[VR-Zone]AMD upcoming Tonga GPU to be released in mid August

[NostaSeronx]

AMD would need the GPU IP to be on the same node of the APU IP. To reduce cost and have time to market. It would be pretty bad to have to delay 20-nm APUs to the 2H of 2015. Just because someone didn't produce IP on 20-nm.

Everything collides with 20-nm being the HBM node.

GPUs have a faster design flow thus would appear much quicker than CPUs or APUs. It would be AMD's fault not to use the parts indirectly created by this guy: http://www.linkedin.com/in/atluruchaitanya

Also, if TSMC's and GlobalFoundries' HBM is only 20-nm LPM/SOC and 16/14-nm FF/FF+/LPE/LPP.

How do we explain this patent image.

Sorry for the image spam but it must be done. (16-nm SOC is a typo of 20-nm SOC)

Check the dates with;
http://www.microarch.org/micro46/files/keynote1.pdf

Die stacking is happening in the mainstream. It is happening now because we need it & It is going to change who and how we build sockets in the future

Ya, tell them how FP4 and SP2 are single-channel.

 

[Cloudfire777]

Yes all fine and dandy, but nobody have confirmed that Tonga will even have HBM.
It could be a R9 2xx card, and 20nm + HBM might be for 300 series only

 

[NostaSeronx]

Iceland, Tonga, and Maui are Volcanic Islands(Rx 2xx) and Pirate Islands(Rx 3xx).
Fiji, Treasure, and Bermuda are Pirate Islands(Rx 3xx).

Iceland is the only one excused from the 20-nm list if it isn't 20-nm. Everything is pointing that Iceland is early because it isn't using High Bandwidth Memory. It isn't a rehashed Oland because it uses Volcanic Island macros, like the ACE F32 units, etc.

DDR4 PHYs for AMD are pure 20-nm as well and they aren't the only ones putting DDR4 on 20-nm nodes.

Southern Islands - 28nm Gen 1 (HP, HPL, LP)
Sea Islands - 28nm Gen 2 (SHP, HPM, SLP)
Volcanic Islands - 20nm Gen 1 (LPM, SOC)
Pirate Islands - 20nm Gen 2 (FF+, LPP, FF, LPE, FD)

 

[Keysplayr]

So it's like 100% speculation then based on things that can't be extrapolated. Cool.

 

[Techhog]

If you think I have insider info you to can have insider info then;

site:linkedin.com
filetypedf
"DDR3/DDR4 HBM"
"GDDR5 6.5 GBps / HBM 256 GBps"

https://www.google.com/?tbm=pts <= Patents
etc

Everything I say is hinted at while I have to guess the correct specifications before AMD launches. As we get more close to the launch more information is displayed.

For giggles;
/ site:linkedin.com GDDR6 Tablets /

--
Topaz;
http://www.amd.com/en-us/products/graphics/notebook/r5-m200#2
http://www.amd.com/en-us/products/graphics/notebook/r7-m200#2

The only GPU I have yet to see launch is Emerald XTX R9 M280X (Bonaire XTX)

A link to Google patent search is meaningless to me, and the GPUs you linked are already released and not Topaz. I don't think you have insider info, though; I think you're just full of it.

Warning issued for inflammatory language.
-- stahlhart

 

[tviceman]

It's not going to have HMB, and I think that is a pretty safe guess. It takes a solid 18-24 months to design a GPU, and I don't think HMB was a viable reality yet for GPU design 18-24 months ago. Also guessing it'll probably be around 250mm^2 in die size.

Anyone wanna bench a 7970 @ 900-1000ghz core but downclock the vram to 4ghz? That's probably about where this chip's performance will end up.

 

[Cloudfire777]

It seems more and more like this slide is coming true.

Anyone remembers R9 370X specs that floated around earlier?
Notice that it says 1536 shaders?

What if Tonga, with 2048 shaders (could be 1536 shaders too if they reduce ALU down to 48/CU...), is the R9 370X? If so, then Seronx might be absolutely right that Tonga is 20nm and may feature HBM. A system builder from Overclockers UK even said that AMD is ready with next generation cards and have been for a while, so he may have been talking about Tonga...

You may also notice that GM206 is mentioned in the slide and will be released soon.
It just happened that GTX 870MX was listed a couple of weeks ago. Which I`m sure is GM206...

Hmmmm

 

[Techhog]

Except that we know that 20nm isn't happening this year...

I just get annoyed when wild speculation is presented as fact. :/

 

[Cloudfire777]

Except that we know that 20nm isn't happening this year...

I just get annoyed when wild speculation is presented as fact. :/

Plans can change, and I`m not sure if Lisa was entirely direct when she responded to David`s question. She was referring to overall product portfolio, which may have meant "in general" but some exceptions.

Maybe. We will see

David Wong- Wells Fargo
Great. And when will we expect you to introduce GPU products at a node below 28 nanometers? Will have you have any 20-nanometer GPUs this year or next year?

Lisa Su - SVP and General Manager of Global Business Units
David, I think what I said earlier sort of what we're doing in terms of technology strategy, we are 28 this year, we have 20-nanometer in design, and then FinFET thereafter. So that's the overall product portfolio.

David Wong - Wells Fargo
That includes GPUs, Lisa?

Lisa Su - SVP and General Manager of Global Business Units
That's the overall product portfolio, so I'm not being specific about graphics versus other products.

 

[f1sherman]

^^ wow that's some reaching.

And if she stated the obvious - yes, overall portfolio DOES include GPU,
then I suppose that still leaves the possibility of not including ALL GPU's? :ninja:

 

[Cloudfire777]

Gotta stay postive you know. She might as well had been referring to everything they release yes, but could also mean "in general", ie some exceptions. Probably not, but maybe

Hope for 20nm, but won`t mind 28nm either considering what Maxwell could do while still on 28nm

 

[f1sherman]

can you link the original post with that AMD/NV roadmap please?

btw it does not mention GM108 which is already out... hmmm

 

[Cloudfire777]

can you link the original post with that AMD/NV roadmap please?

btw it does not mention GM108 which is already out... hmmm

http://www.3dcenter.org/news/kommt-amds-tonga-grafikchip-bereits-im-august

Note: This is not the original site that made the chart. I read it long before it was posted on 3dcenter. It was a Japanese site which I can`t remember the name of

 

[f1sherman]

http://www.3dcenter.org/news/kommt-amds-tonga-grafikchip-bereits-im-august

Note: This is not the original site that made the chart. I read it long before it was posted on 3dcenter. It was a Japanese site which I can`t remember the name of

280 perf @150W is not that big perf/Watt increase, and would leave them significantly behind Maxwell part(s)
AMD needs better than that, and I am guessing they will deliver whats needed.

And before someone says you cant extrapolate perf/W for bigger dies from smaller ones.
Best NV part (excluding non-kepler GM107) is a full blown GK110!

 

[tviceman]

280 perf @150W is not that big perf/Watt increase, and would leave them significantly behind Maxwell part(s)
AMD needs better than that, and I am guessing they will deliver whats needed.]

just because they need it doesn't mean they will deliver. Fermi was late and highly unoptimized despite nvidia needing it to deliver. The 2900xt was late and underperforming, despite ATI needing to drastically catch up.

Keep the last 4-5 years in perspective. Nvidia's biggest priority has been working to build a great 2 watt GPU that can scale up. AMD's biggest priority is to build the ultimate APU. Both strategies have advantages and drawbacks, but if AMD were to actually beat Maxwell in efficiency on 28nm, then Nvidia will have been dramatically shown up.

 

[f1sherman]

just because they need it doesn't mean they will deliver. Fermi was late and highly unoptimized despite nvidia needing it to deliver. The 2900xt was late and underperforming, despite ATI needing to drastically catch up.

Keep the last 4-5 years in perspective. Nvidia's biggest priority has been working to build a great 2 watt GPU that can scale up. AMD's biggest priority is to build the ultimate APU. Both strategies have advantages and drawbacks, but if AMD were to actually beat Maxwell in efficiency on 28nm, then Nvidia will have been dramatically shown up.

Just because NVidia was so vocal about being focused on per/Watt does not mean that that same metrics should be(or was) of any less importance to AMD.

Two of them being neck to neck for so long, I just can't imagine another round of total domination (on either part).
It won't be an an easy task, Maxwell looks that good.
But I just can't imagine AMD throwing the towel and not delivering just as good as Maxwell. If not they are gone from mobile, and will fight for scraps on Desktop.
Perhaps Nvidia/TSMC will help them by canceling chip or two

 

[tviceman]

Just because NVidia was so vocal about being focused on per/Watt does not mean that that same metrics should be(or was) of any less importance to AMD.

Two of them being neck to neck for so long, I just can't imagine another round of total domination (on either part).
It won't be an an easy task, Maxwell looks that good.
But I just can't imagine AMD throwing the towel and not delivering just as good as Maxwell. If not they are gone from mobile, and will fight for scraps on Desktop.
Perhaps Nvidia/TSMC will help them by canceling chip or two

You aren't looking at the whole picture. Perf/watt is definitely one part and Nvidia is placing paramount on it being the most important aspect. But AMD might try to maximize perf/mm^2, which is evident in Hawaii's transistor density. Despite having seen Kepler in action some 18 months in advance, and despite improvements in GCN architecture, Hawaii could not compete in perf/watt vs. Kepler. Instead, Hawaii competes much more effectively in perf/mm^2.

There are more ways to compete than in just perf/watt.

 

[f1sherman]

No, not really when those same parts are covering mobile segment too.

And when you say Hawaii could not compete with GK110 in perf/mm2... the difference is not all that big (lets not wave with benchmarks OK?)
Furthermore that perf/Watt difference pretty much directly translates to difference in absolute performance between the two.
Coincidence? Ofc not because thermal solutions have have a finite upper limit ~300W.

Still, Hawaii's economics should make up for that with smaller die as you correctly point out.
And it most certainly would/does, if it didn't require more elaborate cooling solution than what AMD had in the 1st place -> Even more price to pay for slightly worse perf/Watt, because every Watt counts when you are dancing on it's upper limit.

BTW notice that if 290/X parts weren't trying to compete with 780/Ti on absolute performance basis, they would have identical if not better perf/Watt!!!
(you already mentioned that GK110 is 18 months old design, so it damn well should have)

So much for AMD does not (need to) compete on this fundamental metrics.
AMD needs it just as much as NV, only difference being the latter for some odd reasons gets away with bigger dies, hence AMD needs perf/mm2 more than NV does.
Not that NV minds great perf/mm2 - remember GK104 and THE RISE OF MARGIN

 

[tviceman]

No, not really when those same parts are covering mobile segment too.

AMD has APU's to volume-cover the notebook segment, and it's definitely a priority for them to serve that market more so than the notebook discrete GPU market. AMD's APU's are not competing in the same products as Tegra. Tegra is and has to be much more power-aware, so much so that Nvidia has for the first time that I can think of (ever???) released a small die first with a new architecture.

And when you say Hawaii could not compete with GK110 in perf/mm2... the difference is not all that big (lets not wave with benchmarks OK?)
Furthermore that perf/Watt difference pretty much directly translates to difference in absolute performance between the two.

No, you're off here. GTX780 TI is about 10% faster AND 10% more efficient. That means it's generally consuming less power AND performing faster.

Still, Hawaii's economics should make up for that with smaller die as you correctly point out. And it most certainly would/does, if it didn't require more elaborate cooling solution than what AMD had in the 1st place -> Even more price to pay for slightly worse perf/Watt, because every Watt counts when you are dancing on it's upper limit.

So much for AMD does not (need to) compete on this fundamental metrics.
AMD needs it just as much as NV, only difference being the latter for some odd reasons gets away with bigger dies, hence AMD needs perf/mm2 more than NV does.
Not that NV minds great perf/mm2 - remember GK104 and THE RISE OF MARGIN

It's give and take; I agree to a point. But like I said, if AMD margins are lower than Nvidia's (and they are) then I certainly think that cramming as much performance as they can in a chip might be more important to them than perf/watt.

BTW notice that if 290/X parts weren't trying to compete with 780/Ti on absolute performance basis, they would have identical if not better perf/Watt!!!
(you already mentioned that GK110 is 18 months old design, so it damn well should have)

If AMD had downclocked Hawaii to be as (or more) competitive on a perf/watt basis when compared to GK110, then it would not have the best perf/mm^2 measurement. So I think my argument is correct; AMD had 2 years since Tahiti came out and 18 months since GK110 was on the market (21 months since GK104), yet they didn't make any strides in perf/watt with Hawaii. They chose to concentrate on perf/mm^2 and if you look back at the Hawaii pre-release info, when they bragged about how dense their memory controllers are among other aspects of the chip, I think this goes to show that I am right.

 

[antihelten]

No, you're off here. GTX780 TI is about 10% faster AND 10% more efficient. That means it's generally consuming less power AND performing faster.

No. If it's 10% faster and 10% percent more power efficient, that means it consumes the exact same amount of power and performs faster. Efficiency is performance/power, increase performance by 10% and efficiency naturally goes up by 10% as well, with no change to power consumed.

 

[tviceman]

No. If it's 10% faster and 10% percent more power efficient, that means it consumes the exact same amount of power and performs faster. Efficiency is performance/power, increase performance by 10% and efficiency naturally goes up by 10% as well, with no change to power consumed.

My wording may be off, but it's well documented that GTX780 TI consumes LESS power on average than a r9-290x while outperforming it by around 10%

I'd link hardocp and anandtech but suddenly I don't care to put it all here when anyone else can look just as easily.

 

[antihelten]

My wording may be off, but it's well documented that GTX780 TI consumes LESS power on average than a r9-290x while outperforming it by around 10%

I wasn't really doubting your overall claims about the GTX 780, just that the exact sentence you wrote (wording) was incorrect, which may have been a bit nitpicky of me.

With that being said though there two important things to keep in mind with regards to GTX 780 Ti efficiency vs. 290x efficiency discussion. First of all, at least part of the higher power consumption of the 290x comes down to the inferior cooling solution (since higher temps mean higher power use). Secondly, you should be careful when looking at TPUs performance per watt graphs, since it is calculated using the average performance of their entire benchmark suite, whilst only using the power usage of a single game (crysis 2 for older reviews, metro last light for newer ones).

 

[3DVagabond]

Efficiency depends on metric used. If you use OpenCL, just as an example, AMD destroys nVidia in any metric including perf/W. They are also better in double precision. I'm not trying to create a pissing contest on whether or not those metrics are more or less important than gaming performance, just that there is more than one metric that could be used to determine efficiency and nVidia isn't simply more efficient than AMD.

 

[itsmydamnation]

im confused by some of the language here,

GK-110 is ~551mm sq
hawaii is 438mm sq

113mm sq is not insignificant, GK-110 is 26% larger. power consumption isn't that different, http://www.tomshardware.com/reviews/geforce-gtx-780-ti-review-benchmarks,3663-17.html , 6% in GK-110 advantage. actual performance improvement is a pain in the arse to track with driver updates etc but 9% seems reasonable ( a properly cooled 290x is a couple of points closer).

so it seems to me if your to have a performance/perf/power rating with each having an equal weight hawaii wins by about ~10%.

 

[Cloudfire777]

#mildlyinteresting
http://webcache.googleusercontent.com/search?biw=360&bih=269&tbs=qdr%3Ad&ei=G2S2U-v-DYbY4QSCrYDgCg&q=cache%3Adevgurus.amd.com%2Fmessage%2F1304894&oq=cache%3Adevgurus.amd.com%2Fmessage%2F1304894&gs_l=mobile-gws-serp.12...13937.19848.1.20937.8.7.0.1.1.0.754.2811.2-4j1j0j1j1.7.0.ernk_qsrb...0...1.1j2.48.mobile-gws-serp..1.7.2512.3.oj872hyiErY

ID :6938
Tonga is not just 692x
http://pci-ids.ucw.cz/read/PC/1002