New details about Nvidia`s Maxwell

does Germany even bother with gpus?

should it support EU union that lacks expedients for a proper self support?

 

I seriously seriously doubt that they will leave whole Maxwell on 28nm. If they were going to then there would be no delay with top-end Maxwell line. No way.

 

ckoro grantu Ukraine.

Sherla merla da pomerla

 

No. GTX 750 Ti is a full GM107

 

only gm 107...

 

How do they know GM107 is a part of a different chip? Sounds like vague guessing to me.

Take GK106 for example. One would of course guess that bigger chips will arrive later, but GK106 is not a part of GK104

 

??

Gk 104

Gk 110
.
Only differences are in hardware level.

Something like wider bus, etc can be a proper full version and gm 107 is a cut down version.

gk 106 and gk 104 are both low/mid end parts. But gk 104 is stronger, so yes, there might be bigger Maxwell on 28nm but not all of it..

 

Until Lenovo decides to go ahead and SLI it anyways.

 

LOL.
Lenovo sli-ed the gt 750m and the gt 650m.

 

? care to explain that or just trolling?

 

They didn`t make GK110 first, chop off some parts, call it GK104 and sell that one first.
They go step by step, mature the process and learn along the way, then make bigger chips later.
GK106, GK104, GK110 are all individual chips. None of them are based on the other. They make many ES of each of them, then move on to QS once they found one that meet their criterias.
They do however make a chip, say GK104, then make different GPUs out of that chip, where they remove memory bus, ROPs, SMX, cores etc. But they never use a GK110 (15SMX), chop off 7 SMX, and call it a GK104.


Hence I would like to see why anyone would suggest that GTX 750 Ti (GM107) is a cut down chip from any other upcoming Maxwell chip. Which review said that?

The small chips are the most efficient ones. Once you increase die size, increase core count, that efficiency would go down.
Nvidia made a chart specifically for 1st generation Maxwell (GM108/GM107) and I think it was a reason for that.
They could do 28nm on GTX 750 Ti because its such a small chip. If they did GM104 on 28nm, I`m not so sure if we would have gotten that +35% result compared to GK104.

Their options is:
Make GM204 in 28nm, have a small increase in performance. Enough to market Maxwell as something really good?
Make GM204 in 20nm. Have room for much more transistors, ie cores. 20nm alone should give 20-25% better power consumption. Which they would use toward unlocking better performance.

Personally I think 20nm is the way to go. And TSMC have said that the one process they are using now to create 20nm wafers is good enough for high performing chips as well (they saw neglible differences between Low Power and High Power). I`m positive that some of the wafers they have been producing now for 2 months is being purchased by Nvidia.

 

Cheers for the reply Cloudfire. I thought that Maxwell was nearly 100% more efficient than Kepler on a Watt for Watt basis at 28nm from the initial testing? I also thought that a move to 20nm would nearly double performance too due to being able to fit nearly twice as many transistors per mm squared? If that was all true, then NVidia could just stay on 28nm and get twice the current performance then, and then if they shrunk it to 20nm then they could get 4x the performance of current Kepler. Did something change from the intial testing of Maxwell when the reviews came out for the 750, because I could have sworn those were the conclusions from it? Are you just saying that those performance gains won't scale with bigger chips then (bigger chips than the 750 for instance)? I don't really understand why it can't?


EDIT: Think I realise now why the high end Maxwell can't be 28nm for twice the performance. HTWingnut said the chips would be too big, and I can picture that now, they'd have to nearly twice the size to get double the performance, even if the TDP would be the same (due to the x2 performance per Watt). Yes, so I've got that now! I can see how the next generation of high end Maxwell chips would need to be 20nm just from a chip size point of view rather than a TDP point of view!

 

OC3DTV GTX 750 Ti review.

Desktop gpus do have full part and later being cut down.

Does not matter is it is just an ES or not.


Is there still someone questioning 200 series being 20nm parts? please..

 

Guys it seems that many manufacturers struggled to adapt with the 28nm maxwell integration, as you have probably seen the 860 right now is only sold by 2 manufacturers Clevo/Sager, and MSI, the rest are just promising or putting pre-orders, it will take lenovo till may and possibly alienware, if you add that to nvidia's milking attitude, you would expect only heavy hitters from 28nm maxwell in computex and after the manufacturers finally cope with 28nm maxwell, they will be less likely to rush in 20nm and restructure their laptops once more to accommodate it not to mention that the 20nm's would cannibalize the previous 28nm maxwell's profits..

 

Source?

Not sure how this would cannibalize nVidia's profits since both are nVidia's chips. Again, if there is a surplus, they can rebadge. Greed is good.

 

The GTX 750ti is fully enabled GM107. Maybe you're thinking about the GTX 750 which is cut down GM107, but the 750ti is not.

 

I provided enough to defend myself.

There will be more 28nm Maxwells, I just got a confirmation

 

GTX 750 Ti which is a GM107, is 75% better than GTX 650 which is a GK107. Source.
At the same time, GTX 750 Ti use the exact same power as GTX 650. Source.

Not entirely 2x the performance per watt as Nvidia promised, but not far away either.

BUT here comes the catch. GTX 650 (GK107) is 118mm2. GTX 750 Ti (GM107) is 156mm^2. There is 1.3 Billion transistors in the GTX 650, and there is 1.8 Billion transistors in the GTX 750 Ti. Nvidia had to go up 33% in die size to fit those transistors because they used the same node. The perfomance comes directly from that transistor increase.
One would assume similar would happen with high end as well. GTX 680 (GK104) is 526mm2. If Nvidia built a GTX 880 in 28nm, they had to increase the die size to 700mm2 to see similar gains. Not ever happening. It would just be too big. The reason why Nvidia could do 28nm on the low end Maxwell is because they have room to spare to create these bigger die`s.

That is why Nvidia have to use 20nm for the high end chips. It may be possible to create the midrange Maxwell in 28nm because they are not that big, but I am sure Nvidia would want to harvest the power consumption improvements you get with building them in 20nm as well.

 

____________________________________________________________

Cloud, if that article from the link posted was accurate, it means that high-end 20nm Maxwell for gaming laptops is a LONG ways off, and my guesstimate based on that article, is that we won't be seeing any of those made available for high-end gaming as successors to the 880m until 3rd qtr 2015.
Aside from that, it makes no sense for Nvidia to release a 28nm 880m, and then a 880mx or 980m chip 3 months later.
I would love to be wrong and see these HP 20nm's out this July in my next Sager gaming system, but I just don't see it happening.

 

lots of good info in this thread

 

nVidia should release 20nm in Summer/Fall, and leave ATI in the dust in terms of performance. It will take a long time for ATI to catch up.

 

ATi doesn't exist anymore.

 

LOL. Point taken.

 

Pretty interesting that Nvidia have been testing 20nm chips since April 2012...

THE IMPACT OF PROCESS TECHNOLOGY ON KEPLER

 

I won't listen to that nonsense -- you just shut your mouth ATI will never die.

Can we please go back to 2009 where nV actually had competition? Even better, can we go back to 9800 pro days?

 

I cannot wait to see what you guys can do with Maxwell on an even more advanced manufacturing node!

△ ▽
•
Share ›













Avatar

Joe Greco > BestJinjo • 2 years ago





Thanks for your note BJ. I wouldn’t want to delve into the waters of competing foundry technologies here. That topic is ample enough to be a PhD-level dissertation by itself.

We will satisfy your curiosity on your last point pretty soon

Links to March 25th conference.

 

The WAN Show: New Razer Blade 14 Popcorn Time is Gone?? Haswell-E Coming Soon?? - YouTube
About 1 hour 18 minutes in, Linus mentioned a MAXWELL 870M!!!!

 

That guy must have read my post here from 13-th of March! And video is uploaded on 14-th LOL



Just keeding of course, these thoughts are not that unique. Although it's cool when you are not alone in some suggestions. Even if future will show that some of them were wrong.

 

If there was a Maxwell 870m it would make no sense for nvidia to hide it.


Sent from my bq Aquaris 5.7 using Tapatalk

 

Yeah I highly doubt one exist.
It kinda would make sense that Razer stuffed a cool Maxwell in their 14" and not that hot GTX 870M Kepler, but it would be based on a entirely different chip than GM107 since 860M is a full GM107. And neither GPU-z or AIDA logs have updated any support. Neither have I heard about any. None of the reviewer sites got any other information from Nvidia either.

 

Well it's quite perplexing how both Razer and MSI are able to fit that same GPU in a very thin form factor suddenly when they had enough trouble cooling the 765m.

 

I have to agree with you HTWingNut. We are talking a GTX 680M @ 1GHz here... Strange indeed.
Maybe some new mid end Maxwell are bound to hit us after announcement March 25th when the CEO from Nvidia have its yearly talk about future GPU technology...He will talk in the upcoming GPU conference in a week.

 

On another note, 20nm production is expected to be ahead of 28nm production with 30% just like previously forecasted

They have reached between 50-70% yield rate with 20nm production now. Very good for being only 2 months in production. This didn`t happen until month #6 with 28nm. Only having one process (LP, Low Power) instead of 5 different processes with 28nm, seems to help with production.

I`ve played around with some numbers and:

@ 10 000 Wafers Per Month
Nvidia could make 185 GM204 chips per wafer if they were the same size as GK104. Which means 1 850 000 GM204 chips per month if they bought everything. If Nvidia only bought 10% of those 10,000 wafers, we are looking at 185 000 GM207 chips per month

 

If it is Maxwell 870m, I might be considering the MSI. But if Razer and MSI can do it, maybe Auros will offer an SLI 870m thin form factor rig. I'm beginning to think that Clevo needs to rethink their game. They do offer great bang for their buck, but with GPU TDP's dropping significantly, I wouldn't mind seeing a laptop or two in a thinner form factor with longer battery life.

 

Still very sceptical.
Linus does say GTX 870M will have Kepler version and Maxwell version in the video. He could have thought about 860M but who knows.

We will see

 

U are waiting for Maxwell on 25 of March? 790 will be your Maxwell card and the event kicker.

 

Sounds like GTX 870M is Kepler to me.

Hands-on: Updated Razer Blade and Blade Pro laptops are slim, more supercharged | PCWorld

Hoping for an announcement of atleast future plans yes

 

added needed info

 

You mean MSI and clevo

dazing and confusing via taptalk

 

since when was clevo thin?
He was talking about razer.

 

No, Razer and MSI have thin (sub 1") laptops with powerful components, and it'd be nice to see Clevo try to get in on the thin and lighter side of things, i.e. better thermal engineering.

 

Hoping that it is true about the GTX 870M.

Like I hope for last time :
MXM GTX 870M 6GB - Kepler
Soldered GTX 870M 3GB - Maxwell.

This happened to the GTX 860M.
MXM GTX 860M 4GB - Kepler.
Soldered GTX 860M 2GB - Maxwell.

It's the RAM we're looking at.

MSI and Razer will come with a GTX 870M 3GB.
GTX 870M Kepler leaked GPU-Z is 6GB. Which is predicted 3 months ago to be an MXM on a Clevo's Roadmap.

Just my thought.

 

So far:
MSI GS60: soldered 860M 2GB, Kepler
MSI GE60/70: soldered 860M 2GB, Maxwell
Clevo W230SS: soldered 860M 2GB, Maxwell

It's hard to predict anything at the minute, there's simply no consistency... the only appealing laptop to me right now is the new Gigabyte P34G which will come with 4GB 860M likely to be Kepler ... if not, this is the one.

 

You are right about the MSI GS60.
Too bad.

 

Putting Kepler in a thin and light when you have Maxwell at your disposal is just WEIRD. Yes, I'm talking to you, MSI.

 

People, ask yourself a question: Why would everybody already know everything about Kepler GPUs in 8XXM series and low-end Maxwell chips (no DX12 support) while everybody is still unclear about some other laptops?
Just speculation of course so far.

 

Since the MSI GS60 is already confirmed that is equipped with a GTX 860M Kepler, a GTX 870M Kepler on other thin notebook just doesn't seems weird anymore.

 

Likey the Kepler version was the only chip available when the those laptops were in design and QC -- probably why the laptops with Maxwell 860m are, in general, arriving a bit later. Moreover, I'm guessing those manufactuers who went with Kepler chips got a slight discount since there were more of those lying around than the brand-new Maxwell.

 

Maxwell is 128bit so no 3gb, it would be 2GB or 4gb.

 

Lets put our old Kepler in the most expensive new notebook from us and we will put the new Maxwell in the crappy GE series.
That makes sense...

Nah, I know they probably use MXM card inside the GS notebooks, hence they only have Kepler available to use. Still sucks when you think about heat and stuff. Why not work with Nvidia and get approval of a MXM based Maxwell 860M? Maybe that break the whole "No SLI allowed" thing they got going on about the 860M? I dont know.