why nvdia still not making big monolithic CPU with ARM ?

wahdangun

Golden Member
Feb 3, 2011
1,007
148
106
we know that NVDIA is used to build big monolithic GPU but why nvdia can't build the same thing with ARM ???

why i ask something like this, because i growing tired gaming with windows, I want something new like android or linux to be my new gaming platform (and no i don't think angry bird is sufficient).

if nvdia can make that happen i will glad to build ALL nvdia gaming rig.
 

Abwx

Lifer
Apr 2, 2011
11,846
4,806
136
Because ARM CPUs are not boxing in the same category as X86 CPUs ,
not even comparable with an Intel Atom....
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
What would a "big" ARM chip do? What algorithms would Nvidia be adding to the chip that would require billions of xtors? What are you thinking all those billions of xtors would be put to work doing in the microarchitecture supporting and an ARM-based ISA?
 

jones377

Senior member
May 2, 2004
460
64
91
What would a "big" ARM chip do? What algorithms would Nvidia be adding to the chip that would require billions of xtors? What are you thinking all those billions of xtors would be put to work doing in the microarchitecture supporting and an ARM-based ISA?

It would actually be a pretty good solution for Nvidia's effort into HPC. Running the OS on the ARM core(s) and CUDA on the GPU cores. It should make for a cheaper server as there would be no need to attach it to a Xeon or Opteron CPU. They'd need an I/O bridge though, and they gave up that development many years ago. Essentially the GPU daughterboard of today would become the motherboard in such a system.
 

Genx87

Lifer
Apr 8, 2002
41,091
513
126
I dont know if it is worth designing a large ARM processor just for HPC. Their Tesla designs share R&D with their quadro and consumer lines. I dont know if they would get much sharing between Tegra and an HPC version of Tegra.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
I dont know if it is worth designing a large ARM processor just for HPC. Their Tesla designs share R&D with their quadro and consumer lines. I dont know if they would get much sharing between Tegra and an HPC version of Tegra.

Maybe I am just totally missing the premise of the OP.

I read the OP to be akin to someone saying the following: "why Intel still not making big monolithic CPU with 486?"

Q: What would a 1B xtor 486 CPU actually look like?

A: Nothing like a 486 CPU.

So I'm curious what the OP thinks making a "Big ARM CPU" would actually result in.

A Sandy Bridge CPU is nothing at all like a 486 CPU, but that is what you get when you try and make a 486 CPU be "Big". A "big" arm CPU is going to be something else entirely than a mere ARM cpu.
 

Concillian

Diamond Member
May 26, 2004
3,751
8
81
What would a "big" ARM chip do?

I think it would make a pretty decent candidate for next gen consoles. They don't need a ton of CPU horsepower if they're mated to GPUs that have some compute performance.

A "big" arm, like quad, hex or octo made on HP instead of LP process and pushes voltage up a bit for some clock speeds should be adequate for consoles and leaves the option for easy co-development for mobile platforms.
 

Ben90

Platinum Member
Jun 14, 2009
2,866
3
0
Gaming on OSX/Linux is obviously better than windows. Everyone knows this.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
I think it would make a pretty decent candidate for next gen consoles. They don't need a ton of CPU horsepower if they're mated to GPUs that have some compute performance.

A "big" arm, like quad, hex or octo made on HP instead of LP process and pushes voltage up a bit for some clock speeds should be adequate for consoles and leaves the option for easy co-development for mobile platforms.

Sorry, that's what I get for asking a silly question, I had that backwards.

I did not mean to say "what would a big Arm do", what I meant to ask is "what would you do to ARM to make it a big ARM?".

If I gave you the design plan (microarchitecture and ISA) for a 486 processor and asked you to make a "big 486 CPU", what would you do to the existing 486 such that it became a "big 486"?

This is what is not clear to me. Just saying it is the objective doesn't really fill in that ~300mm^2 of die space that could soak up 3B xtors. So what are you gonna do to ARM to make it a Big ARM?
 

Ferzerp

Diamond Member
Oct 12, 1999
6,438
107
106
And the corollary to that would be how would you expect to do so in a way that would allow you to compete with the 8 million pound gorilla?

Sorry, no, magic pixie dust isn't an answer.
 

wahdangun

Golden Member
Feb 3, 2011
1,007
148
106
Because ARM CPUs are not boxing in the same category as X86 CPUs ,
not even comparable with an Intel Atom....

hmm why do you think like that ??

Sorry, that's what I get for asking a silly question, I had that backwards.

I did not mean to say "what would a big Arm do", what I meant to ask is "what would you do to ARM to make it a big ARM?".

If I gave you the design plan (microarchitecture and ISA) for a 486 processor and asked you to make a "big 486 CPU", what would you do to the existing 486 such that it became a "big 486"?

This is what is not clear to me. Just saying it is the objective doesn't really fill in that ~300mm^2 of die space that could soak up 3B xtors. So what are you gonna do to ARM to make it a Big ARM?


maybe nvdia could added some transistor to make it reach higher clock just like HD 4890 ?? or maybe added some RSIC function to speed up some process, added some instruction set like 3Dnow or SSE instruction set, and maybe adding better branch prediction.

and NVDIA could reviving their chipset business and serve boarder market than just providing chip for a mere tablet and phone witch have a very very razor thing profit.
 
Last edited:

lamedude

Golden Member
Jan 14, 2011
1,222
45
91
A Cortex A15 with a 128bit IMC, a lot of cache and maybe some more ALUs/FPUs is my idea of a big ARM. Don't there's a market for slighty faster than Atom that uses much more power ARM CPU though.
 

dma0991

Platinum Member
Mar 17, 2011
2,723
1
0
because if nvdia pushing android to be gaming platform then we will have alternative and maybe the os business will become competitive.
You must understand that adding more hardware performance isn't going to ensure the success of a platform in gaming. If you understand why PC lost to consoles in terms of number of gamers and game titles, you will understand what I'm trying to say.

The focus of handheld devices should not be solely for gaming. It can be a very versatile device and not limit it's focus on gaming performance alone. This is why I think that PSVita will not be able to outpace lesser performing handhelds despite its strong hardware end, it can't do anything more than play games.
 

Ben90

Platinum Member
Jun 14, 2009
2,866
3
0
because if nvdia pushing android to be gaming platform then we will have alternative and maybe the os business will become competitive.
I will agree with you partly here. I recommend watching this John Carmack Interview until at least about 7:40 which highlights the major problems with the IDTech5 engine on Windows.

I don't believe Android is going to bring us an optimal gaming operating system. It suffers the same problem Windows has in which it needs to support a wide array of hardware with as much program compatibility as possible. Raw performance comes second.

Linux or even *gasp* iOSX is honestly a better bet for creating us our theoretical super performance gaming system. Unlike Windows/Android, they are able to give legacy hardware a giant middle-finger. Linux is also able to cut out a lot of the user friendly bloat in addition.

However, as far as I know, Direct3D is here to stay for quite a while. The eventual shift off traditional rasteration into ray tracing is the next slim chance of a major API shift which might permit Linux to be used for a majority of games.
 

cytg111

Lifer
Mar 17, 2008
25,687
15,177
136
I will agree with you partly here. I recommend watching this John Carmack Interview until at least about 7:40 which highlights the major problems with the IDTech5 engine on Windows.

- Carmack is a friggin rockstar, the nerd I was supposed to be born into.. Love his take on many subjects, and they're often not exclusive to his line of profession! Alot to learn from this man.

edit-> i do feel sorry for the guy interviewing him though .. picture(s) speaks a 1k words, click the link.
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
maybe nvdia could added some transistor to make it reach higher clock just like HD 4890 ?? or maybe added some RSIC function to speed up some process, added some instruction set like 3Dnow or SSE instruction set, and maybe adding better branch prediction.

and NVDIA could reviving their chipset business and serve boarder market than just providing chip for a mere tablet and phone witch have a very very razor thing profit.

Ah, ok, so you are thinking that Nvidia would exand the ISA along with the microarchitecture, much in the same way that Intel and AMD did with x86 CPU's over time.

x86ISAovertime.jpg


There are a couple reasons why what you are asking to happen is impractical (and thus why it isn't happening).

First reason is that ARM owns the IP and rights to decide what "ARM" means. The ARM license holders are dependent on ARM to expand the ISA of ARM.

The idea with ARM is that the licensee builds a SOC with ARM, the IP added to the SOC is by way of the other parts of the chip that is being added.

What makes Tegra unique to all other ARM products is the stuff Nvidia added to the tegra chip above and beyond the ARM core itself. (the graphics stuff, software drivers, power management, etc)

As far as making an uber ARM core, that is up to ARM Holdings. They hold all the rights to do that, and its up to them to do it and to then license it to its customers.

I suppose a company like NVidia could jointly develop such an uber ARM core, but it took Intel 20yrs and tens of billions of dollars to get that 486 chip up to the level of today's Sandy Bridge. Nvidia and ARM aren't going to magically make that step in performance and complexity in a mere 2-4 yrs for $1B-$4B. The barrier to entry is not so easily overcome.
 

wahdangun

Golden Member
Feb 3, 2011
1,007
148
106
I will agree with you partly here. I recommend watching this John Carmack Interview until at least about 7:40 which highlights the major problems with the IDTech5 engine on Windows.

I don't believe Android is going to bring us an optimal gaming operating system. It suffers the same problem Windows has in which it needs to support a wide array of hardware with as much program compatibility as possible. Raw performance comes second.

Linux or even *gasp* iOSX is honestly a better bet for creating us our theoretical super performance gaming system. Unlike Windows/Android, they are able to give legacy hardware a giant middle-finger. Linux is also able to cut out a lot of the user friendly bloat in addition.

However, as far as I know, Direct3D is here to stay for quite a while. The eventual shift off traditional rasteration into ray tracing is the next slim chance of a major API shift which might permit Linux to be used for a majority of games.

umm excuse me but isn't android is a linux variant too ? just like other linux distro ??


Ah, ok, so you are thinking that Nvidia would exand the ISA along with the microarchitecture, much in the same way that Intel and AMD did with x86 CPU's over time.

x86ISAovertime.jpg


There are a couple reasons why what you are asking to happen is impractical (and thus why it isn't happening).

First reason is that ARM owns the IP and rights to decide what "ARM" means. The ARM license holders are dependent on ARM to expand the ISA of ARM.

The idea with ARM is that the licensee builds a SOC with ARM, the IP added to the SOC is by way of the other parts of the chip that is being added.

What makes Tegra unique to all other ARM products is the stuff Nvidia added to the tegra chip above and beyond the ARM core itself. (the graphics stuff, software drivers, power management, etc)

As far as making an uber ARM core, that is up to ARM Holdings. They hold all the rights to do that, and its up to them to do it and to then license it to its customers.

I suppose a company like NVidia could jointly develop such an uber ARM core, but it took Intel 20yrs and tens of billions of dollars to get that 486 chip up to the level of today's Sandy Bridge. Nvidia and ARM aren't going to magically make that step in performance and complexity in a mere 2-4 yrs for $1B-$4B. The barrier to entry is not so easily overcome.


i think you forget one thing AMD currently taking approach to open up their architecture for third party. so maybe (although its unlikely) they will licensing several AMD patented CPU technology to nvdia ??? like 3Dnow etc.

btw is 3Ghz ARM really difficult to build ?????maybe with just increasing tegra frequencies, and without expanding the ISA, nvdia could in theory make high performance ARM based CPU ???
 

grkM3

Golden Member
Jul 29, 2011
1,407
0
0
id like to see them make a big arm chip with 8 a15 cores on one die clocked to 4 ghz each.i can see this powering a gaming rig easily.

I think the op meant a bigger faster clocked arm chip with power draw not holding it back.

Like an arm with a 65watt tdp
 
Last edited:

cytg111

Lifer
Mar 17, 2008
25,687
15,177
136
Ah, ok, so you are thinking that Nvidia would exand the ISA along with the microarchitecture, much in the same way that Intel and AMD did with x86 CPU's over time.

x86ISAovertime.jpg


There are a couple reasons why what you are asking to happen is impractical (and thus why it isn't happening).

First reason is that ARM owns the IP and rights to decide what "ARM" means. The ARM license holders are dependent on ARM to expand the ISA of ARM.

The idea with ARM is that the licensee builds a SOC with ARM, the IP added to the SOC is by way of the other parts of the chip that is being added.

What makes Tegra unique to all other ARM products is the stuff Nvidia added to the tegra chip above and beyond the ARM core itself. (the graphics stuff, software drivers, power management, etc)

As far as making an uber ARM core, that is up to ARM Holdings. They hold all the rights to do that, and its up to them to do it and to then license it to its customers.

I suppose a company like NVidia could jointly develop such an uber ARM core, but it took Intel 20yrs and tens of billions of dollars to get that 486 chip up to the level of today's Sandy Bridge. Nvidia and ARM aren't going to magically make that step in performance and complexity in a mere 2-4 yrs for $1B-$4B. The barrier to entry is not so easily overcome.

Interresting to see tegra, being a 5 core unit, it is the first in its space.. one could theorize that while arm and tegra scales the cores on the uber low wattage level, the high level wattage industry (x86), makes the neccesary adjustments towards parralel computation and when the PC is finally able to take generic abuse of n number of cores, the x86 generation will still be sitting on 4 cores while arm and company is +? .. (or dont mind me)