NVIDIA Maxwell Speculation Thread

Denver is the CPU, but maxwell generation of GPU it will be used as controller.
Seems that would probably in maxwell-generation on the GPU side small OS of its own work, and thereby resolve the draw call.
Change of use of an existing game is not required.

I heard this bs on many forum. This is only possible if they make an API for it, that work exactly this way. But running the game on the CPU, and the API on the GPU is more problematic, than just creating a Mantle alternative.
 
Or to put it in another way: what would those arm cores be capable of (in regards to the draw call bottleneck) that you couldn't do driver side with regular cpu ressources? If the bottleneck was on the driver/gpu side of directx it would be easily solvable.
 
I don't get what is wrong about that sentence, other than being redundant. It is not talking about using Mantle on XBone, but on PC, which was already a given. Although I would consider the writing style a bit anti-AMD...

in my country we have a saying : "A madman throw a stone into a lake and five wisemen jump into the water to take it out" ... This is a Rumors&Speculations thread of course but a rumor must have at least a thin perfume of credibility to deserve a discussion around it ... those guy from "OnLiveSpot" mixed tech infos (whispered to him by sources, moles and little birds) with Star Wars lore and other things, so let me repeat myself : Are we talking about delusional now?
 
Last edited by a moderator:
in my country we have a saying : "A madman throw a stone into a lake and five wisemen jump into the water to take it out" ... This is a Rumors&Speculations thread of course but a rumor must have at least a thin perfume of credibility to deserve a discussion around it ... those guy from "OnLiveSpot" mixed tech infos (whispered to him by sources, moles and little birds) with Star Wars lore and other things, so let me repeat myself : Are we talking about delusional now?

In my land we also say that when someone criticizes something it must be substantiated. Care to elaborate where is the Star Wars lore there? I am not saying that the site is credible, but that sentence had nothing factually wrong in it.
 
Maxwell will be same as xeon phi and BroadCom's Videocore.

Even if the
Shader Compile on the GPU side
Remote procedure calls

Also, complex scheduling, such as dynamic warp formation, dynamic warp subdivision becomes possible.
 
I don't see any contradiction. :rolleyes:
On the first sentence he is talking about drawing calls.
On the second sentence he is talking about specific game effects.

So this "special sauce ARM" can magically reduce API overhead and drastically increase draw calls without any changes to the API or the engine/game but has to be specially coded to work in Nvidia's own middleware/library which would then have to have changes made to the game code?

I know you really, really want to believe but please put on your critical thinking cap for a minute.
 
The "1 million draw calls under DirectX without code changes" has probably even less truth to it than the pastebin "leak" a while ago. But he likely got the idea from there.
 
How about a speculative estimate how many clusters NV would have to sacrifice from the biggest chip of the family in order to get 16 ARM cores in there?

Well if the pictures of K1 are remotely accurate, the 2xDenver is at most 1/2 a SMX, so 4SMXs.
On the other hand, what can they decrease and/or remove from the die because of the Denver cores? That will be much harder to calculate and we may never get all the details if that is the case.
 
I heard this bs on many forum. This is only possible if they make an API for it, that work exactly this way. But running the game on the CPU, and the API on the GPU is more problematic, than just creating a Mantle alternative.

The way it looks remotely credible would be if everything runs on the Denver + GPU and the game
engine is built around it. i.e. like programming something specifically for a HSA APU.

Of course I agree it sounds like something merely wrote "16 cores" and "ONE MILLION BILLION BAJILLION DRAWCALLS!!" to get attention.
Dropping the Steam OS name in the article is suspect : it's sensible and Steam OS might technically run.. but what about Steam itself, and the games? There aren't that many games on Steam linux x86 already. I'm sure Valve would port its game to ARMv8 and it can be a long term solution but most 3rd party games could be "orphans". I can't even buy the Quake games for Steam linux i386.

I don't say desktop/laptop gaming on ARMv8 will never exist but I can't see how it would be announced so early.
 
ask at the expert ones: how can a cpu Arm eliminates a D3d11's bottleneck?

It probably cannot, unless the API is hacked in a major way. The problem many developers are facing right now is - if i understand correctly - d3d11's threaded rendering only shifts the bottleneck to a point where all the commands have to be queued into a single thread again to be fired off to the GPU. Now, if you had that much control in a GPU and if you could get Microsoft to issue an update to DX to allow the last queue construction to be eliminated (i.e. sorted out on the GPU die), then it could work. Don't know if such a thing would be possible in OpenGL via extensions or if you could emulate this on the GPU die.
 
If driver is the one communicating with a GPU, isn't it possible to have driver which gives jobs/data to ARM cores to do the job without need for DX/GL knowing?
 
Well if the pictures of K1 are remotely accurate, the 2xDenver is at most 1/2 a SMX, so 4SMXs.
On the other hand, what can they decrease and/or remove from the die because of the Denver cores? That will be much harder to calculate and we may never get all the details if that is the case.

Oh you mean the photoshop artwork? Besides why would you want to use performance and not area optimized CPU cores in a high end GPU chip in the first place?

The point was that it's likely not going to exceed the 8 cores mark for the top dog and of course that there's again an unbelievable amount of rubbish surrounding relative speculation.

Project Echelon was some sort of "what the future could bring" exersize". Early versions of it mentioned 16 CPU cores integrated, while after a while it bounced down to 8 and that was meant for 2017 and NOT for 2014/5 LOL.... :p
 
If driver is the one communicating with a GPU, isn't it possible to have driver which gives jobs/data to ARM cores to do the job without need for DX/GL knowing?

The problem is the DX API that sits between the application and the driver. Thus, the most optimized driver in the world still couldn't remove the bottleneck that DX11 injects in the middle.

The only way to circumvent it is to NOT USE the DX API calls in the first place, or else "fix" the scaling issues within DX itself. Hence, Mantle.
 
From tmall.com: GTX 750 specs. The image on the page containing the specs is also here (it's fairly large).

I can't read Chinese but it appears that the 750 has 768 CCs, 1020 MHz base clock, 1085 MHz boost clock, 5 Gbps memory, 1 GB, a 128-bit bus, and DX11.2.

There are also some OC benchmarks here.

UjGa12J.jpg
 
From tmall.com: GTX 750 specs. The image on the page containing the specs is also here (it's fairly large).

I can't read Chinese but it appears that the 750 has 768 CCs, 1020 MHz base clock, 1085 MHz boost clock, 5 Gbps memory, 1 GB, a 128-bit bus, and DX11.2.

There are also some OC benchmarks here.

UjGa12J.jpg

If this chip really does not require any external power it looks pretty damn impressive. HD6950 power for less than 75W?

EDIT - I wonder what's that "4+2" thing?
 
Back
Top