Nvidia CPU

[Goku1814]

Could Nvidia make their own CPUs for computers that can rival Intel or AMD?

[Guest]

They make SoCs. If they wanted to make true CPUs, they would.

[Goku1814]

2 minutes ago, Den-Fi said:

They make SoCs. If they wanted to make true CPUs, they would.

Could they possibly be better than AMD or Intel?

[Guest]

Just now, Goku1814 said:

Could they possibly be better than AMD or Intel?

No one know since they don't exist.

I'm sure if they thought they could overtake either one, they would have done so.

Instead they opt for areas they know they'll profit.

[Goku1814]

Just now, Den-Fi said:

No one know since they don't exist.

I'm sure if they thought they could overtake either one, they would have done so.

Instead they opt for areas they know they'll profit.

are their GPUs the best tho?

[Guest]

3 minutes ago, Goku1814 said:

are their GPUs the best tho?

They currently have the best GPU in terms of performance.

 

[zassou]

they already making cpus, google tegra

 

they just not design it from the ground up, they buy blueprints and build a product out of it, and have tsmc to manufacture the chips for them.

[Rolling2405]

6 hours ago, comander said:

With x86 compatibility - it'd be very hard legally.

This is something I don't get why are AMD and Intel the only ones allowed to make x86? Shouldn;t the patent have expired already allowing any company the right to make x86 CPUs? Why is there a legal block and will it ever expire eventually like a normal patent.

[Rolling2405]

28 minutes ago, comander said:

from the early 2000s which isn't expired. 

So the patent or licence will never expire and only Intel and AMD will ever have the rights to x86???

[zassou]

AMD makes x86 cpu because back in the days intel supply cpus for ibm for their pc line, and ibm needed a second supplier, so intel contacted amd.

so companies start making x86 chips, including amd, cyrix, nec, national semiconductor etc. most of them are clones of intel chips, some later develop their chip, aka clean house design.

 

fast forward to 2000s, amd designed the highly sucessful x86-64 instruction set, which is compatible with 32 bit x86, therefore quickly adopted by inductry. while intel had their own 64 bit chip, the itanium (RIP), it's not compatible with x86. so intel did a cross patent deal with amd, that license x86-64 from them (like companies buy design from ARM), in change for extend x86 license to amd.

 

via bought centuar and cyrix (which bought from nat semi) so they have their own flavour of x86 design, and also has some sort of cross patenting thing with intel. while in the 90s, transmeta was the fourth manufacture of x86 chips, altho it's only compatible, not native, and it's limited on really low powered applications.

 

so basically, there are only 3 providers of x86 chips right now, intel, amd and via (zhaoxin). i can give some unbased but educated guesses about the chinese x86 situation if you interested.

[kokakolia]

On 5/19/2020 at 1:22 AM, Goku1814 said:

Could Nvidia make their own CPUs for computers that can rival Intel or AMD?

Maybe. But the desktop market isn't very attractive in the long, long term. Everything is slowly going mobile on the consumer-level. And I expect the desktop market to rely heavily on professionals and enthusiasts (e.g. PC gamers). 

 

It would be smart for Nvidia to branch out of the desktop market into other fields. And that's what they're quietly doing already with self-driving cars. 

[Windows9]

On 5/19/2020 at 12:56 AM, comander said:

1. With x86 compatibility - it'd be very hard legally. Emulation might be viable but that's its own can of worms. 
2. Using ARM/RISCV/POWER - maybe but WHY? It'd be a huge risk and I don't see it paying off. Many-core CPUs are becoming increasingly memory limited. Over time they'll become more and more commodity-like.

GPU cores are extremely powerful, so if you can repurposed them as CPU cores, you will have double the speed of a good CPU out there, providing you can do it efficiently

[Windows9]

2 hours ago, kokakolia said:

Maybe. But the desktop market isn't very attractive in the long, long term. Everything is slowly going mobile on the consumer-level. And I expect the desktop market to rely heavily on professionals and enthusiasts (e.g. PC gamers). 

Even PC gamers are starting to want smaller, more compact and portable devices. Look at how many Mini ITX build they are. Soon we will probably be using mobile CPU's in compact desktops

[SavageNeo]

On 5/19/2020 at 2:30 AM, Goku1814 said:

are their GPUs the best tho?

NO. They have very good performnace but they dont have much good bang for the buck gpus (meaning that in certain budget their gpu makes no sense)

good bang fro the buck gpus: rx 570 4gb/ rx 580/ 1660 super/ rx 5600XT/ rx 5700/ rx 5700XT/ rtx 2080 super

[Anden]

54 minutes ago, Windows9 said:

GPU cores are extremely powerful, so if you can repurposed them as CPU cores, you will have double the speed of a good CPU out there, providing you can do it efficiently

 CPU's and gpu's are fundamentally different... The gpu's can't do the same things as a cpu...

[Eigenvektor]

42 minutes ago, Windows9 said:

GPU cores are extremely powerful, so if you can repurposed them as CPU cores, you will have double the speed of a good CPU out there, providing you can do it efficiently

A GPU core is not the same thing as a CPU core by a long shot. A GPU's cores are tiny and only useful for a handful of instructions. A GPU has different cores optimized for different tasks and uses them as needed (think CUDA, RTX, Stream, …)

 

The reason a GPU is so fast at what it does is that it has been purpose built for the very thing it is used for: Processing large sets of data with the same limited number of instructions over and over again. The speed comes from the fact that you have thousands of cores that can process data in parallel. Individual cores aren't very fast at all.

 

A CPU on the other hand can run far more complex instructions and is well suited for branching code. Emulating a CPU with a GPU is pretty much impossible, because the cores can't do what a CPU can. A CPU on the other hand can absolutely emulate a GPU. The reason it is slower is because it's cores are far larger, more power hungry and it doesn't have thousands of them to process data in parallel.

[Windows9]

3 hours ago, comander said:

1. GPU cores are extremely weak. There's just 1000x more of them. 

2. For some tasks

If you can make an OS that utilises lots of cores, then it's good. One core for each process

[atxcyclist]

Cyrix tried.

[Unimportant]

On 5/21/2020 at 6:23 AM, atxcyclist said:

Cyrix tried.

And succeeded! Their 6x86 was competitive and cheaper. They were undeservedly killed off because they got unlucky with some of Carmack's and Abrash's magic in quake. (carefully hand-crafting code to run floating point and integer instructions in parallel by taking advantage of a quirck in the pentium).

[atxcyclist]

2 minutes ago, Unimportant said:

And succeeded! They were undeservedly killed off because they got unlucky with some of's Carmack's and Abrash's magic in quake. (carefully hand-crafting code to run floating point and integer instructions in parallel by taking advantage of a quirck in the pentium).

I had a Cyrix 233mhz and it was poop in many ways, not just Quake. 

[Unimportant]

30 minutes ago, atxcyclist said:

I had a Cyrix 233mhz and it was poop in many ways, not just Quake. 

They were highly competitive, just often not given a fair chance. A computer is much more then just a CPU, especially in those days where the choice of motherboard still made a huge difference. Many a Cyrix CPU, being the cheaper choice, found itself on a motherboard with a crappy chipset and no cache (the COAST module was the first to go when cutting corners). Given equal platforms they gave Intel a run for their money. Most reviews are gone so it's hard to find good evidence. Managed to find this for what it's worth: https://www.fermimn.edu.it/inform/materiali/evarchi/cyrix.dir/perfchrt.htm

 

Then came Quake, which exploited the Pentium quirk mentioned above to run floating point and integer in parallel, making the Pentium fly in quake and giving the Cyrix a reputation of "poor floating point performance". Which was not true at all, their floating point was fine, it just could not run in parallel with integer instructions.