R9 370 Details leaked - Finally a new architecture from AMD?

You do realize that it is already confirmed that AMD will use 14nm for their CPUs and APUs in 2016 right?

 

Idk what to think whenever you say the word "confirmed". It's like "HL3 confirmed!" know what I mean?

 

No, I am being serious, it will be Samsung/GloFo 14nm.

 

See, this is exactly what I mean. You can't seem to separate rumor/educated guess from fact. Confirmed means fact, unless I'm from a different planet.

 

I thought it was 16nm?

 

Lets put some context out there:

The newest APUs from AMD, Carrizo, based on the new Excavator architecure, is 28nm.
Intel`s Skylake/Broadwell is 14nm.

You actually think AMD will leapfrog and suddenly be neck in neck with Intel, which have one of the most advanced foundries out there, which have always been way ahead of AMD?
Nope.

AMD will first go to TSMC`s 16nm FinFET which is 20nm (Intel`s 14nm is 14nm), then go to 14nm later once TSMC move there. 14nm process at Globalfoundries is for SOC`s, not big APUs with power hungry cores and IGPs.
SOCs is typically up to 4W ish, APUs from AMD is all the way up to 100W for desktops. GlobalFoundries/Samsung 14nm SOC process cannot deal with the power involved in APUs. TSMC 16nm FinFET or GlobalFoundries 20nm SOI can.
AMD make desktop chips first, thats their priority, then bin them and use lower voltage on those that have good enough silicon, and make those mobile APUs.
So Zen, AMDs next APUs, will be in 16nm FinFET (20nm).

Rory Read said this in late 2014:

AMD is without foundry. They will go to 14nm when TSMC/GloFo have that ready for them in the process it requires. Intel however control their own destiny. Thats why they are ahead of AMD

 

AMD is not going to use TSMC 16nm, it is going to use 14nm Samsung/Globalfoundries in 2016, it doesn't matter if it's only partial 14nm/20nm, they call the process 14nm Samsung so that's what I will call it also, and Carrizo uses 28nm and that is also Globalfoundries and not TSMC. Doesn't matter if 14nm Samsung right now is for SoCs they have a better process coming. There have been reputable sources confirming it (including people from AMD if I remember correctly) that Zen and it's derivatives (APUs) will use 14nm. GPUs are not confirmed 14nm but I doubt they will be based on tSMC 16nm. TSMC has lost this round, AMD wasn't even in their customer list for 16nm.
We should be getting true 8 core Zen CPUs with much more threads (maybe four threads per core similar to IBM) at 95W TDP, with Intel Skylake already being 95W and most likely 4 cores only. I don't know how good the IPC of Zen would be compared to Skylake but it would be way better than Excavator it is replacing, maybe even faster than Broadwell. Heck there is even a possibility for us to get 8 core mobile CPUs and that's a great thing, considering that if leaks are true about 8 cores at 95W so there is no reason why AMD can't put that in mobile CPUs just like Intel does for their CPUs and even AMD for their current APUs just by lowering clocks a bit (Kaveri FX 7600P performance is already close to desktop Kaveri, and the massive efficiency improvements we are supposed to get with Carrizo will close that gap even closer). It will be interesting to see the Carrizo improvement over mobile Kaveri and that's when both are 28nm, all I can say right now is that the potential for Zen is huge, will it be another athlon for them or will it just be a Phenom (first gen), whatever the case it won't be another Bulldozer)

As for your quote Cloud, Reed does mention 14nm so I'm sure he knew something when he said it.

 

Lets take this again Link.

The upcoming 14nm at GlobalFoundries/Samsung is either LPE or LPP process. Both is for low power SOCs (ARM, Qualcomm, Samsung, Altera, Nvidia - Tegra etc) with power draw maxing at 5W ish max.
28nm Carrizo APUs was indeed produced at GlobalFoundries but its a 28nm SHP process, Super High Performance. That process can deal with the power required for APUs.

Next step for AMD is 16nm FinFET (20nm) from TSMC or 20nm from GlobalFoundries (if GloFo offer them). Those nodes are the only ones available to use for chips that require high power (APUs, CPUs, GPUs).





I really hope AMD is sucessful with Zen and that it rocks as well, but Im extremely sceptical and don`t think AMD is able to match Skylake. Hopefully they can match i5-2500k (Sandy Bridge) atleast. That will be nice

 

That doesn't mean that 14nm LPP won't be better than TSMC 16nm, even at 100W. Don't forget that TSMC 20nm high performance failed to outperform their own low power process, even while using more power. We haven't even seen what 14nm LPP can do yet, everything so far has been LPE, which has still been impressive with Samsung's own Exynos chips being clocked higher than ever even at lower power.

I also found this http://hexus.net/tech/news/cpu/79758-amd-already-designing-products-14nm-node/. It's a bit old article but since AMD was already designing products they will be using them, it would be a massive waste of resources to design chips and then change the process and do it over again. Anyway we should find out a lot more once May 6th arrives. And also besides tablet chips like Mullins which are based on high power consumption 15W+ chips AMD doesn't design 5W TDP chips.

 

Still nothing about a 970M competitor. How depressing...

 

I can tell you what I do know:
R9 M385 is a Tonga card/rebrand. Ive been comparing the OpenCL information I have found with many desktop cards, and the M385 is identical only with R9 285 and M295X. That is pretty much confirmed. Which one, R9 285 rebrand with 1792 shaders, or R9 M295X rebrand with 2048 shaders, I can`t tell.
The thing is that rumor has it (plus recent driver from AMD) that the Fiji chip is only new. The rest, below R9 390, is rebrands of 200 chips.
So it could mean that M385 is R9 285 rebrand while M385X is R9 M295X rebrand.

M390X is not confirmed by anyone yet, so nobody knows if its even coming. Mobile cards up to M385X is confirmed.
So 2015 might be just rebrands from AMD which is quite shocking

 

Wont call it exactly hard evidence, but let's go with it. Depends on how you look on the matter (and I'll obviously be on the positive side) it shouldn't be a bad thing. As I said in my previous post (or the one before that), if it comes coupled with a new process would it be a bad thing? It would be a 970m rival, even slightly better (for the fun of it and for the science of course), while pushing some real efficiency. The one that doesn't go out through the door for a walk when you start pushing it - i.e. once you flash non-stock BIOS. Then comes the R9-M39something series, if there ever is such thing, which would be the heavy weight 100~120W GPU champion.

With all that said I'll be selling my Clevo 7970m v1.1 soon and I see at least 780m in my future since HP seems to hate AMD and my machine is rather incompatible with them ("lovely" BIOS thing).

 

What is Amd holding on the 6th of May???

 

Isn't there any information about R9 M390? I really hope it won't be rebrand too

 

Analyst day or something like that, they are expected to reveal official roadmaps.

 

There really haven't been any leaks that include even the slightest mention of it, at this point we have no idea what it is. We do know that M380/X will be based on Trinidad (although no concrete info on whether it's a new GPU or a rebrand) and we also know 385/X will be based on Tonga. The big mystery here is whether Grenada is a new GPU (like Tonga was to Tahiti) or Hawaii on a better process. If it is the former it is possible to be efficient enough to fit into mobile TDP at lower clocks while still offering great performance (at 75% of the clock it should beat 980M at 1440p) and it would be even better if it has HBM. If Grenada is a rebrand the only thing left is Fiji, the only way for which to get into mobile is at considerably lower clocks. At 500-525MHz (Depends whether desktop 390X runs at 1000 or 1050MHz) Fiji should have 80% of the performance of desktop 290X and at 625MHz it would match 290X. These are just scaled results from calculations of course and the real thing may have a 5-10% performance delta.
Still whatever M390 cards end up being based on it will not be a rebranded mobile GPU.

 

^Let your imagination run wild

 

It's called math not imagination.

 

What does it matter? At the end of the day you're still wrong and AMD still can't make a competitive mobile GPU.

 

No you are wrong because you make baseless statements like that, if they don't release 390 and 390X then they can't compete, but there is no information about it so you can't say anything.

 

Thanks for putting words in my mouth. Also I love the irony, coming from you.

 

Come on guys where is the love!?

I just really hope AMD give us something!!! It has been a looooooooooong wait

 

Just wait 2 more months and you will know exactly what they have, I have already said what was possible in my previous posts, no one knows exactly what they have. Don't listen to these skeptics, their opinions are based on BS.

Edit: What I did say is that it was a possibility, nowhere in my previous post did I say anything was a fact, of course those who read properly already know this

 

Please Delete.

 

Something about the R9 m385 being a straight rebrand is too off putting for me. The only chip it could be a rebrand of is the m295x, which is a 100+ watt GPU. I know Nvidia has rebranded the GTX 580m as the GTX 675m, so it isn't a stretch to imagine a 100w 2nd tier mobile GPU (one under the best), but... the m295x is a bit too power hungry to be a 3rd rank GPU. Unless Fiji makes some great power efficiency gains from Tonga, I have doubts this is a straight rebrand but either a downclocked version of the R9 285 (because a stock R9 285 consumes ~190w) or simply a new Tonga GPU.

 

It can be Tonga on a different process, like Globalfoundries 28nm instead of TSMC to reduce power consumption. But I agree with Cloudfire on this one, M385 and M385X would be the perfect numbers for Tonga rebrand.
Also Cloud can you provide the links for the info you have found?

 

It makes absolutely no sense to switch processes without a node shrink. It isn't like AMD has extra money to throw around. I do not know what they expect to achieve on mobile but getting any card out will give consumers another choice. I just hope it doesn't cause spontaneous combustion of the notebooks it goes into

I think at this point it looks like AMD has no chance for the performance crown in 2015 so their best strategy would be to get parts out that compete with the 960M and 965M and focus on 2016. There are more gamers that will end up with 960M or 965M than will end up with 970M and 980M machines. If AMD has tweaked the GPU a bit they could have improved it well enough to get the TDP down.

 

Spontaneous combustion?
I realize that AMD didn't have a stellar track record in mobile space lately, but their mobile Tonga M285 didn't burn down any laptops (yet).
Furthermore, proper cooling hasn't been put in place for that GPU considering its higher thermal requirements, along with less than adequate power supply (which was noted by several people on these forums to have been restricting the GPU in achieving its maximum - the gpu doesn't get enough power to run properly, and its not adequately cooled, resulting in throttling).

Anyway, its not as if Nvidia didn't produce similarly problematic products in the past.

As for AMD... my guess is that their new mobile lineup (M3xx series) will be produced on the enhanced 28nm process (like Carizzo was), which could result in power savings either way.
In addition to that, and information inside drivers aside, we don't know if most of the 3xx series will be rebrands (apart from top of the line gpu's).
And even if they were, it is possible that if the rebrands are made on the enhanced 28nm process, it might provide some enhancements... whereas the top end models will probably be new architecture (Fiji with HBM).

We won't know for sure until AMD releases the products in question.
They are very tight lipped about what's coming out lately, and most 'predictions' (apart from some leaked info) that were made by others (not AMD) turned out incorrect.

 

Well, the mobile Tonga does reach about 90 C in the AW15 according to a few owners, the only laptop to house the GPU. Compare that with a GTX 970m in the same laptop that only reaches to about 70 C. I'm not sure what one could take from this since, like you said, the m295x is being held back, but given that owners of the laptop have actually returned the AMD AW15 because it was hotter than the closest Nvidia counterpart, ehh... it could be worse like the iMac 5K.

 

The spontaneous combustion was poking fun. As someone who has owned ATi/AMD and nVidia cards and had issues with both brands over the years, it was coming from a place of neutrality. I'm tired of seeing GPUs that can heat a small apartment for the winter. I had the 880M and know first hand how hot those got and I regularly see complaints about 90+ on the *throttling* M295X. At least the 880M had a throttling performance algorithm that kept the temps around 87C under load.

I have no faith in AMD pulling a rabbit out of their hats this year for the high end mobile sector and choose to keep my prediction that they will target the 960M and 965M to get something on the market and focus on 2016. It would be nice to see some competition from AMD aimed at the 970M but I just can't see it without a die shrink and a new architecture. 980M? Forget about it...

 

I hadn't stated that AMD will be pulling a rabbit out of their hats at all.
All I said was that its possible their higher end lineup might be based on Fiji and possibly feature HBM... plus that until we get more information that will give us actual data, all we can do is speculate.
As for the enhanced 28nm process, that's not so far fetched.
We've seen the power efficiencies achieved on it with Carrizo... perhaps if the current mobile lineup is produced on the same process, it might result in similar efficiency gains, and there's also a question whether AMD will actually release mostly rebrands again and reserve the higher end for new architecture?

 

Well I was wrong. It get worse. R9 M380 and R9 M385 was listed in new driver today
They are both rebrands. I thought it would be Tonga rebrands earlier. Guess again....
Sorry people :/

 

My doubts were correct. Phew... thought AMD would be crazy.
However, the fact that the R9 28x series is a rebrand of the m280x isn't all that bad. The R9 m280x is only available in one laptop, some random ASUS laptop called the N551ZU (click on specifications). So all this means is it'll get more exposure than before.
Also tried checking if the m280x is a rebrand of any other GPU, but nope. It's unique... well, until the m38x comes along. But the good news is that if anybody wants to upgrade or use the GPU, you can buy and install one yourself.
EDIT: Uhh... scratch what I said about the m280x being unique. It's just a downclocked HD 8770/7790. Well then, this sucks.

 

WOW AMD...! What on earth are they thinking?!

 

Yeah its absolutely horrible. This is AMD giving up.
The entire 2015 will be nothing but rebrands from AMD for mobile like speculated earlier.

I had hopes for M390 and M390X being brand new if M385 was Tonga.
But now that M385 being M280X rebrands I think M390 and M390X will be M29*X rebrands. M390 will probably be M290X rebrand, lol you know, 7970M rebrand (jesus). And M390X a M295X rebrand.

GG AMD
GG Mobile gaming
GG competition

Guess we will have to wait all the way to 2016 and 16nm from TSMC to see AMD with something new for us again :/

 

This is one time I feel bad about being right.. This is not a good thing...

AMD would be better off not releasing anything. Nobody who does any research will buy a rebrand over the 970M unless they are either insanely cheap or the purchase is by someone who is team red til they die...

 

Short AMD stock. Just kidding. Can't believe this, lol.

 

So the 2015 mobile lineup consists of Bonaire, Pitcairn, and Tonga? GG AMD. Better luck next year.

@Link4:

 

GG indeed

 

This probably means that the R9 370X series isn't strong enough to beat the current line, or isn't actually power efficient enough to be thrown into the mobile section. GG.

 

Or it just means they wont bother with mobile users. Shaking my head. Really disappointed. The moment I saw your post cloud, that was my thought exactly - if 380 is going to be 280 rebrand, we are so out of luck, even if there is 390 or whatever, it would be... yeah, REALLY DISAPPOINTED.

 

I wonder if they'll actually boost the clocks on their Pitcairn rebrand this time or if they're going to pull another 8970M - > M290X rename and keep the 8970M performance for 3 generations... What a joke!

 

Would it matter? I never got the whole rebranding thing at all - you know, it's the same chip, but with a bit higher clocks, and probably, but not certainly, a bit more mature technology, so it's 1~2% more efficient. Who cares? Just step-up and say - we'll rain check this year, so don't bother to check the specs of our "new" chips. This goes to all, not limited to AMD. In the days of internet and instant information, it's sickening how they still treat us like morons.

On the other side, if that's the only way to stay away from gnusmas, it's kinda OK, but I really don't see how it would help them. I think it would be exactly the opposite.

 

Rebranding is a way of trying to trick people. Nvidia does it too, but they never go this extreme and have an entire new gen with nothing but rebrands like AMD plan to do now.
Just imagine if M390 is M290X rebrand. We have the following line of rebrands:
7970M 2012
8970M +50MHz 2013
M290X 2014
M390 2015

lol


Also I wonder if more than Dell this time will use M390X if its a M295X rebrand.

 

How depressing! I guess I will upgrade SSD first then...lol

 

Given that the m295x is unable to be crossfired, doubtful. Maybe some Clevo reseller, but most likely it'll just be Apple and Dell/Alienware (and I think Apple will abandon the m295x after their 8K iMac (rumor started by LG)).

 

Why not get Maxwell? GTX 980MX might hit mobile this summer although zero leaks are found. Damn it, wouldnt surprise me that the lack of AMD in mobile means Nvidia wont release a full GM204 because of that

Hm, thats interesting. I didnt know M295X lacked Crossfire.
Why not get XDMA to mobile then? Its better than CF anyway?

 

I remember from the m295x thread that somebody said it lacked the physical connectors for an XFire bridge.
However, according to Wikipedia,

So XMDA should be possible since the 295x is the mobile Tonga.
In other words, disregard what I said. I need to do more research.
But then one thinks, "Well why does someone need a 250w dual GPU solution in their laptop? And who is going to buy it anyways?" I mean, the MXM m295x costs 360 pounds from Zentrica. $1000 for dual m295x with free-sync and mantle support or $1000 for dual GTX 980m with a smaller PSU, DSR (unofficial) and MFAA?

 

Yeah I know, Nvidia have a huge advantage with Maxwell. If 980M cost more, they can just buy GTX 970M that will match M295X in price anyway and have better performance and heat and overclocking and the features you wrote about. Probably why only Dell is selling M295X I guess.

Yep, XDMA is basically a physical engine on the GPU that can communicate through PCIe with the other GPU to get the best scaling possible. But we may not know if its possible until 2016 because AMD seems to be more interested in freaking rebrands and we know Nvidia won`t have similar function (called NVLink) until Pascal is here in 2016.

 

If we go according to this generation of AMD mobile GPUs, then the m395x should at least be based on Fiji. The m390x could be based on the m295x .
Not sure about the m290x/7970m. m385x perhaps? Guess it'll be wishful thinking for the time being, I suppose.