AMD will use 28nm for their GPUs in 2015

Forget about 20nm.

GlobalFoundries only have 28nm HP process for GPUs, which AMD will use in 2015. So both Nvidia and AMD will use 28nm process throughout 2015 and won`t switch to newer nodes until 2016.
AMD = 28nm SHP Global Foundries
Nvidia = 28nm HPM TSMC

AMD to Switch to GlobalFoundries' 28 nm SHP Node in 2015 | techPowerUp

 

Well hopefully the SHP process allows AMD to increase clock speeds due to lower voltages and helps to increase the performance.. If they can get HBM memory in, it should make up for the shortfall from not using the 20nm process...

 

To Bad, no new Notebook for me this year :/
But ist good to, no need to decide between broadwell and skylake

 

If they can roll out an efficient architecture like Maxwell it will be good.
But for those who was waiting huge leaps will probably be dissappointed. Smaller transistors is required for that

 

There was a rumor from Chiphell that only the 390X will be on 20nm, and everything else will still use 28nm. Maybe that really was just a rumor. However I definitely remember reading somewhere that this SHP 28nm from GloFo will have ~30% performance increase compared to existing 28nm chips.

 

As long as it is MXM non soldered, we should be ok.

 

The wait is torturous! Come on AMD release something already!

I guess we might hear something very soon I hope! New year is approaching afterall......

 

I would take all this with a grain of salt... The Italian website that TPU cites says that 16nm GPUs expected near the end of 2015/beginning of 2016, so there's still the possibility of a new node in 2015. Also, the original article doesn't have a clear source and reads more like a bunch of deductive reasoning than an actual leak.

 

There's always the possibility to skip a node

Nothing out of f@#$in' nowhere to be honest, those damn smartphones and tablets are the sole reason for pushing back everything else but their SoCs manufacturing. Soldered crap indeed, and you, yes you, who buys every new phone they throw at you, you are even more guilty for making this happen. So don't whine about everything else being hanging.

 

I'm still using a flip phone so don't yell at me.

 

Tape outs for 16nm is expected begin late 2015. But add another 2-4 months for enough inventory to finally launch the products to us consumers.

 

These are all rumors so take them as they are, rumors. As far as nVidia is concerned they won't have anything besides 28nm for the entire year. AMD plans are a bit more complicated. AMD themselves claimed that they will have 2 (if my memory isn't betraying me) different 20nm products, one being 20nm console chip shrink (if that even counts) and the other is a mistery. Since Carrizo and Carrizo-L are confirmed 28nm that leaves low power server CPU/APU and GPUs as the only possibility (semi-custom obviously in the same boat as console chips). There are conflicting leaks that claim AMD will have 28nm SHP and 20nm GPUs next year, but this may not be conflicting at all. There is the possibility of them refreshing some GPUs (370X and below, this includes mobile line) and maybe some new ones (new mobile GPU codenames floating around) at GloFo 28nm SHP very early this year (Q1) and then making 20nm GPUs for high end (390X, not sure where 380X goes since it was supposed to launch in February but may be affected by 20nm delay) by late Q2. Note that those who claim that there will be no 20nm GPUs base their claims solely on the fact that TSMC 20nm isn't suitable for GPUs but then there is Globalfoundries. I believe that AMD just gave up on TSMC completely for new chips in 2015 and will rely completely on Globalfoundries as they recently praised GLoFo for their improvements in management which led to best relationship between the companies yet. Also don't forget the leak of AMD GPU plans last year where GPU designs were being transferred from TSMC to GloFo and all the newer chips were based on GloFo and TSMC was completely absent. There is also quite significant amount of evidence that they dropped TSMC altogether even in 2016 and beyond (with the exception of ARM chips possibly) and the biggest factors are Samsung/GloFo alliance which led them to beating TSMC 16nm+ to market with their 14nm LPP and don't forget the fact that AMD was completely absent from TSMC's list of 16nm customers. So no, most likely there won't be AMD 16nm at all and nVidia won't have 16nm GPUs in 2015 either.
As for who will be first to release FinFet GPUs (AMD 14nm vs nVidia 16nm) is hard to say, as GloFo claims their 14nm will beat TSMC 16nm to market but AMD's GPUs will still be relatively new by early 2016 so I'm not sure if they will wan't to rush things while nVidia will be desperate because their GPUs won't be very competitive by then (especially Tesla cards since Maxwell is unsuitable for DP and they won't be used there according to a nVidia employee). There is also the fact that AMD claimed that they won't be on the leading edge of manufacturing process due to cost and yields and as always mobile chipmakers saturating the entire supply early on.

 

I'm still using an iPhone 4 so don't yell at me.

Ah, flip phones. Those things had battery life for days!

 

Can I get a TL;DR of this unpunctuated ramble?

Also, the mention of Tesla at the end there, is that a typo? Tesla architecture came before Fermi.

 

I just don't see AMD matching Maxwell with lower voltage and increased clock speed for some reason, fingers crossed for them though (however this seems like bad news for AMD).

 

No, I was talking about supercomputer GPUs.

 

GG Nvidia product naming schemes.

 

I also read that GloFo are working on a different technology as well - FD-SOI (cheaper than FinFET, easier to implement, scales good), so it can only get better for AMD with their siding with GloFo.

Same here, as well as Nokia N93, this thing with its old battery lasts a week, granted I'm not pushing it as much, just calling.

 

Well maybe that's what the 20nm process for high end GPUs will be.

Still using a Galaxy S3 but it is showing it's age, especially the battery. Want to get a Note 4 to replace it but I want the Exynos variant, tired of Snapdragons. The problem is US keeps getting the Snapdragon variants so it will cost me $700 to buy the Exynos variant from Ebay.

 

Hey hey what's wrong with Snapdragon? I'd never buy an Exynos varient phone again.. Poor dev support so its pretty much impossible to get a custom ROM and massive lag fests... I had a S3 (international varient) with this processor and totally turned me off.. Snapdragon processors have been epic for me the last 2 years.. Anyways you might as well wait for the Note 5 or the Note 4 with the Snapdragon 810.. That will be fun

 

Not everyone cares about custom ROMs, it's just a phone at the end of the day, not to mention that you are throwing all that extra functionality on a Note device by doing it. As for lag Exynos is faster than the cr@p that is Snapdragon 805 so it won't lag.

Also about the S3, I don't think the Snapdragon version is any better. 810 is good because it doesn't use the weak old Krait cores based on A15 but that version of Note 4 will probably be overpriced and released only in Korea.

 

Custom ROM's can be quite good actually.
In reality, they can easily speed a smartphone UI up by a sizeable degree where there's a noticeable difference for many users.
Plus, certain improved functionalities such as more optimized rendering of the applications through the GPU or games, or a number of other things such as optimized OS for battery life.

Granted, I got myself the THL W200 phone, and while it was good for the price I paid (still is mind you), I later on learned it doesn't have the option for upgrading the ROM because MTK never released the source code for their SoC's.
Overall, I was able to clean up the stock ROM by removing all of its pre-installed Chinese apps that I never used to begin with (without affecting functionality), and the phone has been running smoothly.
I don't like that I'm stuck on Androind 4.2 (for now), but still, it still works quite nice.

Anyway... back to the topic at hand.

I've read that AMD will mainly use 28nm for most of its products, but that only 'some' of its products will be done on 20nm.

Sigh... oh well, we'll have to wait and see what AMD comes out with in a month or so for the mobile section.
I'm hoping it will be HBM.
If not... then I might as well wait for Broadwell to come out and see what will be on offer then.
It's not like I'm in a hurry (just yet).

 

HBM will be the only thing that can help them.. If they don't include it, NVIDIA will just trample them...

 

If R9-M295X is any indication, I'm quite on the opposite side. HBM would only help, indeed, but it wont be make it or break it. We'll see. I still have some savings to do, so I'm not exactly in a rush.

 

Agreed.. The R9 295MX hardly inspired any confidence in AMD for me... I just hope they don't make anything that an overheating POS like the 880M..

 

Well, even if AMD uses HBM for a modified R9 295MX, it would still mean a sizeable decrease in chip size and power consumption (from HBM alone), while at the same time possibly increasing performance (especially the bandwidth).

But, in that case, I think it would be more sensible for AMD to release HBM on a new architecture as opposed to sticking with Tonga, because I think AMD is way overdue for a new architecture that's more efficient.
I certainly wouldn't complain if they released an architecture that's comparable to Maxwell AND has HBM on top of that.

 

If it were the latter case with new architecture+HBM, then NVIDIA would be in some trouble

 

So I just saw this in an article then checked AMD's website and confirmed this:
"4K and beyond – AMD will be demonstrating exceptional ultra-high-definition experiences at 4K resolutions and beyond, spotlighting 4K gaming using AMD Radeon™ R9 graphics and AMD Eyefinity technology, the Apple iMac with 5K Retina display powered by AMD Radeon R9 graphics, and the new Alienware 15 notebooks with 4K display support powered by new AMD Radeon R9 M295X graphics."
http://www.amd.com/en-us/press-releases/Pages/2015-ces-2015jan05.aspx

So a new Alienware 15 with R9 M295X (hopefully properly cooled this time) so we can see what it can do in a proper setup, but makes me wonder if they release this now won't it just get outdated really quickly. The desktop R9 300 series seems to be promising with rumors claiming R9 380X to be faster than 980 (not very hard to do) and 390X being 33% more powerful than that, and most likely beating GM200 thank to HBM. All that will launch by May apparently (some rumors claiming 20nm GPUs being delayed by 2 months from February/March to April/May) so I wonder what the mobile market will get. They already have new mobile GPUs codenamed Litho XT (MXM A) and Strato Pro/XT (MXM B) but these are slower than M295X and will be midrange cards. These will probably be 28nm parts and launch sooner but there will probably be a new high end mobile GPU with HBM later on (by summer) based on 20nm process if something doesn't arrive earlier based on 28nm.

 

hmm? regardless of what their desktop is their flagship mobile GPU is still tonga based, which means whatever you see will be around the r9 285 range, and that's probably gonna only wrestle with 970m, optimistically.

 

Praying for a return barrage of fire from AMD, nVidia has been winning the war for far too long. If the new cards run as cool as the 7970M and perform as a 980M, I will be happy

 

I'd add 'be cheaper than a 980M' to that list.

But yeah, I agree. AMD needs to bring some heat to compete (or...cooling...I guess), especially in the CPU and mobile GPU markets.

As a sidenote, do they have any plans for new mobile CPUs?

 

Can you link the source of that? It's not like everyone in the HPC market must have DP.

 

Well the most info so far for their new APUs can be found here: First Look: AMD's Carrizo APU Notebook Design.

 

Here you go NVIDIA Planning To Ditch Maxwell GPUs For HPC Purposes Due To Lack of DP Hardware - Will Update Tesla Line With Pascal in 2016, Volta Arriving in 2017. And since when does anyone place GPUs with anemic DP performance in supercomputers.

 

Thanks.

Since the very beginning of time, when GPGPU involves remapping OpenGL buffers and DP on GPU didn't exist.

At least we'll get Maxwell Quadros right?

 

Hmmm... Thanks for the source.
I don't like how they're going on about gaming power with an APU...

 

Well if they use HBM with FX Carrizo APUs it will be impressive, might run games at medium-high setting at 1080p. The FX-7600P is already quite good, sadly though it's bottlenecked by memory bandwidth.

 

Yes but they won't be getting into any supercomputers.

 

Guess Nvidia is hedging its bets on Pascal and NVLink for HPC.

 

Well so far NVLink only works with IBM and they just dug an even deeper hole with Maxwell. The reality is FirePros have been the superior cards for compute ever since GCN was released so AMD is slowly but constantly gaining market share. The fact that they top the Green 500 list and highest ranking nVidia GPU based supercomputer is only 3rd on that list (I don't know what the second place uses but nVidia can't even compete with that) with the FirePro S9150 based server being 18.5% more efficient than the most efficient Tesla K20x server despite being larger and using higher clocked Xeons (means less energy efficient CPUs at 3GHz vs 2.1GHz) means that more server makers will be interested in them. Looks nVidia are betting it all on NVLink, not sure how that will help though once there is a proper PCIe based NVLink competitor.

 

Uh, no. AMD has by far the worst OpenGL drivers of the big 3 IHVs (AMD, Nvidia, and Intel). Their OpenGL driver team's ineptitude is laughable. And FirePro still has barely a fraction of Quadro's market penetration.

 

AMD employee on LinkedIn: Desktop R9 380X is 300W with stacked die and getting ready for production.

Swedish summary: AMD-anställda röjer Radeon R9 380X – nytt flaggskepp med staplade minnen - Grafikkort - SweClockers.com
German source: 3DCenter Forum - Einzelnen Beitrag anzeigen - AMD/ATI - Pirate Islands: Bermuda, Fiji, Treasure | mobile: Strato, Litho - 2015

 

Good find Mr Naisman.
It says the GPU is "largest in the king of the hill". Plus the TDP is 300W (+50W from R9 280X), so this pretty much confirms 28nm for 300 series like the original article in OP says

Better be good performance from that GPU or else they have nothing to counter Maxwell with considering the TDP have increased?!

 

If 380X is 300W, what's 390X gonna be?

 

Well 290X is 290W. Maybe 340W? To take up the fight against 165W GTX 980
This does not bode well for mobile GPUs. And both 380X and 390X will probably need to be bundled with water cooling/hybrid.
It almost sounds like they are maybe using the same architecture as 200 series but increasing the core count (hence the TDP increase) and adding HBM (stacked memory).

Ugh I hope this is not true. AMD need much more efficiency to have a figthing chance against Nvidia

 

AMD won't have mobile GPUs until a die shrink, so for now, they at least should unleash a top performer on the desktop side.

To be honest, at current prices, 290x is still a good GPU.

 

Its sad to see AMD falling behind like this. Nvidia didnt need die shrink to make Maxwell?!

GTX 970 got 290X prices covered, with GTX 980 for those who want the best. Thermal and noise is greatly reduced with 970. PSU requirements and size of rig is reduced with 150W GTX 970 against 290W 290X.

GM200 will further be built with Maxwell cores. Against 390X it wont be pretty, if 390X even got a chance if they continue with the little efficient GCN architecture. There is a limit on how much heat they can pack in a die until they got to quit the race because they meet a thermal wall

 

I don't think 380X is a 300W card, at least there is nothing to indicate that. The 380X and 300W are completely different things, the 380X is a finished product, while the 300W 2.5D HBM card is an Engineering Sample (maybe for 390X). Still even if 380X is 300W most likely 390X will be 20nm and use less power (this case is much less likely than the first).

 

So apparently 390X will be on 28nm as well?!

If true, 2015 is going to be one boring year.

 

Well I dunno if the new AMD GPUs are even using a next gen GCN or not. Maxwell is a new architecture, which was already an improvement on its own regardless of die shrink.

I am not saying R9 290x is better value compared to a GTX970 etc, merely that it is competitive performance wise. I don't think the average user cares that much about the very low TDP of maxwell so far, only people that do tweak and manually overclock their GPUs etc.

Sure AMD is behind in performance per watt, but the important part is that they are still competitive performance/price wise. New archs and dieshrinks will take care of their insane tdp hahaha