Zhaoxin's ZX-F/KX-7000/KH-40000 and beyond

NostaSeronx · May 9, 2019

I think this might as well get its own discrete thread.

Refresher;
-> 7nm @ TSMC / 2020 high volume production
-> DDR5 / PCIe 4.0
-> Competitive to Zen/Dhyana (US-sources)

The new info from an interview;
Wang Weilin's 2019 goal is clear, "complete the development of the next generation of 7nm validation chips ..."
^-- Chief Engineer at Zhaoxin. // https://www.chainnews.com/articles/158019324090.htm -- https://translate.google.com/transl...ews.com/articles/158019324090.htm&prev=search

amd6502 · May 9, 2019

NostaSeronx said:
I think this might as well get its own discrete thread.

Refresher;
-> 7nm @ TSMC / 2020 high volume production
-> DDR5 / PCIe 4.0
-> Competitive to Zen/Dhyana (US-sources)

The new info from an interview;
Wang Weilin's 2019 goal is clear, "complete the development of the next generation of 7nm validation chips ..."
^-- Chief Engineer at Zhaoxin.

Why not 12FDX?

NostaSeronx · May 9, 2019

amd6502 said:
Why not 12FDX?

GlobalFoundries' is limited by the state on when it wants to start Chengdu.
Also, SOITEC just recently opened an office/channel in China. //https://www.soitec.com/en/news/press-releases/soitec-announces-direct-sales-operation-in-china

With that and VIA/Zhaoxin already announcing a partnership with TSMC for ZX-E(16FF)/ZX-F(7FF). Zhaoxin using FDSOI is outside the norm, since 2017. VIA is unlikely to use because Taiwan. They have already paid the fee to use the same chips rebranded to Nano for worldwide as well.

thigobr · May 9, 2019

Competitive with Zen is a big thing

NTMBK · May 10, 2019

thigobr said:
Competitive with Zen is a big thing

It's also not backed up by the article- that is just Seronx.

NostaSeronx · May 10, 2019

NTMBK said:
It's also not backed up by the article- that is just Seronx.

https://fuse.wikichip.org/news/733/zhaoxin-launches-their-highest-performance-chinese-x86-chips/
"Zhaoxin has stated they intend on reaching AMD level of performance with KX-6000’s successor, KX-7000. That is, they want the KX-7000 to match the performance of Zen 2."

There is also an echo chamber for it as well, in other languages as well!
-> Zhaoxin hoopt met deze generatie in de buurt van de performance van Zen 2 te komen. // Zhaoxin hopes with this generation to come close to the performance of Zen 2.

Most recent translation; The large semiconductor industry network quoted Wang Weilin, the chief engineer of Zhaoxin, as saying that the performance of the KX-7000 is aimed at the extent of AMD in the same period, so is Zen 2 or Zen 3?
^-- Opinion piece, etc --^

dark zero · May 10, 2019

But where is KX6000? There are any performance scores?

Also they expected KX6000 by 2018. So KX7000 might come by 2021

coffeemonster · May 10, 2019

funny that they would publicly target Zen2 as a performance goal when the public doesn't even know what zen2's performance will be yet. Perhaps they meant Zen+ (Ryzen 2000)?

NostaSeronx · May 10, 2019

dark zero said:
But where is KX6000? There are any performance scores?

Also they expected KX6000 by 2018. So KX7000 might come by 2021

https://browser.geekbench.com/v4/cpu/compare/13037999?baseline=12754288
If I remember correctly, the U before the numbers means laptop-orientated. Where Intel/AMD place the U after the numbers.
AMD Ryzen 3500U median/mean => 3682 MHz/3680 MHz
Zhaoxin Kaixian U6780A median/mean => 2669 MHz/2669 MHz

The KX-6000/KH-30000 have recently launched in 2019, but were validated back in 2018. So, if based on the KX-6000 and statements by the chief, then we should see benchmarks by the fourth quarter of 2019.

amd6502 · May 10, 2019

NostaSeronx said:
https://browser.geekbench.com/v4/cpu/compare/13037999?baseline=12754288
If I remember correctly, the U before the numbers means laptop-orientated. Where Intel/AMD place the U after the numbers.
AMD Ryzen 3500U median/mean => 3682 MHz/3680 MHz
Zhaoxin Kaixian U6780A median/mean => 2669 MHz/2669 MHz

The KX-6000/KH-30000 have recently launched in 2019, but were validated back in 2018. So, if based on the KX-6000 and statements by the chief, then we should see benchmarks by the fourth quarter of 2019.

I would think competitive with excavator, or zen 1 at best.

Also, competitive could mean various things. piledriver 8350 was competitive with sandy i7 in various loads. It might win a few benchmarks in MT int and MT mixed, but generally would lag overall. Still, you could consider it "competive" if it would win or tie a few specific races.

fkoehler · May 11, 2019

8 cores vs 4, and it still is only 66%.
4 core vs 4, or 1 to 1 should scale to 33%.

I fail to see what the interest is in this, as its primarily directed at in-China use to get national and internal systems off of Western equipment.
If they try to export it, they're going to be hit with trade suits for an IP they've 'borrowed' from elsewhere since I assum VIA has limited access to x86 IP.
Unless they somehow are able to duplicate/exceed both Intel and AMD's R&D, to get to and remain competitive means copying IP.
Do that, and they aren't going to be able to export.

amd6502 · May 11, 2019

fkoehler said:
I assum VIA has limited access to x86 IP.
Unless they somehow are able to duplicate/exceed both Intel and AMD's R&D, to get to and remain competitive means copying IP.
Do that, and they aren't going to be able to export.

Maybe I'm very wrong about this but I thought Via had full use of x86 after they acquired Cyrix. That x86 designer was the first to bring out of order superscalar to x86, with tricks like register renaming. However, thinking about it, the amd64 extensions, they may not have the rights to use those.

fkoehler · May 11, 2019

IIRC, Cyrix/Via has x86 IP, however that is the old stuff. I don't believe they have any right to use anything new Intel/AMD has developed after that. Thats all Intel/AMD's own IP.
If that were not true, then they would be snapping at Intel's heels using that newer IP, and since they aren't, the assumption that they aren't allowed too seems reasonable.
They are going to be playing catch-up forever and a day, unless they dump ton's of R&D dollars into copying Intel/AMD from public source data, or simply copying....

jpiniero · May 11, 2019

fkoehler said:
They are going to be playing catch-up forever and a day, unless they dump ton's of R&D dollars into copying Intel/AMD from public source data, or simply copying....

China did acquire the Zen 1 IP...

NostaSeronx · May 11, 2019

The exact license agreement;
Section III.C.1 requires that Intel offer to extend the capture period in VIA’s x86 license until April 7, 2018. Section III.C.4 prohibits Intel from conditioning its offer on any other change to VIA’s x86 license. And Section III.E makes it a violation of the Decision and Order for Intel to breach its license with VIA, AMD, or NVIDIA. This extension means that Intel lacks a good-faith basis to bring patent litigation against VIA until at least April 2018, as VIA has a license to all of Intel’s intellectual property for x86-instruction-set compatible CPUs.
=> Under the license between VIA and Intel, each company gains a perpetual license to any patent reading on x86-instruction-set-compatible CPUs obtained by the other company during the capture period.

Zhaoxin has an inherited license through VIA.

///
ZX-A through ZX-C were mostly the work of Centaur, with ZX-D through ZX-F being mostly the work of Zhaoxin's research and development. Somewhere in 2017, VIA gave credit to Zhaoxin, for which I assume is for ZX-D through ZX-F cross-licensing. So far only the ZX-C/KX-5000(ZX-D) is outside of China, being displayed for Europe buyers in Ukraine in 2017.

dark zero · May 12, 2019

At last they are on Intel Core tech levels but needs a dedicated GPU in order to compete

Arkaign · May 13, 2019

This stuff is always fascinating. The idea of a relevant 3rd party CPU maker with x86 makes me a bit misty eyed thinking of the days when one board could run Intel, AMD, Cyrix, and maybe even VIA lol. Super 7 era was probably the most diverse mobo capability ever.

Then Intel went slot 1, AMD slot A, and we've been on a less flexible path ever since.

SPBHM · May 13, 2019

Via had some presence with the C3 and C7 in the 2000s, but after ARM became so popular and Atom/AMD Bobcat and so on it was pretty much gone.

Nano looked good against the first Atom but it never had the availability/pricing to compete.

fkoehler · May 13, 2019

jpiniero said:
China did acquire the Zen 1 IP...

Ah wait, is that from AMD's joint partnership venture talked about last year, with THATIC.
What is their relationship with Zhaoxin that allows Zhaoxin to use any of that?

And, VIA has IP capture through 2018 only, correct?

OK, I stand corrected. VIA/Zhaoxin have closed the gap with the partnership with VIA.
And, the Chinese have gotten their hands on AMD technology through THATIC.

I think Siu made a real mistake there, penny-wise and pound foolish.

Kosusko · Nov 10, 2019

KX-7000 ZX-F OctaCore 2000MHz @ Geekbench 5.0.3 Tryout for Windows x86 (64-bit)

Single-Core Score: 469
Multi-Core Score: 3 264

Shanghai Zhaoxin Semiconductor Co., Ltd. CHA001 MB - Geekbench

Benchmark results for a Shanghai Zhaoxin Semiconductor Co., Ltd. CHA001 MB with a ZX-F OctaCore 2000MHz processor.

browser.geekbench.com

VirtualLarry · Nov 10, 2019

fkoehler said:
I think Siu made a real mistake there, penny-wise and pound foolish.

I think so too. Like handing our ICBM technology over to the Russians (hypothetical analogy).

Gideon · Nov 10, 2019

Kosusko said:
KX-7000 ZX-F OctaCore 2000MHz @ Geekbench 5.0.3 Tryout for Windows x86 (64-bit)

Single-Core Score: 469
Multi-Core Score: 3 264

Shanghai Zhaoxin Semiconductor Co., Ltd. CHA001 MB - Geekbench

Benchmark results for a Shanghai Zhaoxin Semiconductor Co., Ltd. CHA001 MB with a ZX-F OctaCore 2000MHz processor.

browser.geekbench.com

While it's obviously far from perfect for 7nm and its single threaded score leaves a lot to be desired, It's actually not that shabby all things considered. It has no turbo and it ran at 2Ghz the entire test, therefore if we normalize the MT score for clock speed and cores(which obviously isn't quite correct since it dosen't scale linearly) we get:

R5 2500U = 2670 MT score / 3.5 / 4 = 190.714 MT points per core per Ghz
KX-7000 = 3264 MT score / 2.0 / 8 = 204 MT points per core per Ghz
R7 3700x = 8752 MT score / 4.35 / 8 = 251.4 MT points per core per Ghz

So obviosly not competitive with zen 2 (even per-clock with the big assumption that the benchmark scales linearly), but quite similar to zen 1.
That of course is with the big assumption that it gets a turbo and that the final achievable clocks are at least in the 4GHz ballpark
https://browser.geekbench.com/v5/cpu/compare/526995

Kosusko · Nov 10, 2019

to Gideon:

Thank you, but please provide exact results. Not scaling.

IPC @ 2.0GHz as this:

Kosusko said:
IPC @ 2.0GHz:

Intel Atom C3958 @ 2.00 GHz (68,66%)
1 processor, 16 cores
GenuineIntel Family 6 Model 95 Stepping 1

vs

KX-7000 ZX-F OctaCore 2000MHz @ 2.00 GHz (+45,66%)
1 processor, 8 cores
CentaurHauls Family 6 Model 71 Stepping 1

link: https://browser.geekbench.com/v5/cpu/compare/526995?baseline=458824

or this

IPC @ 2.0GHz:

AMD FX-8300 3.3GHz @ 2.00 GHz (65,25% )

vs

KX-7000 ZX-F OctaCore 2000MHz @ 2.00 GHz (+53,27% )

link: https://browser.geekbench.com/v5/cpu/compare/526995?baseline=562975

NostaSeronx · Nov 10, 2019

Got to watch if there is a KX-S7000 model... like the server side: https://browser.geekbench.com/v5/cpu/341928 (2x KX-6000 dies)

I heard that it was a possible that KH-S40000 models will do multiple BGA w/ multiple dual-dies. Up to 8P being 8 BGA packages with 2x 8-cores => 128 cores.

ehume · Nov 10, 2019

VirtualLarry said:
I think so too. Like handing our ICBM technology over to the Russians (hypothetical analogy).

What's not so hypothetical is that the US gave the Russians info on safeguarding nuke release. Somebody in our intelligence bureaucracy noted that the Soviets never went on alert. One thing led to another and somehow Soviet spies were allowed to find out how to safeguard nuke launches. Apparently our guys breathed a sigh of relief when the Sovs were able to exercise their missiles.

Zhaoxin's ZX-F/KX-7000/KH-40000 and beyond

Diamond Member

Senior member

Diamond Member

Senior member

Lifer

Diamond Member

Platinum Member

Senior member

Diamond Member

Senior member

Member

Senior member

Member

Lifer

Diamond Member

Platinum Member

Lifer

Diamond Member

Member

Member

No Lifer

Golden Member

Member

Diamond Member

Golden Member