP are separate designs from the start.

I find it hilarious that 20nm planar which was designed for mobile SoCs can now make an Xbone SOC at high power... but a new 14nm node specifically designed for performance cannot, according to you.

Apparently it can't.. but then it can.

20nm wasnt designed for mobile SOC. TSMC's 20nm was, and is for low power and still is as far as I know.

Dont mix GlobalFoundries in this.

Re m375 shows 1024 shaders. My m275 shows that as well when not under load. Once its under load it shows the proper shader count.

Really? Can you post screenshot?

Suffice to say that I'm very skeptical of this 20nm rumor. It contradicts most of what we've heard previously; all indications are that 20nm is suitable only for low-power portable devices, not high-performance CPUs or GPUs. Is GloFo even doing 20nm at all? I thought they were going straight to 14nm FinFET licensed from Samsung.

That said, if the XB1 and PS4 APUs can be manufactured on 20nm, then there wouldn't be any reason I can think of why a discrete GPU couldn't be. The PS4 APU basically contains a Pitcairn GPU with two CUs disabled, only slightly downclocked (800 MHz). But there hasn't been any official confirmation of APUs on 20nm, either; this could have been an experiment that didn't pan out. On the other hand, if AMD is contractually obligated to do a 20nm APU for Microsoft and Sony, then using the same process for discrete GPUs to close the perf/watt gap with Nvidia starts to look very tempting. In for a penny, in for a pound.

IF AMD does somehow manage to pull off 20nm GPUs, it would explain a lot, including the delays and the company's silence in the face of rumors. GCN is still a viable architecture - a node shrink would match if not beat Maxwell in perf/watt even if no other improvements have been made since Tonga, and this would buy AMD some time to catch up on the design side before Nvidia executes their own die-shrink.

All that said, I think the balance of the evidence still is that AMD's 300 series GPUs will be manufactured on 28nm - probably GloFo, because it was designed specifically with GPUs and APUs in mind and thus may be more efficient for these applications than TSMC, and because AMD has to buy their wafers anyway. Still, the GPU-Z screenshot in the original post is encouraging because it contradicts the "everything but Fiji will be a rebrand" rumor that's been floated. It looks like we might actually get a full new lineup from AMD, and that's great news. Hopefully, these cards will include hardware H.265 decoding, as Carrizo's iGPU has been confirmed to do. At the moment, the only shipping product that does full H.265 in hardware is the GTX 960.

Well Im a bit sceptical too but open to it based on the rumors about Xbox One and PS4 APU is being produced on 20nm. I think Ive read that PS4 draw about 140W while on load so the APU is drawing quite a bit energy.
The only company that can make 20nm for AMD must be GloFo but I too was wondering if they even make it. But they sort of have to if they supply the chips for Sony and Microsoft.

Global Foundries have 8 fabs in total and only 1 of them are working with Samsung to produce 14nm FinFETs. Meaning they have 7 other working on other processes. 28nm is probably the main part of them, but who knows, maybe 20nm is being produced for 300 series?

Fingers crossed. I mean, dual Fiji (R9 395X2) on 28nm? Sounds scary regarding temps and TDP on such a beast. 20nm however, thats a different story

 

[96Firebird]

Re m375 shows 1024 shaders. My m275 shows that as well when not under load. Once its under load it shows the proper shader count.

Yup, looks like the GPU-Z results for the M275 are a little bugged (shader count and VRAM)... Wonder if this carried over to the M375, if it is indeed just a rebrand with higher clockspeeds.

Source

 

[Cloudfire777]

Yup, looks like the GPU-Z results for the M275 are a little bugged (shader count and VRAM)... Wonder if this carried over to the M375, if it is indeed just a rebrand with higher clockspeeds.

Source

Thanks man. Well that rules out GPU-z reading M375 wrong because its 20nm. Its a rebrand since Device IDs the same.

Would GPU-z be able to read M375 as rebrand if they made Cape Verde on 20nm? The hardware is still the same right?

 

[waffleironhead]

Here they are

 

[RussianSensation]

Yup, looks like the GPU-Z results for the M275 are a little bugged (shader count and VRAM)... Wonder if this carried over to the M375, if it is indeed just a rebrand with higher clockspeeds.

Source

Good catch. It does look like the same chip with GPU clocks increased from 925mhz to 1015mhz. Considering R9 M275 was announced on August 1, 2014, it's pretty lame to release a card 10% faster by summer 2015 unless AMD is dropping this into 1-2 price levels below where R9 M275 sat. I mean it makes no sense why AMD would delay this card for 11-12 months just for a 10% bump in GPU clocks. What a lame strategy.

 

[JDG1980]

Good catch. It does look like the same chip with GPU clocks increased from 925mhz to 1015mhz. Considering R9 M275 was announced on August 1, 2014, it's pretty lame to release a card 10% faster by summer 2015 unless AMD is dropping this into 1-2 price levels below where R9 M275 sat. I mean it makes no sense why AMD would delay this card for 11-12 months just for a 10% bump in GPU clocks. What a lame strategy.

AMD really shouldn't be reusing any GCN 1.0 parts at all, regardless of the clock speeds. These should have been put out to pasture a year ago or more. That said, I think that Cape Verde in particular may remain, because AMD doesn't want to create new ultra-low-end GPUs. They make APUs with up to 512 SPs, so they see Cape Verde as an obsolete product and probably don't have any plans to replace it. It's the same reason that Nvidia still uses Fermi (!) for some of their bottom-of-the-barrel offerings: these cards aren't profitable and don't sell well, so they won't get refreshes. Hopefully everything from Bonaire on up will be refreshed. In the case of Tonga, this may simply entail porting to GloFo, though I hope they at least add a HEVC decoder while they're at it. If they can get fully enabled Tonga into a 150W power envelope (one PCIe connector), it would be a much stronger competitor to the GTX 960 than it currently is. The Tonga FirePro W7100 already has a 150W TDP and one power connector; I'd be interested to see if it is because of better binning, or if it just throttles under heavy loads to achieve that. The clock speeds and RAM speeds are virtually identical to those on the R9 285.

 

[RussianSensation]

AMD really shouldn't be reusing any GCN 1.0 parts at all, regardless of the clock speeds. These should have been put out to pasture a year ago or more. That said, I think that Cape Verde in particular may remain, because AMD doesn't want to create new ultra-low-end GPUs.

It's probably unrealistic for AMD which is so strapped for cash and engineering resources to do an entire new stack of mobile and desktop SKUs when this generation is really shaping up to be so short. I don't see how it's even physically possible. NV is primarily a graphics/GPU-focused firm and they have multiple teams working on different SKUs and different GPU architectures concurrently. AMD has so many other core projects going on, GPUs isn't their primary focus by any stretch of imagination. Therefore, it's a guarantee that we will see some re-branded products on both desktop and laptop side. But imagine if R9 370X = 270X re-badge and sells for $109-119, well that's 43% faster than the $120 GTX750Ti. That means if you want a budget gaming card, you will not be able to beat this type of value. One can argue that because NV's sub-$300 cards are so weak in performance, AMD can almost get away with straight up re-badging everything from Pitcairn to Tonga XT. A $199 Tonga XT with 2048 cores and 1Ghz clocks would wipe out GTX960. This is a risk strategy though since consumers for the first time in decades are foregoing massive performance advantages AMD offers in favour of perf/watt. So it's possible this strategy will backfire badly because even a card 50-60% faster than the 960 4GB doesn't sell as well.....

 

[jpiniero]

Good catch. It does look like the same chip with GPU clocks increased from 925mhz to 1015mhz. Considering R9 M275 was announced on August 1, 2014, it's pretty lame to release a card 10% faster by summer 2015 unless AMD is dropping this into 1-2 price levels below where R9 M275 sat. I mean it makes no sense why AMD would delay this card for 11-12 months just for a 10% bump in GPU clocks. What a lame strategy.

OEMs want new products. New products is what they get.