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

Page 18 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hitman928

Diamond Member
Apr 15, 2012
5,259
7,855
136
New KH40000 results:

2 Processors, 32 Cores
ZHAOXIN KaiSheng KH-40000/16
source: https://browser.geekbench.com/v5/cpu/15706425

CentaurHauls Family 6 Model 71 Stepping 2 vs CentaurHauls Family 7 Model 11 Stepping 3
CNS vs KH-40000
L1 Instruction Cache is 64KB per core and this is differend from Centaur CNS
L3 Cache is 8MB per eigt core cluster and this is differend from Centaur CNS with 16MB L3 cache per eight core
source: https://browser.geekbench.com/v5/cpu/compare/12878360?baseline=15706425

Comparison to a state of the art Epyc:

 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,542
14,498
136
average 709% faster in multi-core, 257.2% in single core. And thats Milan, nothing new. I really don't know why anybody would care about such a slow processor. My $1100 7452 EPYC would beat this thing easy. 32c64t. And no mention of power, mine is one processor, 155 watt. Who knows what these others are power wise.
 

Kosusko

Member
Nov 10, 2019
161
120
116
I believe that they are primarily intended for the Chinese market and secondarily for enthusiasts around the world.
There are enthusiasts who do not want to be mainstream. They want to be different.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,542
14,498
136
I believe that they are primarily intended for the Chinese market and secondarily for enthusiasts around the world.
There are enthusiasts who do not want to be mainstream. They want to be different.
If they are for the China market, why do you keep pushing them here ? Getting ANY EPYC is not mainstream. If you want to be different, that is the way you would go, not for something much slower and most likely takes a lot more power.
 

Kosusko

Member
Nov 10, 2019
161
120
116
Refreshed processor: higher frequencies and especially a new integrated graphics card

ZHAOXIN KaiXian KX-6000G (4C 3.2GHz, 4MB L2) with ZX C1080 GPU (SM6.0, 512MB, PCI) (D3D 11)
source: https://ranker.sisoftware.co.uk/sho...af92a385edd0e5c3bb86b791f491ac9cbac9f4cc&l=en
source: https://ranker.sisoftware.co.uk/sho...9bab8de4d9ff97aa8cf4c9ef8aefd2e2c4b78ab2&l=en


Interesting performance for an integrated graphics core as a successor to IP S3 Graphics/VIA.
 
Last edited:

Kosusko

Member
Nov 10, 2019
161
120
116
CENTAUR CHA IN THE TEST
The last x86 processor of its kind

167757-345058-345051_rc.jpg


source: https://www.golem.de/news/centaur-cha-im-test-der-letzte-x86-prozessor-seiner-art-2209-167757.html

P.S. There is an interesting comment in the discussion that it is allegedly possible that the former Centaur Technology team is working on a project (Sierra Forest?)
 

prosty_mirek

Junior Member
Nov 1, 2020
13
4
51
So Zhaoxin skipped CNS, AI cores, that CHA concept (though owning patents), and all that post 16nm nodes, to sneak peek regression on CPU+GPU integration. SMIC 7-ish nm needs small chips to better yields?
That kx-7000 GPU looks big to cores.
Zx-1080 is nice improvement from c-960 on synthetics. But MIA.
DGPU transfered to Glenfly Arise and also MIA.
Still that should be a big jump for chinese market.
 
  • Like
Reactions: Kaluan

NostaSeronx

Diamond Member
Sep 18, 2011
3,686
1,221
136
Been rummaging around for a bit. VIA/Zhaoxin might have switched over to RISC-V for CPU.

KX-7000 (last x86 processor) -> KX-8000 (first RISC-Five[RISC-V] processor)
VIA-Zhaoxin Alliance(VZA PC and AI products) implies KX-7000 is 6nm, while KX-8000 is 5nm.
 

prosty_mirek

Junior Member
Nov 1, 2020
13
4
51
Been rummaging around for a bit. VIA/Zhaoxin might have switched over to RISC-V for CPU.

KX-7000 (last x86 processor) -> KX-8000 (first RISC-Five[RISC-V] processor)
VIA-Zhaoxin Alliance(VZA PC and AI products) implies KX-7000 is 6nm, while KX-8000 is 5nm.
Looks like Kx-7000 was skipped.
 

Attachments

  • Screenshot_2023-10-13-18-43-01-449_com.android.chrome-edit.jpg
    Screenshot_2023-10-13-18-43-01-449_com.android.chrome-edit.jpg
    401.8 KB · Views: 27
  • Like
Reactions: lopp09

DrMrLordX

Lifer
Apr 27, 2000
21,626
10,835
136
Wow, blast from the past. Sad to see Zhaoxin go that route, but with all the uh troubles for Chinese designers, is anyone surprised?
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,686
1,221
136
Wow, blast from the past. Sad to see Zhaoxin go that route, but with all the uh troubles for Chinese designers, is anyone surprised?
Zhaoxin is a founding member of:
"The China RISC-V Industry Consortium (CRVIC) was co-established in Shanghai on September 20, 2018 by 65 main RISC-V enterprises and institutions in China. As the first China local RISC-V alliance, CRVIC supports and assists the global RISC-V Foundation in several aspects, and has a friendly collaborative relationship with China RISC-V Alliance (CRVA), which was set up in November 2018."

Which was reported here:

Also listed here:

With that:
ZX-F (KX-7000) ~ x86 and CNS-based => 256-bit units (could have AVX512) // VIA-Zhaoxin Alliance 6nm product
ZX-H (KX-8000) ~ RISC-V and CNS-based => 256-bit units, definitely has 2048-bit vector width support. // VIA-Zhaoxin Alliance 5nm product

My rummaging around the dirt lead me to VIA being the one launching the product globally, thus at TSMC. With a similar venture of PC-1 and APC in the works. Leading to something like VPC/PC-V for the RISC-V processor. Which would also be a return to CNR/ZX-C, yr 2015 times, where VIA and Zhaoxin launched the same product. I'll be watching if VIA finally finalizes the Neo-ITX standard of boards, with VIA Labs coming out with a OCulink or USB4 thing on it.
 
Last edited:
  • Like
Reactions: lopp09

Dubble

Junior Member
Jul 1, 2022
9
4
41
Some Chinese OEMs I know have tested the KX-7000.
The Zhaoxin KaiXian KX-7000 processor is a complex SOC with various units and pipelines to handle different types of instructions.

Introduce KX-7000 architecture in here

Decoder Units:
  • Four decoder units.(0,1,2,3)
  • 1-uop instructions can be decoded by any of the four decoders in one cycle.
  • 2-uop instructions can be decoded by decoder 0, decoder 1, or decoder 2 in one cycle.
  • Complex (microcode) instructions can only be decoded by decoder 0.
Pipeline Structure:
  • Ten pipelines in total: Integer execution pipelines (4), Address calculation/load/store pipelines (2), and Floating-point pipelines (4).
The KX-7000 looks like a lower-power version of the i7-6900K, or Zen+ with GPU.

The KX-7000 is an engineering sample (ES version) featuring eight cores and eight threads. The single-core frequency is 4 GHz and more, while the overall core frequency ranges from 3.0 to 3.6 GHz. It adopts a CCD and IOD separate configuration and utilizes an LGA packaging.

The eight cores are interconnected via a ring bus, and they share a 32MB L3 cache.

The I/O Die incorporates a GPU with DirectX 12 support, a modified version of the c1080. The video encoding and decoding performance has been enhanced. The processor supports dual-channel DDR5 memory and provides 16 PCIe 4.0 lanes, along with compatibility for USB4 interfaces.

In terms of performance, the CPU-Z single-core score is approximately 400, and the multi-core score is around 3000. The average power consumption falls within the range of 60-100 watts.

QQ图片20231114225803.jpg
 
Last edited:

prosty_mirek

Junior Member
Nov 1, 2020
13
4
51
Some Chinese OEMs I know have tested the KX-7000.
The Zhaoxin KaiXian KX-7000 processor is a complex SOC with various units and pipelines to handle different types of instructions.

Introduce KX-7000 architecture in here

Decoder Units:
  • Four decoder units.(0,1,2,3)
  • 1-uop instructions can be decoded by any of the four decoders in one cycle.
  • 2-uop instructions can be decoded by decoder 0, decoder 1, or decoder 2 in one cycle.
  • Complex (microcode) instructions can only be decoded by decoder 0.
Pipeline Structure:
  • Ten pipelines in total: Integer execution pipelines (4), Address calculation/load/store pipelines (2), and Floating-point pipelines (4).
The KX-7000 looks like a lower-power version of the i7-6900K, or Zen+ with GPU.

The KX-7000 is an engineering sample (ES version) featuring eight cores and eight threads. The single-core frequency is 3.7 GHz, while the overall core frequency ranges from 3.0 to 3.6 GHz. It adopts a CCD and IOD separate configuration and utilizes an LGA packaging.

The eight cores are interconnected via a ring bus, and they share a 32MB L3 cache.

The I/O Die incorporates a GPU with DirectX 12 support, a modified version of the c1080. The video encoding and decoding performance has been enhanced. The processor supports dual-channel DDR5 memory and provides 16 PCIe 4.0 lanes, along with compatibility for USB4 interfaces.

In terms of performance, the CPU-Z single-core score is approximately 400, and the multi-core score is around 3000. The average power consumption falls within the range of 60-100 watts.
View attachment 89489
View attachment 89490
So that Schrodinger's CPU is zx-g based now? Or they changed arch behind zx-f, once again? Because previous zx-f leaked results (from 2020?) was definitly CNS. Yongfeng is already different than CHA.
It is missleading. Kx-7000 is few years late now. There are public traces of kx-8000, not kx-7. Zx-f was CNS, but kh-40000 is not (or not purely).

So, to clear things:
- Zx-f, CNS based, not released due to transfer Centaur's team to Intel (Zhaoxin is still patents holder).
- Zx-f2, Yongfeng, only for kh-40000, non CHA concept.
- Zx-f3/zx-g, Yongfeng+ with chiplets, for kx-7000?
 

prosty_mirek

Junior Member
Nov 1, 2020
13
4
51
Introduce KX-7000 architecture in here

Decoder Units:
  • Four decoder units.(0,1,2,3)
  • 1-uop instructions can be decoded by any of the four decoders in one cycle.
  • 2-uop instructions can be decoded by decoder 0, decoder 1, or decoder 2 in one cycle.
  • Complex (microcode) instructions can only be decoded by decoder 0.
Your description is basically a copy-paste from Yongfeng:
The rules for the decoder are simple:
;; - an instruction with 1 uop can be decoded by any of the four
;; decoders in one cycle.
;; - an instruction with 2 uops can be decoded by decoder 0 or decoder 1
;; or decoder 2 but still in only one cycle.
;; - a complex (microcode) instruction can only be decoded by
;; decoder 0, and this takes an unspecified number of cycles.
 

Dubble

Junior Member
Jul 1, 2022
9
4
41
Your description is basically a copy-paste from Yongfeng:
Yes, KX-7000 is based on yongfeng, it may be a modification of yongfeng.
In recent years, Zhaoxin has thoroughly understood the design of the CPU by rewriting important modules, iterating Centaur's CNS and the old yongfeng architecture, and many modules in the pipeline have been implemented with new designs. But it seems that the architectural improvements have not brought about improvements in IPC, and performance has improved through frequency.
 

Dubble

Junior Member
Jul 1, 2022
9
4
41
Some Chinese OEMs I know have tested the KX-7000.
The Zhaoxin KaiXian KX-7000 processor is a complex SOC with various units and pipelines to handle different types of instructions.

Introduce KX-7000 architecture in here

Decoder Units:
  • Four decoder units.(0,1,2,3)
  • 1-uop instructions can be decoded by any of the four decoders in one cycle.
  • 2-uop instructions can be decoded by decoder 0, decoder 1, or decoder 2 in one cycle.
  • Complex (microcode) instructions can only be decoded by decoder 0.
Pipeline Structure:
  • Ten pipelines in total: Integer execution pipelines (4), Address calculation/load/store pipelines (2), and Floating-point pipelines (4).
The KX-7000 looks like a lower-power version of the i7-6900K, or Zen+ with GPU.

The KX-7000 is an engineering sample (ES version) featuring eight cores and eight threads. The single-core frequency is 3.7 GHz, while the overall core frequency ranges from 3.0 to 3.6 GHz. It adopts a CCD and IOD separate configuration and utilizes an LGA packaging.

The eight cores are interconnected via a ring bus, and they share a 32MB L3 cache.

The I/O Die incorporates a GPU with DirectX 12 support, a modified version of the c1080. The video encoding and decoding performance has been enhanced. The processor supports dual-channel DDR5 memory and provides 16 PCIe 4.0 lanes, along with compatibility for USB4 interfaces.

In terms of performance, the CPU-Z single-core score is approximately 400, and the multi-core score is around 3000. The average power consumption falls within the range of 60-100 watts.
View attachment 89489
View attachment 89490
So that Schrodinger's CPU is zx-g based now? Or they changed arch behind zx-f, once again? Because previous zx-f leaked results (from 2020?) was definitly CNS. Yongfeng is already different than CHA.
It is missleading. Kx-7000 is few years late now. There are public traces of kx-8000, not kx-7. Zx-f was CNS, but kh-40000 is not (or not purely).

So, to clear things:
- Zx-f, CNS based, not released due to transfer Centaur's team to Intel (Zhaoxin is still patents holder).
- Zx-f2, Yongfeng, only for kh-40000, non CHA concept.
- Zx-f3/zx-g, Yongfeng+ with chiplets, for kx-7000?
There was an error in the introduction of the previous test. The KX 7000 was not at the 3.7GHz frequency, but at a higher frequency.
The KX7000 is no ordinary KX-7000 processor, but an engineering sample with more options.
The Best score in 4GHz and more, use GCC 14, may be Overclocking frequency or max boost frequency.
The new test switches the optimization option, O2 to Ofast, and enables more vector instructions.
40 -> 44.3
1701445860582.png
 

Kosusko

Member
Nov 10, 2019
161
120
116
Zhaoxin Kaixian® KX-7000 Series Processor


CgAGbGV31GOASTNvAAS2g6SRSgk001.jpg



Zhaoxin launches Kaixian KX-7000 series high-performance desktop processors with x86 Century Avenue architecture



• 8 cores / 8 threads
• maximum working frequency of 3.7 GHz
• 4 MB L2 cache and 32 MB L3 cache
• supports dual-channel DDR5 memory with a maximum capacity of 128 GB
• integrated high-performance C-1190 iGPU graphics card, supports the latest APIs such as DX12, OpenCL 1.2 and OpenGL 4.6, support H.265 HEVC/H.264 AVC. DP/HDMI/VGA output, supports dual-channel 4K
• support for 24x PCIe 4.0, 2x USB 4 + 4x USB 3.2 Gen2 + 2x USB 2.0, 3x SATA Gen3 interface
• compatible with x86 32/64-bit instructions, SSE4.2/AVX/AVX2 extended instructions, SM2, SM3 and SM4 accelerated instructions
• supports virtualization technologies such as VT-x and VT-d 2.5
• LGA 1700 (45 mm x 37.5 mm)


CgAGbGV3zT2AGRftAAHNBqd8smc615.jpg



source: https://www.zhaoxin.com/prod_view.aspx?nid=3&typeid=593&id=2757
 
  • Like
Reactions: lightmanek

NTMBK

Lifer
Nov 14, 2011
10,233
5,015
136
Zhaoxin Kaixian® KX-7000 Series Processor


CgAGbGV31GOASTNvAAS2g6SRSgk001.jpg



Zhaoxin launches Kaixian KX-7000 series high-performance desktop processors with x86 Century Avenue architecture


• 8 cores / 8 threads
• maximum working frequency of 3.7 GHz
• 4 MB L2 cache and 32 MB L3 cache
• supports dual-channel DDR5 memory with a maximum capacity of 128 GB
• integrated high-performance C-1190 iGPU graphics card, supports the latest APIs such as DX12, OpenCL 1.2 and OpenGL 4.6, support H.265 HEVC/H.264 AVC. DP/HDMI/VGA output, supports dual-channel 4K
• support for 24x PCIe 4.0, 2x USB 4 + 4x USB 3.2 Gen2 + 2x USB 2.0, 3x SATA Gen3 interface
• compatible with x86 32/64-bit instructions, SSE4.2/AVX/AVX2 extended instructions, SM2, SM3 and SM4 accelerated instructions
• supports virtualization technologies such as VT-x and VT-d 2.5
• LGA 1700 (45 mm x 37.5 mm)


CgAGbGV3zT2AGRftAAHNBqd8smc615.jpg



source: https://www.zhaoxin.com/prod_view.aspx?nid=3&typeid=593&id=2757
I hope we'll see some proper comparison benchmarks. Maybe Anandtech will get a sample? Fingers crossed.