Will AMD Carrizo offer a considerable performance increase over Kaveri?

[Sonicadvance1]

(09-19-2014, 07:50 AM)NaturalViolence Wrote: Isn't the definition of an x86 cpu a cpu that can execute x86 code natively? I don't know of any cpus that can run multiple ISAs at the physical level.

You're describing a Transmeta CPU which weren't very popular, but they were fairly good.
Nvidia owns a license for a the Transmeta design which they are using for their ARMv8 compatible CPU.
Transmeta got in trouble and was barred from releasing CPUs that run x86 code after a while.
With the company no longer existing, I don't expect Nvidia to be attempting to fix the licensing nonsense to try and make their CPU work with x86 instead of ARMv8

[NaturalViolence]

Transmeta used software emulation for x86 though. They used a JIT dynarec for it if I recall. What I believe DatKid was asking for is a cpu that can run both x86 and ARM code without any software layer abstraction.

And their performance was very lackluster which is why they never caught on even before all that nonsense happened.

[Sonicadvance1]

They translated x86 in to their own microcode. Basically a JIT recompiler built in to the CPU, not a userspace recompiler. So for all intents and purposes, the host OS thought it was running on a x86 CPU.

The Loongson MIPS CPU can execute both x86 and MIPS64, which is at the hardware level.

[NaturalViolence]

Sonicadvance1 Wrote:They translated x86 in to their own microcode. Basically a JIT recompiler built in to the CPU, not a userspace recompiler. So for all intents and purposes, the host OS thought it was running on a x86 CPU.

But how can it be considered "built into the cpu" if it's software that runs in main memory and is loaded off a ROM on the motherboard: http://www.realworldtech.com/crusoe-exposed/2/

It may not run in user space and it may translate it into its own microcode but it definitely isn't hardware emulation according to all of the documentation that I can dig up.

As for loongson I'm not really sure since it's not a well known series of cpus. What little information I can find on it seems to imply that it's a mixture of software and hardware level emulation. With a VM being needed but specific instructions existing to assist in x86 emulation. Then again searching for "x86 loongson" on google doesn't really turn up anything that I would consider a scholarly source.

[omega_rugal]

Quote:"built into the cpu"

is that a requisite?

so what if it used main RAM to work? it still doesn`t run x86 without being translated first, also since the translation code was in RAM, the processor had the ability to optimize the execution as time passed, making small fixes to the translated code with each execution, something you didn`t mention btw...

Quote:definitely isn't hardware emulation

whatever man, whatever.

[NaturalViolence]

omega_rugal Wrote:is that a requisite?

so what if it used main RAM to work? it still doesn`t run x86 without being translated first, also since the translation code was in RAM, the processor had the ability to optimize the execution as time passed, making small fixes to the translated code with each execution, something you didn`t mention btw...

All of that is true but irrelevant. Datkid specifically asked for native hardware level emulation. As in built into the cpu. The example of the transmeta crusoe provided by sonicadvance does not meet that requirement which I pointed out.

[AnyOldName3]

NaturalViolence's idea about just buying a separate x86 CPU is the sensible one. If my motherboard can have an x86 and an ARM (which isn't user-programmable, but instead uses a mixture of ROM and network driver to get stuff to do, and then all it does is optimise network traffic in some placebo inducing way) so can other things.

[NaturalViolence]

I never said have two seperate cpus. DatKid was talking about nvidia breaking into the x86 market with their fast IGPs and saying they should make a cpu that runs both x86 and ARM code without emulation. I said that a much better idea would be for nvidia to buy an existing x86 cpu in bulk to package with the IGP. This would be much cheaper, much less risky, and they would end up with a much faster product. They used to make Intel/AMD montherboards with nvidia IGPs but competition from AMDs own IGPs killed things on the AMD side and Intel repeatedly trying to sue them over it killed things on the Intel side.

Or they could work with Intel to integrate an nvidia gpu into an Intel cpu. But that will never happen. The CEOs of both companies are way too power hungry to allow such a joint project and insist on doing everything completely by themselves. Actually with AMDs APUs on the horizon and Intel in need of good IGPs to compete Intel actually did offer to merge with nvidia a few years back to do just that. But the deal never went through because Jen Huang (CEO of nvidia) insisted on being CEO of the new company. But Paul Otellini (Intels CEO at the time) wasn't having any of that. Even though this would have objectively been good for both companies since huang has been an amazing manager everywhere he's gone. Fun fact, he used to work for AMD as a microarchitecture designer back in the 80s before he founded Nvidia. He must have had no idea that his former employer, an x86 cpu company, would one day be his primary competition in the GPU market.

[DatKid20]

(09-21-2014, 09:53 AM)NaturalViolence Wrote: I never said have two seperate cpus. DatKid was talking about nvidia breaking into the x86 market with their fast IGPs and saying they should make a cpu that runs both x86 and ARM code without emulation. I said that a much better idea would be for nvidia to buy an existing x86 cpu in bulk to package with the IGP. This would be much cheaper, much less risky, and they would end up with a much faster product. They used to make Intel/AMD montherboards with nvidia IGPs but competition from AMDs own IGPs killed things on the AMD side and Intel repeatedly trying to sue them over it killed things on the Intel side.

Or they could work with Intel to integrate an nvidia gpu into an Intel cpu. But that will never happen. The CEOs of both companies are way too power hungry to allow such a joint project and insist on doing everything completely by themselves. Actually with AMDs APUs on the horizon and Intel in need of good IGPs to compete Intel actually did offer to merge with nvidia a few years back to do just that. But the deal never went through because Jen Huang (CEO of nvidia) insisted on being CEO of the new company. But Paul Otellini (Intels CEO at the time) wasn't having any of that. Even though this would have objectively been good for both companies since huang has been an amazing manager everywhere he's gone. Fun fact, he used to work for AMD as a microarchitecture designer back in the 80s before he founded Nvidia. He must have had no idea that his former employer, an x86 cpu company, would one day be his primary competition in the GPU market.

I actually wanted a x86/ARM CPU for mobile since Intel is MIA when it comes to Smartphones. If it were able to run both ARM and x86 code with ARM's efficiency I would be even more happy. I know it'll never happen but I just wanted to mention it.

[MayImilae]

datkid20 Wrote:If it were able to run both ARM and x86 code with ARM's efficiency

That's not really how it works. Efficiency improvement is doing more with the same amount of power, or doing the same amount of work with less power. That all boils down to IPC. Intel's amazing IPC is superior to any ARM fab thanks to a their insane R&D and a serious head start. The difference is of course that x86 was too big and bloated for smartphones when they first appeared, and ARM was stripped down. Basically, doing less means using less power, even with inferior IPC, so ARM gave smartphones something x86 couldn't. But as smartphones become more powerful and do more things, ARM is getting fatter to support more functionality, while Intel is working on x86 to make it more flexible and and leaner. With ARM getting bigger, and x86 getting leaner, Eventually they are going to cross. Intel's raw IPC advantage means they'll have a technically superior CPU for smartphones, with better battery life and performance. Of course ARM has a massive market cornered all to themselves right now, so even with a superior CPU it's going to be tough as nails for Intel to penetrate into smartphones.

It's going to be very interesting to watch.

Spoiler: (Show Spoiler)

I'm sure NV is going to say this is all wrong, but hey, life has to have some spice.