Maxwell is not DirectX 12 Next compatible

[Kaapstad]

Specification still under construction

This should not come as much of a surprise to most people in the loop, but Nvidia’s next generation GPU architecture codenamed Maxwell does not support DirectX Next aka DirectX 12. The reason is rather simple as the new DirectX next is still under development and Microsoft still hasn’t locked down the final specification.

From what we heard, DirectX Next actually fixes a lot of latency related issues that are present in DirectX 11 and earlier versions. The new DirectX should have lower driver latency something that developers have complained for quite some time but we are not aware of any major feature set that will come with the DirectX Next.

Since the Maxwell core launches in Q1 2014, probably March, support was not possible and most likely you will have to wait for Volta graphics to support it in a year or two. In case AMD’s Mantle delivers as much as 45 percent performance boost as AMD claims in Battlefield 4, this might put a lot of pressure on Microsoft to speed up development. This might force Microsoft to solve long standing issues including driver latency. On a side note, we don’t have any info on AMD’s DirectX plans, either.

Maxwell is expected to deliver huge gains in performance per watt and will put another flavour on the graphics market. It will also be used in next generation Tegra products and Nvidia is also planning to license Maxwell IP to other ARM players, taking on the likes of Imagination, Vivante and ARM's Mali business.

http://www.fudzilla.com/home/item/33638-maxwell-is-not-directx-12-next-compatible

 

[shankly1985]

It's going to be interesting how Microsoft tackle Mantle.

 

[andybird123]

There's two ways to look at this, the article seems to be saying that Maxwell cards definitely wont be dx12 compatible, but if dx12 hasnt been finalised the question and answer might just be that we dont know either way as you wont get a straight yes/no until after DX12 is actually finalised

Technically Kepler cards are not FULLY dx11.2 compliant, however they support all the major features and dx 11.2 games run fine on Kepler cards

Good news that MS are actually working towards making good on their blog promise

 

[Silent_Scone]

As in the first line, not surprised.

 

[Orangey]

How is Maxwell launching in Q1 if 20nm process won't be ready with timeslots available at foundries until Q3?

 

[LtMatt]

I hope Mantle ridicules DirectX 11 in terms of performance.

 

[MooMoo444]

I hope Mantle ridicules DirectX 11 in terms of performance.

This, simply so MS raise their game and we get something good. DX10 to DX11 didn't really seem to deliver much imo.

 

[drunkenmaster]

How is Maxwell launching in Q1 if 20nm process won't be ready with timeslots available at foundries until Q3?

It's low end parts, maybe for mobile only, considering this is the market Nvidia has recently done their best to pretend they don't care about at all and is meaningless to them... same way they didn't go for console contracts because they are worthless.

Low end 28nm parts, finally, but considering 20nm will take a while to mature(well in fact it till take a year to get to 16nm update then that will take a while to mature, making something now optimised for a updated and improved 28nm(compared to when it launched) then it's not a bad idea anyway.

When you only have 1/2 shader clusters, the shaders vs rest of core ratio is really bad, and several parts of that don't scale great with process shrinks, like mem controller. The highest end has the most to gain from process shrink, the lowest parts have the least to gain.

Then you factor in the low volume of 20nm for the first 6-12 months and then you're talking about 28nm being more available and even potentially cheaper. If you can only get 1000 wafer starts on 20nm, the low availability increases cost and you stand to make more profit from 1000 wafers of high end gpu's than low end.

Basically everything points to low end maxwells coming at 28nm.

 

[PhysicsMan94]

How is Maxwell launching in Q1 if 20nm process won't be ready with timeslots available at foundries until Q3?

The speculation is that low end Maxwell cards will be 28nm process based and will be released Q1, whereas the higher end will be released later (Q4 at earliest) and be based on 20nm process.

 

[Rroff]

I'm still extremely skeptical that Maxwell will appear on 28nm at all, they had the design kits from early on for 20nm and unlike previous processes there is no easy optical shrink 20<>28nm or vice versa so they'd have to redo a ton of work and debugging seperately, unless they were really really desperate they wouldn't even consider it as it would add another 1/3rd onto their development costs for Maxwell.

AFAIK production wise last I heard nVidia has production slots alongside qualcomm for the 2nd round of 20nm production as well as later production about the same time as AMD around Q3 (EDIT: Not sure if that was Q2 or Q3 off the top of my head) - with Apple having the first round with delivery inside Q1 then nVidia around the end of Q1 ish.


EDIT: One thing I do think is likely is that we will see GK110 rebadged under whatever designation they use for Maxwell as part of the first lineup as I suspect if there are 20nm parts they probably won't be able to produce a full lineup first time around.

 

[drunkenmaster]

That assumes they want to do a low end maxwell at 20nm. They aren't quick to a low end chip, haven't been for years. How long was it after the 280gtx launched they next did a low end chip?

Also 1/3rd for development costs for an architecture to tape out a low end chip on another process......

 

[Rroff]

Also 1/3rd for development costs for an architecture to tape out a low end chip on another process......

With the difference in lithography (and a few other differences in regards to debugging, etc.) required between 28nm and 20nm there are significant additional costs required to tape out a design for 20nm and 28nm compared to say the 65 to 55nm shrink of G92 which was relatively straight forward optical shrink. (Some stuff has to be layed out differently and so on).

EDIT: Its not the same as taping out 2 entirely seperate designs but with the difference between ArF and EUV etching and so on your going quite a few steps down the road towards the work required for taping out 2 seperate designs - hence theres a fair bit of extra outlay compared to a traditional shrink.