News The only Chinese chipmaker with an x86 license releases surprisingly modern new chips made with a mystery process node —Xhaoxin KX-7000 CPU launche...

Status
Not open for further replies.
"Whether KX-7000 beats or even levels the playing field isn't the whole point for China. The ultimate goal is achieving technological autonomy"
exactly, and in the meantime US companies will lose a big market.
this is why Nvidia is selling them as much as they can, CPU is bound to be replaced, GPGPU+IA accelerators will take a longer time.
edit:typo
 
Last edited:
Now I want one for testing, im sure its going to be ridiculously expensive for what it is when it becomes available to buy from Aliexpress, but it would be fun to benchmark and compare. It would be nice if they built a whole socketed platform around it, but that may be putting the cart before the horse for now.
 
> Speaking of which, the US has restricted China's access to leading process node technologies, making it hard for the country to create chips.

No, it makes it hard to compete with the latest chips from the west... it does not make it hard to make computer chips. All it means is their software devs will have to git gud at optimizing.
 
Now I want one for testing, im sure its going to be ridiculously expensive for what it is when it becomes available to buy from Aliexpress, but it would be fun to benchmark and compare. It would be nice if they built a whole socketed platform around it, but that may be putting the cart before the horse for now.
probably can run on existing MoBo (be Intel or AMD), otherwise they would throw away whole ecosystems and make them harder and more expensive to replace in the short term.
 
  • Like
Reactions: artk2219
Fellow Chinese CPU maker Loongson achieved Zen 3-like IPC with its 3A6000, so there's no reason Zhaoxin couldn't do the same.
Well... Loongson did kinda make its own ISA (debatable ties to MIPS - let's not get sidetracked on that, please). Making a fast x86 CPU is going to be more difficult, due to the complexity of the ISA.

Although VIA hadn't been particularly relevant in the CPU space for several years, it crucially is one of the three companies with a license to the x86 architecture, alongside Intel and AMD.
I haven't seen anyone debate whether Via can legitimately make x86, but didn't they get that by buying Centaur? And didn't they recently sell off the remnants of Centaur to Intel?

This gives its CPUs a big leg up since they can natively access the massive library of PC-related software made for x86 processors.
Even so, it's unlikely to run as fast as Apple's processors can execute x86 code under emulation.

Boost Frequency 3.7GHz
You know the first x86 CPU to reach 3.7 GHz was a Pentium 4, right? So, clock speed definitely doesn't tell the whole story!

So, referring back to the previous generation this is supposedly 2x as fast as...

3aKTCkoNWK6nWdJQXfDb6g.png


AJ7wSwYDZE2WvAmAN3XBC8.png


Source: https://www.tomshardware.com/features/zhaoxin-kx-u6780a-x86-cpu-tested/4

Uh... 2x of that would be usable? I think that's the best I can say.
 
I feel like you should have added more information about VIA/Centaur's CNS project. It's almost certain these are evolved versions of the Centaur CNS CPUs which were never publicly released.
Eh, that's not what they say.

Here are some details about a prior, 28 nm SoC they developed 6 years ago.

"WuDaoKou is largely a brand new architecture designed by Zhaoxin. This is a departure from earlier microarchitectures such as ZhangJiang which were a lightly modified version of VIA Technologies (Centaur) architecture. WuDaoKou is a new and complete SoC design."


That's the claim, anyway. Yes, they had access to Centaur's IP. ...so, legitimate questions remain.
 
Last edited:
probably can run on existing MoBo (be Intel or AMD), otherwise they would throw away whole ecosystems and make them harder and more expensive to replace in the short term.
The previous generation that Toms actually reviewed was BGA (soldered):


You do know the Chinese are pretty good at making motherboards, right? That's not the hard part!

Yes, running Windows on the thing would require new drivers, if you don't reuse an existing chipset, but I somehow doubt either AMD or Intel is going to give them support for reusing an existing chipset. Plus, Via was once famous for making its own x86 motherboard chipsets!
 
While I have been surprised at the level of advancement China has made for CPUs+ in recent history, I expect we are getting cherry picked data here and the real performance will be far less than what is advertised. Regardless their advancements are nothing to be sneezed at and I hope it will push other companies to do better. Certainly the US based CPU/GPU companies need to step things up or risk losing the edge they currently have. These embargos have really lit a fire under China's back side to become independent of US and other 'foreign' powers chip making technologies.

Ultimately this is disheartening yet exciting news (from a tech stand point) all at the same time. I'll be curious to follow their progress.
 
Uhhh the 7700x supports 192Gb of ram as well via 48gb ram chips why make the Intel chips look better than they really are there has been no real arch change since 12th gen Intel. Plenty of reviews showing a 2 to 6% performance increases from 12th to 14th gen Intel sad days can't wait for AMD 8***/9*** series in Feb March April to really put a lot of pressure on Intel
 
Well... Loongson did kinda make its own ISA (debatable ties to MIPS - let's not get sidetracked on that, please). Making a fast x86 CPU is going to be more difficult, due to the complexity of the ISA.


I haven't seen anyone debate whether Via can legitimately make x86, but didn't they get that by buying Centaur? And didn't they recently sell off the remnants of Centaur to Intel?


Even so, it's unlikely to run as fast as Apple's processors can execute x86 code under emulation.


You know the first x86 CPU to reach 3.7 GHz was a Pentium 4, right? So, clock speed definitely doesn't tell the whole story!

So, referring back to the previous generation this is supposedly 2x as fast as...
3aKTCkoNWK6nWdJQXfDb6g.png
AJ7wSwYDZE2WvAmAN3XBC8.png

Uh... 2x of that would be usable? I think that's the best I can say.
Quite usable, better than a lot of low power chips that are being used these days. It won't be a power house, but it should be a decent everyday web machine, basic media, light gaming. If it's within 30% or so of Intels efficiency cores, that's super usable.

 
I haven't seen anyone debate whether Via can legitimately make x86, but didn't they get that by buying Centaur? And didn't they recently sell off the remnants of Centaur to Intel?
They did, and they did, but word regarding x86 was nil. It sounded mostly like Intel got whatever they wanted out of Centaur and left IP with Via. I imagine we'll only find out if they start trying to sell CPUs in the west as fighting over it in China wouldn't be worth the effort.
 
Now that I think about it this should put it at better than piledriver IPC. It would actually be neat to put it head to head against like an FX 8320, same number of threads, similar clock speeds, maybe throw in sandy bridge as well, that would be fun. Man I wish they made an 8 core bristol ridge apu, but we got zen instead so I can't really complain hah. I also don't think I've ever seen a detailed ipc comparison between the different bulldozer generations, Bulldozer, piledriver, steamroller, and excavator.
 
  • Like
Reactions: ivan_vy
Unless the rendered image of the kx-7000 is just an artists interpretation then it looks very very very much like they got their hands on some intel CPU....
We had a story like that some months ago.

KX-6000
kvPELvKTBLnSeoALEA5myA-970-80.jpg.webp

KX-7000
9mFVfuFQQXuiAzuNttqth6-970-80.jpg.webp


29SYgsA2NeeQXLSXXP5PJV-1200-80.jpg
Or they decided to use an identical chip packaging because that would allow reusing machines that were used to solder/assemble Intel chips on motherboards, and reuse existing cooler mounts.
 
Certainly the US based CPU/GPU companies need to step things up or risk losing the edge they currently have.
From what I've seen, nobody has been resting on their laurels since Intel got served up a rude awakening by Zen. Intel had stalled out on its new process nodes, but that pipeline finally seems to have gotten moving, at long last.

In terms of microarchitecture, I think Golden Cove was a "swing for the fences" effort, with Raptor Cove representing a worthy refinement. I do slightly worry about middling IPC improvements in Meteor Lake & Arrow Lake. We'll see what comes after, but iniatives like APX tell me that Intel is digging deep.

I think the real battleground will be RISC-V. x86 CPUs, like these, will probably serve mainly as stopgap solutions.
 
Quite usable, better than a lot of low power chips that are being used these days. It won't be a power house, but it should be a decent everyday web machine, basic media, light gaming. If it's within 30% or so of Intels efficiency cores, that's super usable.
Going by the graphs I cited, 2x performance would give it Cinebench 2020 performance on par with AMD A10-9700 (Bristol Ridge) - still far below even Zen1!

In terms of web performance, it'd draw equal with Athlon 220GE - a Zen1 APU with max clock speed at 3.4 GHz. Since its clockspeed is a little higher, we can say that it doesn't quite have the IPC of Zen1.

Yes, I consider that usable. That's what I said: usable. Not good, though. Especially when you conisder Zen1 was made on 14nm and this is supposedly being made on TSMC N7, yet it doesn't even match Zen1's IPC!
 
Unless the rendered image of the kx-7000 is just an artists interpretation then it looks very very very much like they got their hands on some intel CPU....
We had a story like that some months ago.
That first photo was from a Toms review, which I linked. After seeing those benchmarks, if you still want to claim it's an Intel processor... I guess you can have at it? I doubt Intel would appreciate the suggestion of having anything to do with it, however!
: D
 
Or they decided to use an identical chip packaging because that would allow reusing machines that were used to solder/assemble Intel chips on motherboards, and reuse existing cooler mounts.
No, the socketed CPUs he pictured were indeed rebadged Intel processors. It was likely done by somebody else, though - not Zhaoxin, themselves.
 
Last edited:
Going by the graphs I cited, 2x performance would give it Cinebench 2020 performance on par with AMD A10-9700 (Bristol Ridge) - still far below even Zen1!

In terms of web performance, it'd draw equal with Athlon 220GE - a Zen1 APU with max clock speed at 3.4 GHz. Since its clockspeed is a little higher, we can say that it doesn't quite have the IPC of Zen1.

Yes, I consider that usable. That's what I said: usable. Not good, though. Especially when you conisder Zen1 was made on 14nm and this is supposedly being made on TSMC N7, yet it doesn't even match Zen1's IPC!
Correct, it actually looks like it should be right around Sandy Bridge in terms of IPC, and I still know tons of people that are still using one of those CPU's at least somewhere in their lives, heck I still know people that use one for gaming. Don't get me wrong, thats definitely still very old and much slower than a modern mid range CPU, but you could do worse. I'm sure they will be thrilled when they hit Zen 1 levels of performance, but we cant expect them to just jump there. They do have the benefit of working with Via which has been making x86 CPU's for decades, but not HIGH POWER x86 cpu's. This is the first time that team has actually been getting the infusion of money and resources thats been allowing them to design chips that could eventually be "pretty good" instead of just "good enough". The architecture and platform are definitely still works in progress, but im very curious to see where they will be within another 5 years. I doubt they will be fully caught up, but they should have some novel and interesting solutions none the less.
 
The architecture and platform are definitely still works in progress, but im very curious to see where they will be within another 5 years. I doubt they will be fully caught up, but they should have some novel and interesting solutions none the less.
I wonder how many side-channel attacks it's vulnerable to. That could swing my assessment, somewhat. If fewer than Zen 1, I could forgive them for not hitting Zen 1 levels of IPC.
 
Status
Not open for further replies.