New details about Nvidia`s Maxwell

Here is for future reference to show that GTX 850M and GTX 860M is the same. Note that GTX 850M comes with both DDR3 and GDDR5.

 

Interesting... so, the 850m is just a 860m (Maxwell) that runs with a core 313mhz slower? Are all 850ms Maxwell, or are some Kepler? If so, buying a machine with an 850m and clocking the gosh-diddly-arn heck out of it (if it will in fact clock as high as an 860m) may be the way to go. In any case, it would open up options for those who want a Maxwell 860m chip, but their preference only has a Kepler chip.

 

850m is only available in maxwell. I think its a great card

 

Question: The 845m is listed under notebookcheck, but on Nvidias site it isn't. Is there any proof that the 845m exists? If it does performance wise, what would it match in the 700m series?

Also, sites like anandtech are claiming that the 740m DDR5 version should still trump the 840m due to the fact that it's ddr3 in terms of performance.. Most likely true?

 

What are you writing? those clock difference was between 2 gpu-z vesions. If you had looked better you would have seen that 860m has "lower clocks" not 850m

 

So, looking at those two images and not being an expert one could draw the conclusion that the 850 would outperform the 860? ...but that is not the case or?

 

Right, I'm confused as well... how is it the 850m has much faster core clocks and ~60% better texture and pixel filrate than the 860m in the GPU-Zs (while having the same number of shaders and ROPs)? Shouldn't those numbers be reversed?

 

Probably the boosting limits of the 860 are higher than the 850 , most likely the 850 doesn't have boosting

Sent from my LG-D800 using Tapatalk

 

Not sure about 850M but 860M boosts all the way up to 1020MHz.

 

I'm confused too, how is that proof they're the same chip anyway. ROPS/TMUs are different, only one of them says GM107, which part shows it's the same chip? (The 640 shaders and BIOS number that is nearly the same?)

 

It could be that the performance state that gpu-z reads is the 3d non boost in which the 860M is just set lower at. I would assume that once the clocks go higher the 860 would go to a higher clock. Also could point that the 850 doesn't have boost since it has a higher non boost performance state.

Sent from my LG-D800 using Tapatalk

 

I will jump on one leg screaming like a chicken if 850m doesn't have boost.
Don't forget about basics which can't be edited.

 

True , what I should have said was that it doesn't have as high boost as the 860 , also this is all speculation till the cards are actually in peoples hands and we can get proper info. But other that it doesn't change the outcome which is the 860 is faster than the 850

Sent from my LG-D800 using Tapatalk

 

Yes correct. If you find a cheaper machine with GTX 850M with GDDR5, buy that one. You should easily be able to change max clocks to the same as GTX 860M. Either just reflash with a new vbios or do it manual.
All GTX 850M is Maxwell.

The thing is that GM107 (GTX 850M and GTX 860M) have 2MB L2 cache which helps a lot with the bandwidth and performance. It the sole reason why GM107 which is low end can compete against mid end Kepler with 192bit bus. I`m not 100% sure, but according to the test I posted in the early pages in this thread, GM108 (GT 840M) have a sucky memory bandwidth because it lost to the GK107 GT 740M`s but beat GK208 GT 740M`s by a lot. GK107 had 128bit memory and GDDR5, GK208 64bit and DDR3. GM108 have 64bit and DDR3 and low L2 cache which doesnt help.
You should look for GT 845M for sure. It will be GM107 and should beat GT 740M by a good margin.

Sadly I have no GPU-z or anything regarding that GPU other than a couple of announcements from notebook OEMs displaying models with that GPU.
Even Anandtech and the rest got no information from Nvidia at launch regarding GM108. You might have to wait for a while.

EDIT: There was some mention that some GT 840M will be GM107 too.

GTX 860M have a base clock of 520MHz but it can go all the way up to 1020MHz.
GTX 850M score 4500 in 3DMark11 (GPU score) while GTX 860M is at 5000.

Same core count. Same memory bandwidth.
GTX 750 Ti also had the same ROP/TMU reads.
16/40 here
16/53 here

Those two GPU-z screenshots I posted is definitive proof that GTX 850M = GTX 860M
According to Anandtech both should have GPU Boost 2.0, but looking at that GPU-z of GTX 850M, one cant wonder if it runs at 863MHz flat.

 

off topic but this game will be a must buy Steam Greenlight :: BoneTown: Mature Edition

hope 880 will run it

 

a good chance that gtx 850m and 860m is exactly the same.
gddr5 gt 650m, 660m, 750m and 755m are all the same as well.

 

Hi
I don't know if those tests were linked before but here You have anyway: GeForce GTX 880M i GTX 860M – test nowych mobilnych kart graficznych Nvidii :: PCLab.pl. Some info about architecture and few tests. MSI Gt70, Xnote (Clevo) P177SM with gtx880m and GE60 with gtx860m. It's not much but always something.

 

I have a Clevo with 880m in hand now (NP8268) and the 880m does not boost, hardly ever, and cannot be overclocked through conventional means (i.e. Afterburner, nVidiaInspector, etc). That being said it's about on par with a heavily overclocked 780m with comparable thermals and power consumption. The 880m seems to completely turn off when not in use, resulting in better power consumption. I plan on running it using battery boost, and see if the technology even works or even worthwhile for this beast of a GPU.

 

Is here any overclocking guy who can explain how difficult is to overclock gtx 850m to gtx 860m levels. I am looking at MSI GE70 2PC or MSI GE70 2PE, price difference is
almoust 200 euros????????????????????

 

We won't know until we see actual parts in the wild. Per typical nVidia fashion, should at least be able to overclock to +135MHz limit on GPU with any software overclock utility. Beyond that will likely required a modded BIOS. But the +135 will take stock clock from 876 to 1011, close to 860m 1029 clock. With modded BIOS, who knows, all depends on quality of silicon and if you want to overvolt.

 

Cooling and quality of components on the system will matter too.

 

I think the GTX 850M and 860M are exactly the same minus the GDDR3 version of the GTX 850M and difference in vBIOS voltage tables to achieve higher boost in the GTX 860M.

Personally, I'm waiting on a Acer V3-772G refresh with the GTX 850M (hopefully GDDR5 version) since I was meaning the get the 2013 version with the GTX 760M anyways when it was on sale during the Winter Holidays.

The GTX 850M should be equal (maybe slight faster) to a GTX 650 Kepler Desktop both in performance and power consumption, will probably have a high overclocking potential too if it's indeed the same as a GTX 860M with a slightly different vBIOS (mainly the voltage tables and core clocks).

 

Great point Meaker. You have to take the notebook in to consideration wether you can run the GTX 850M on the same clocks as GTX 860M. An Ultrabook might not be ideal running 1GHz+ clocks.

Nemix also mention things that you might need to do to achieve same clocks as GTX 860M. GTX 850M might run on lower voltage than GTX 860M, so you might need to increase the voltage slightly to get the same clocks.

HTWingNut:
There was some mention earlier by sasuke that GTX 750 Ti have no limit on overclocking with Afterburner etc, if you have the newest driver.
Not sure if GTX 860M/850M is also included here, but one can hope since they are using the same chip

This is what the driver log says

 

The 850M and 860M will simply differ like the 4700MQ and 4800MQ, same silicon design but binned differently and with the frequencies coded differently yes.

 

About the gtx 750 to overclocking:

http://www.pcper.com/reviews/Graphi...p-EVGA-Galaxy-and-PNY-Overclocked/Overclockin



Sent from my LG-D800 using Tapatalk

 

holy ****. 1.4ghz. Is this like the highest clock speed for a graphics card??
btw, direct x 12 will be able to be run by all nvidia cards that are currently running on DX11.
http://blogs.nvidia.com/blog/2014/03/20/directx-12/

 

Thanks mate. Looks like its true, there is no +135MHz limit on these Maxwell cards

Yups, Maxwell is insanely good with overclocks. People can easily reach 1300Mhz with it. Something that was almost impossible with Kepler

Can`t wait until I get my 880MX

 

It seems they could get up to 1.6 or maybe 1.8 ... The consumption only jumped by 4 W from overclocking lol modded vbios will make the gtx 750 TI a beast lol

Sent from my LG-D800 using Tapatalk

 

If anyone is interested, Gigabyte have 3 new models with GTX 860M/GTX 850M
GIGABYTE Debuts New Gaming Laptops with NVIDIA GTX 800M Series Graphics

 

Good find!
The memory configs on the p34v2 seem to imply a kepler card. 870m in the p35 is also interdasting.

 

No it is a Maxwell GTX 860M.
4GB VRAM and apparantly runs on 1020MHz flat without any boost



http://forum.notebookreview.com/gigabyte/747756-new-gigabyte-p34-v2-2.html#post9601034

 

Yet another sub 1" laptop with 870m (P35W v2)! There had to be something to that. No way three manufacturers miraculously are able to fit a 100W TDP GPU in a slim laptop where they weren't able to before.

Beamed from my G2 Tricorder

 

Good to hear! That p34 will hopefully give the blade a run for its money.

 

i am a it confuse about the GTX 870M and maybe someone can educate me. So, why is there some laptop with 870M 3GB and some with 870M 6GB. I haven't been able to find any info about that so far

 

The 3GB version is soldered on the board. The 6GB version is the MXM removable board. That is the only difference in regards to the GTX 870M that we know so far.

The GTX 860M comes in 2GB/4GB versions as well and are soldered/MXM. The 860M 2GB is a Maxwell where the 860M 4GB is Kepler.

Confused yet?

 

Microsoft announces DirectX 12 for computers, smartphones and Xbox One - TechSpot

No Maxwell

More Info - NVIDIA: DirectX 12 Is Huge Improvement for Gaming - Guru3D

Microsoft Spills the Beans on DirectX 12, SupportedTitles Scheduled for Holiday 2015 - MaximumPC

 

but didn't they tell in other presentations, that all NV chips up from even Fermi (so the Dx11-chips) will be working with Dx12?

 

Ok guys, I think this would be of interest for all of us: Waiting on 20nm graphics cards from Nvidia and AMD? Don't. | TechSoda
The predictions could be wrong, but the problems listed are real. Other than TSMC being proud to deliver 10000 wafers/m, which is ahead of expectations, this doesn't help nvidia right now.
Fact of the matter is, I believe the most recent iterations of Kepler are not as far apart from maxwell as people think. nVidia has clearly done a lot of progress in optimizing the 8xx kepler series for very high clocks, while appearing to be able to keep TDP and power consumption under control, if the newly released laptops are any indicators.
Think of how powerful the 880m can overclock. It's not quite there yet, still a bit under by like 10-15%, but it's nearby in speed during gaming to the stock specs of the desktop GTX-Titan.
This is phenomenally impressive, in my opinion. I don't believe there will be a high-performance oriented maxwell release until 20nm finFET's are in full force, as we know that engineers were contracted in porting AMD's 28nm TSMC GCN-core fabs to GloFo 20nm. This takes time, and I'm sure the same is true for nVidia.

It's highly likely that the current 20nm line is simply not as advantageous in terms of cost vs performance offered than the existing mature technology on 28nm fabs, and so long as nvidia can do what the 750 Ti demonstrated, that is to make a gigantic performance per watt jump on a single architecture Tock cycle, then I believe this will push 20nm maxwell even further, while nvidia squeeze as much profit and performance as possible out of 28nm, so long as AMD doesn't posit a true contender to the scene. I think this is good and bad at the same time for the consumer. Performance-wishing gamers may have to postpone their high-end upgrade cycle a few months, since I believe the 290X and the 780 Ti will stick around a bit longer, but on the mobile side we get highly efficient Kepler with a high-performance design, and an even more efficient maxwell that's oriented toward the entry-level and mid-range line.

Logically, the entry-level and mid-range line are nvidia's biggest cash cows, and should be captured first. We can't really claim that performance hasn't increased in the high range as well, so long as AMD has no contender at the moment and, and even though it's another reband, the 880m has clearly made steps forward in terms of potential for high clocks.

So I think that for the next few months, the focus of our interest will be on 880m/870m overclocking, and seeing how far the 840m/850m/860m maxwell chips can go, and maybe keep on eye on AMD (who I think won't release a new flagship chip until 20nm as well, the 290X is still well positioned and they're making innovation with TrueAudio/Mantle that I greatly anticipate).

 

I'm sitting here with an 880m in a Sager NP8658 (thanks to LPC Digital for the review sample btw) and the thermals are on par with an overclocked 780m with comparable performance, only the 880m uses a bit less power, but most importantly it never boosts! And can't be overclocked with conventional means, a modded BIOS is required. This isn't a good sign.

I don't think the 880m at 28nm is a great solution at the moment. It's still too hot and uses too much power if compared with the 860m on a performance and temperature per watt basis.

Great article though BTW. If this is true, then I hope nVidia is busy redesigning Maxwell for the faster chips to fit on 28nm process. Because it would be a disappointment if the 880m is the cap for the next year or more.

 

I think it will be sometime too before we see 20nm in notebooks. I think late Q4 this year, which is what I've thought ever since I first read about this a few months ago. I too think the notebook GPU market has become too saturated with recent high performance releases to warrant a replacement for the 880M arriving anytime before the end of the year, especially given the lack of competition from AMD that you mentioned.

 

It looks like even the top end Maxwell will be 28nm. Curious the size of these chips then, lol:

" Nvidia recently launched their first Maxwell graphics card and for some reason a lot of people were really surprised to see it being built on 28nm again. Even so, the reasoning was that 20nm isn’t quite ready and the bigger, faster cards from Nvidia will be coming on 20nm at the end of the year. That makes no sense at all given the information we have.

It’s not like it’s a big issue anyway – the 750 Ti has excellent performance per Watt and that is partly the result of the now very-mature 28nm node. I have real doubts that Maxwell on the current 20nm would surpass this – and most importantly the cost (to Nvidia) would be far, far higher. I’m am absolutely certain that Nvidia will roll-out Maxwell top-to-bottom on the same 28nm process node they’ve been using up till now."

 

Where did you get those quotes HTWingNut, who said them? Earlier this week you might remember I was briefly flirting with the idea that Maxwell might be exclusively limited to 28nm given the x2 performance per Watt, but we agreed the chips would be too big to be cost efficient & that they would be 20nm for the high end chips.

EDIT: finished reading that article linked by sangemaru in post #939. Found out your quote was from page 2 of that article, from the author of the article Jim DOTCOM, so it seems that quote is his reasoned opinion. Oh well, I thought initially your quote might have come from an NVidia leak or something. I guess we'll see if Maxwell stays on 28nm. I think I'll stay on my idea of 20nm Maxwell for the end of the year for now.

 

Yea, same opinion here. Late Q3, or Q4 for 20nm maxwell, same probably expected for AMD cards, although it's interesting to see how GloFo will release in comparison to TSMC. I really hope they have their fab process up to par, I want to see Pirate Islands 20nm.

 

If you can't overclock then Clevo might have messed up their bios once again. Same for boost btw. The 880m boosts and overclock fine with the stock vbios. Which driver are you using and which vbios version is on the card if I may ask?

 

The author of that article is SiliconWars from Anandtech forums. Not exactly the neutral type to say atleast.

His arguments that price for 20nm wafers is more expensive than 28nm wafers is a moot point.
Price per wafer on 20nm will go up from $2500 to $3000. Yes I am aware this is 2015, but the graph show late 20nm production against late 28nm. One would assume same goes for early stages of both.


But SodaTech does not take in to consideration that you can make more chips per wafer. You can get around 160 GK104 sized chips out of one 12" wafer. $500 divided on 160 chips. Or even better, say 320 GK106 chips...

Second of all, a 100mm2 die will cost $2 more on 20nm than on 28nm. Say a bigger Maxwell chip in worst case scenario will cost $20 more. How can this not be countered by increasing the GPU price by $20? Nobody would care about that small price increase.



Third, we need 20nm to build the bigger Maxwell`s. GM107 is 35% bigger than GK107 since they both are on 28nm. Nvidia can`t build a very big die. Also the bigger die, the less effective it becomes. 20nm is crucial here to make them more efficient.

Jon Peddie have this to say:

Next generation is GM206, GM204 and GM200.
Nvidia delivers next-generation mobile gaming GPUs : GraphicSpeak

Another important point is that if Nvidia released 20nm Maxwell in late 2014, the 16nm FinFET would be way too close. TSMC is already doing risk production now and mass production will begin early next year. Which means 16nm would be available a year after Nvidia released 20nm.
Does not make sense. It usually takes several years between each node jump. Nvidia is already late as it is today, so you better be sure they are working hard on getting the 20nm out to public.

Trust me, 20nm will be here in the summer. Anything else would be stupid.

 

Cloudfire is right. I mentioned it as well, Maxwell 20nm is imminent for summer release. There is a reason why only gm 107 showed up and is called 1st gen Maxwell.

A bit more info:

AMD: I was PMed about saying that I was wrong and dx12 demo was run on Titan. Ok, lemme clarify, the DX 12 what was shown to be public is not the that was introduced behind the scene.

Maxwell: Got a hold of with an old secured "source", First 20nm Maxwell chips will arrive as early as May, with availability in june/july time. We are now in pre mass production step, but it is about to change in a few weeks time Also, the 880 desktop numbers I talked about earlier are no no longer valid.
Since the card with these performance is no longer called 880, it stands below it. Oh, boy, if I could only share with u guys. But a bit by a bit..

 

May you say? Well that is good news indeed. Leaks are probably bound to happen anytime soon I hope.
Can`t wait to see what we get. I just hope Dell is coming out with higher res displays than current 1080p.

880MX SLI will be sick overkill

 

OK folks. You've convinced me that anything is possible and summer release is quite likely.

I have $2000 saved up. I know that will be enough for an MXM 880MX but will Clevo be compatible with the new card? That's what I'm afraid of! So if not, that $2000 will have to be used on a whole laptop. No easy feat in this darn country where prices are a third higher than in NA! At least I won't have an SSD to worry about this time around, saving me a couple of hundred. And the Bluray drive too. And the RAM. Might be tempted by Broadwell if it's out - that DDR4 power!

Don't you just hate the waiting game?

 

Why not just buy a pair of discounted plane tickets in advance for like 50$ and fly over there and purchase what you need when you need it?

 

I know for a fact that Atlantic flights cost more than 50 USD!!!