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

[Kosusko]

It's just the right choice of multi-theared tests.
Then it may be true.

 

[DrMrLordX]

We'll see if Dr. Cutress' tests fare any better on it. Tom's ran a fairly standard suite of tests, and it frankly looked awful. At least it's marginally-faster than Bristol Ridge.

 

[Kosusko]

This was 3.0GHz KX-U6880A vs Core i5-7400

https://imgsa.baidu.com/forum/w%3D580/sign=c61b184d9f45d688a302b2ac94c37dab/c3023ea95edf8db1cef5ad1a0423dd54574e745c.jpg

source: https://tieba.baidu.com/p/5917211233

Yes, we will see. But you can tested single core performence so result small core at only 2.7GHz is clear...

 

[DrMrLordX]

The 7-zip numbers square, but honestly who uses CBR11.5 to bench anything these days?

 

[Kosusko]

It was tested when was only R15 and older R11.5. R20 was later.
And with R15 had VIA / Centaur problems with compatibility.

 

[lightmanek]

I think that it is an interesting starting point to move performance upward. No one is expecting Zhaoxin's CPU's to win any battles with big AMD or Intel cores, but from my hobbyist point of view this is really interesting family of processors.
I wish people testing these units spent more time and effort on properly investigating all aspects of these cores and integrated graphics, but I understand that audience for this type of in-depth articles is very small.

Still, I would like to see at least clock for clock comparisons across more than just a bunch of tests. Go back to old applications, test under Windows 7, run old games, from times where this core architecture was conceived. Add to your test rigs some ancient CPU's like Phenom or Core 2 Duo to really illustrate how far or close behind these CPU's are.

If I had more time on my hands I would love to dive into quirky hardware like this and post my findings on forums

Anyway, I'm glad we at least have something from Tom's and more to expect from Anand!

 

[Hitman928]

I think that it is an interesting starting point to move performance upward. No one is expecting Zhaoxin's CPU's to win any battles with big AMD or Intel cores, but from my hobbyist point of view this is really interesting family of processors.
I wish people testing these units spent more time and effort on properly investigating all aspects of these cores and integrated graphics, but I understand that audience for this type of in-depth articles is very small.

Still, I would like to see at least clock for clock comparisons across more than just a bunch of tests. Go back to old applications, test under Windows 7, run old games, from times where this core architecture was conceived. Add to your test rigs some ancient CPU's like Phenom or Core 2 Duo to really illustrate how far or close behind these CPU's are.

If I had more time on my hands I would love to dive into quirky hardware like this and post my findings on forums

Anyway, I'm glad we at least have something from Tom's and more to expect from Anand!

You do realize these CPUs were just released this year on 16 nm, right? Even so, the 8 core CPU is more a competitor to AMD's current 2C/4T CPU in many multi-threaded applications and has performance per clock significantly behind AMD's Excavator which released almost 5 years ago. If you look at perf/w it gets even worse. They fell far, far short of where they claimed their performance to be. There's not much interesting here to see except that they're trying I guess.

 

[lightmanek]

You do realize these CPUs were just released this year on 16 nm, right? Even so, the 8 core CPU is more a competitor to AMD's current 2C/4T CPU in many multi-threaded applications and has performance per clock significantly behind AMD's Excavator which released almost 5 years ago. If you look at perf/w it gets even worse. They fell far, far short of where they claimed their performance to be. There's not much interesting here to see except that they're trying I guess.

Of course I know that, but you do realize these cores are based on old VIA Nano cores ...
You probably are missing my point of treating this CPU as a fun thing to play with and learn more about history of various x86 cores, not as a product fit to compete with modern processors.

 

[Kosusko]

You do realize these CPUs were just released this year on 16 nm, right? Even so, the 8 core CPU is more a competitor to AMD's current 2C/4T CPU in many multi-threaded applications and has performance per clock significantly behind AMD's Excavator which released almost 5 years ago. If you look at perf/w it gets even worse. They fell far, far short of where they claimed their performance to be. There's not much interesting here to see except that they're trying I guess.

Do you realize the difference between the small and BIG core microarchitecture? This is eight small cores.

ZX-C, ZX-D, ZX-E = small core
Jaguar = small core
Excavator = BIG core
Zen = BIG core
Centaur CNS (ZX-F) = BIG core

Here is performance diferend betwean eight small cores KX-6000 at 2.7GHz (ZX-E) vs. eight BIG cores KH-40000 (ZX-F) at 2.0GHz
Single-Core Score : 366 vs 469
Multi-Core Score: 2 167 vs 3 264

source: https://browser.geekbench.com/v5/cpu/compare/526995?baseline=1635708

 

[Hitman928]

Do you realize the difference between the small and BIG core microarchitecture? This is eight small cores.

ZX-C, ZX-D, ZX-E = small core
Jaguar = small core
Excavator = BIG core
Zen = BIG core
Centaur CNS (ZX-F) = BIG core

Here is performance diferend betwean eight small cores KX-6000 at 2.7GHz (ZX-E) vs. eight BIG cores KH-40000 (ZX-F) at 2.0GHz
Single-Core Score : 366 vs 469
Multi-Core Score: 2 167 vs 3 264

source: https://browser.geekbench.com/v5/cpu/compare/526995?baseline=1635708

Doesn't matter much when your small core uses as much power as your competitor's big core at a small fraction of the performance. The KH-4000 is also the unreleased, next-gen architecture made on 7 nm, so a bit disingenuous to compare the two as if it's just a difference between big and small cores.

 

[DrMrLordX]

Of course I know that, but you do realize these cores are based on old VIA Nano cores ...
You probably are missing my point of treating this CPU as a fun thing to play with and learn more about history of various x86 cores, not as a product fit to compete with modern processors.

In the end, someone's got to want to buy these things, though. I would almost rather have Goldmont+ than one of these.

Doesn't matter much when your small core uses as much power as your competitor's big core at a small fraction of the performance.

Exactly. I'm surprised that they managed to produce something that chews up that much power.

 

[Hitman928]

Of course I know that, but you do realize these cores are based on old VIA Nano cores ...
You probably are missing my point of treating this CPU as a fun thing to play with and learn more about history of various x86 cores, not as a product fit to compete with modern processors.

I guess I just don't see what is interesting about that? The 6000 series CPUs went through (reportedly) some major architectural changes as well as being designed on a not too old FinFET process. I mean, you could also say that Sandy Bridge was based off of the old Pentium 3 cores but if it had less than half the performance of its competitors, would anyone really care?

 

[Kosusko]

I guess I just don't see what is interesting about that? The 6000 series CPUs went through (reportedly) some major architectural changes as well as being designed on a not too old FinFET process. I mean, you could also say that Sandy Bridge was based off of the old Pentium 3 cores but if it had less than half the performance of its competitors, would anyone really care?

This is not the same.

But, yes I agree power consumption in review is higher but this is full form facthor Zhaoxin developement board also with 40nm I/O chip ZX-200 @ 6W TDP etc. who is not suitable.

- 16-lane PCIe slot (lane width is x8, though).
- one 4-lane
- three single-lane PCIe slots,
- one old-school PCI slot.
- VGA,
- HDMI
- DisplayPort outs
- Four SATA 3.0 connectors
- One PCIe M.2
- One USB 3.1 Gen 2 port on one Type C connector
- One USB 3.1 Gen 2 port on one Type C pin header
- Two USB 3.1 Gen 1 ports on one Type A connector
- Two USB 3.1 Gen 1 ports on one pin header
- Two USB2.0 ports on one Type A connector
- Eight USB2.0 on x4 pin header
- Two UART ports
- One Audio Codec ALC662

70W TDP for EightCore KX-U6780A 2.7GHz is too much by 16nm FinFet CMOS technology.

 

[lightmanek]

I guess I just don't see what is interesting about that? The 6000 series CPUs went through (reportedly) some major architectural changes as well as being designed on a not too old FinFET process. I mean, you could also say that Sandy Bridge was based off of the old Pentium 3 cores but if it had less than half the performance of its competitors, would anyone really care?

Sentiment to old VIA ... (and Centaur)
I like retro computers and would be equally interested in revisiting PowerPC derivatives

 

[NostaSeronx]

Still wondering if... they'll ever go back to HLMC/HH Fab.

ZX-D is on 28nm HLMC.
ZX-E is on 16nm TSMC.
ZX-F is on 7nm TSMC.

28HKC+: +15% perf/-30% power
22ULP: -10% area/+10% perf/-20% power
22ULL: -10% power
22FD/14FF+/12FF: ~+55% performance/~-50% power(both from 28LP/HKC) //FF is a full shrink and 12FF should be like 22ULP with -10% area vs 14FF+.

ICRD is doing 7nm/5nm FinFET. => https://www.researchgate.net/publication/340345456_A_Device_Design_for_5_nm_Logic_FinFET_Technology (Reference is TSMC/ ICRD is HH Fab version)
HLMC seems to be set to adopt those processes in a third fab eventually.

There is even High Mobility SOI FinFET and Stacked Nanosheet papers on IEEE explore with HLMC tagged.

 

[Hitman928]

This is not the same.

But, yes I agree power consumption in review is higher but this is full form facthor Zhaoxin developement board also with 40nm I/O chip ZX-200 @ 6W TDP etc. who is not suitable.

- 16-lane PCIe slot (lane width is x8, though).
- one 4-lane
- three single-lane PCIe slots,
- one old-school PCI slot.
- VGA,
- HDMI
- DisplayPort outs
- Four SATA 3.0 connectors
- One PCIe M.2
- One USB 3.1 Gen 2 port on one Type C connector
- One USB 3.1 Gen 2 port on one Type C pin header
- Two USB 3.1 Gen 1 ports on one Type A connector
- Two USB 3.1 Gen 1 ports on one pin header
- Two USB2.0 ports on one Type A connector
- Eight USB2.0 on x4 pin header
- Two UART ports
- One Audio Codec ALC662

70W TDP for EightCore KX-U6780A 2.7GHz is too much by 16nm FinFet CMOS technology.

Tom's measured (approximately) 55 W of power used by the CPU which has a 70W TDP, this has nothing to do with the board power. The Zhaoxin chip has absolutely abysmal perf/w compared to its contemporaries. I honestly think they might just be better off using as modern of an ARM design as they can and then emulating x86 for compatibility.

 

[Kosusko]

I underclocked to 2.7GHz my Eight Core processor AMD FX-8300 3.3GHz

Zhaoxin KaiXian KX-U6780A 2.7GHz vs AMD FX-8300 3.3GHz @ 2.7GHz


Cinebench R20

Multi-threaded
Zhaoxin KX-U6780A 2.7GHz: 982 (+11,72%)
AMD FX-8300 @ 2.7GHz: 879 (89,51%)

Single-threaded
Zhaoxin KX-U6780A 2.7GHz: 127 (94,07%)
AMD FX-8300 @ 2.7GHz: 135 (+6,30%)


Geekbench 4

Multi-threaded
Zhaoxin KX-U6780A 2.7GHz: 9 128 (+8,99%)
AMD FX-8300 @ 2.7GHz: 8 375 (91,75%)

Single-threaded
Zhaoxin KX-U6780A 2.7GHz: 1 780 (94,53%)
AMD FX-8300 @ 2.7GHz: 1 883 (+5,79%)

 

[Markfw]

I underclocked to 2.7GHz my Eight Core processor AMD FX-8300 3.3GHz

Zhaoxin KaiXian KX-U6780A 2.7GHz vs AMD FX-8300 3.3GHz @ 2.7GHz


Cinebench R20

Multi-threaded
Zhaoxin KX-U6780A 2.7GHz: 982 (+11,72%)
AMD FX-8300 @ 2.7GHz: 879 (89,51%)

Single-threaded
Zhaoxin KX-U6780A 2.7GHz: 127 (94,07%)
AMD FX-8300 @ 2.7GHz: 135 (+6,30%)


Geekbench 4

Multi-threaded
Zhaoxin KX-U6780A 2.7GHz: 9 128 (+8,99%)
AMD FX-8300 @ 2.7GHz: 8 375 (91,75%)

Single-threaded
Zhaoxin KX-U6780A 2.7GHz: 1 780 (94,53%)
AMD FX-8300 @ 2.7GHz: 1 883 (+5,79%)

Underclocking an 8 year old (not sure the year they came out) to allow this cpu to win does not speak well for it. And the underclocked 8 year old CPU won in single thread.

I wouldn;t be bragging about that. Here is the current 8 core AMD cpu. 5738 single core and 36,846 multi-threaded. 45 times as fast in single thread, and 38 times as fast in multi-thread.

https://imgur.com/a/aWlsBfb

 

[Kosusko]

I'm glad that you are happy with your 7nm AMD Ryzen 7 3700X 3.6GHz @ 4.4GHz Turbo (8C/16T).
But the point is, that every solded Zhaoxin processor is one unsold AMD or Intel processor.
And which of the two companies will it hurt more !?!
"Houston, we have a problem"
Take care.

 

[Markfw]

I'm glad that you are happy with your 7nm AMD Ryzen 7 3700X 3.6GHz @ 4.4GHz Turbo (8C/16T).
But the point is, that every solded Zhaoxin processor is one unsold AMD or Intel processor.
And which of the two companies will it hurt more !?!
"Houston, we have a problem"
Take care.

If you look at my signature, I don't even have a 3700x. I just googled for a benchmark to compare the current 8 core cpu.

And I don't understand your statement about an unsold CPU. Who would want one that gets beaten my a downclocked 8 year old CPU, or completely devastated by the current 8 core CPU ??????

 

[Kosusko]

I'm sorry. You have the Ryzen 3900X.
Maybe only a minority of Chinese users.
It can also be quite enough.

P.S.
And don't remember that the KH-40000 up to 32cores (two slot solutions up to 64cores) is on its way.
Following will be the KX-7000 Series

 

[DrMrLordX]

But the point is, that every solded Zhaoxin processor is one unsold AMD or Intel processor.
And which of the two companies will it hurt more !?!

That's not really true. The people in the market for these things probably can't or won't afford a machine from Intel or AMD. They're the same crowd that's been buying VIA/Zhaoxin CPUs (or worse, Loongson/Godson CPUs) for ages. It's a market that remains largely untapped for companies like Intel that demand such high margins that they (Intel) would never consider selling to it seriously, except maybe via contra revenue (see: Baytrail).

AMD is busy trying to take over the server sector. Ask them if they care about Zhaoxin CPUs eating into their margins. Here's a hint: they don't.

 

[Kosusko]

We will see.
I look forward to the anandtech review.

 

[eek2121]

I'm sorry. You have the Ryzen 3900X.
Maybe only a minority of Chinese users.
It can also be quite enough.

P.S.
And don't remember that the KH-40000 up to 32cores (two slot solutions up to 64cores) is on its way.
Following will be the KX-7000 Series

The issue is that you can get faster, older Intel/AMD chips for much less money. Multicore is only worth it if your single core performance is up to par.

 

[moinmoin]

I guess the biggest relevancy for Zhaoxin CPUs is for the Chinese government agencies' push for self sufficiency in IT, which got accelerated due to the trade war with the US. With both Intel and AMD being US companies (and THATIC being frozen out due to said trade war) they aren't a possible choice anyway. China wants to reach 100% self sufficiency with 20 million units by the end of 2022 so we can expect increased demand and development for Zhaoxin and other homegrown IT companies through that vector even if price/performance is really not there atm.