Huawei's HiSilicon Taishan V120 Server Core Matches Zen 3 Performance

[AleksandarK]

Huawei's new server CPU based on the HiSilicon Taishan V120 core has shown impressive single-threaded performance that matches AMD's Zen 3 architecture in a leaked Geekbench 6 benchmark. The Taishan V120 is likely being manufactured on SMIC's 7 nm process node. The Geekbench 6 result posted on social media does not identify the exact Huawei server CPU model, but speculation points to it being the upcoming Kunpeng 930 chip. In the benchmark, the Taishan V120 CPU operating at 2.9 GHz scored 1527 in the single-core test. This positions it nearly equal to AMD's EPYC 7413 server CPU based on the Zen 3 architecture, which boosts up to 3.6 GHz and which scored 1538 points. It also matches the single-threaded performance of Intel's Coffee Lake-based Xeon E-2136 from 2018, even though that Intel chip can reach 4.5 GHz boost speeds, scoring 1553 points.

The Taishan V120 core first appeared in Huawei's Kirin 9000 smartphone SoC in 2020. Using the core in server CPUs would allow Huawei to achieve competitive single-threaded performance to rival AMD's last-generation EPYC Milan and Intel's older Skylake server chips. Multi-threaded benchmarks will be required to gauge the Kunpeng 930's overall performance fully when it launches. Huawei continues innovating its ARM-based server CPU designs even while facing restrictions on manufacturing and selling chips internationally due to its inclusion on the US Entity List in 2019. The impressive single-threaded results versus leading x86 competitors demonstrate Huawei's resilience and self-reliance in developing homegrown data center technology through its HiSilicon division. More details on the Kunpeng 930 server chip will likely surface later this year, along with server configurations from Chinese OEMs.



View at TechPowerUp Main Site | Source

 

[Space Lynx]

Impressive, sort of... when you consider the methods and means, maybe... ha ha ok I'll stop

 

[ncrs]

[...]AMD's latest EPYC Milan[...]

Milan's from 2021. The fourth generation EPYC - Genoa was launched in 2022.

 

[stimpy88]

AMD will one day come to regret the day it sold the early Zen architecture to the Chinese.

 

[AleksandarK]

Milan's from 2021. The fourth generation EPYC - Genoa was launched in 2022.

Meant last generation lol. Thanks for pointing out!

 

[TumbleGeorge]

AMD will one day come to regret the day it sold the early Zen architecture to the Chinese.

What you mean?

 

[Daven]

AMD will one day come to regret the day it sold the early Zen architecture to the Chinese.

Does that mean Intel will also regret licensing x86 to AMD?

 

[ncrs]

What you mean?

There was a joint venture that resulted in customized Zen 1 designs for the Chinese market. As far as I understood the agreement it was AMD that made the modifications internally, and GlobalFoundries manufactured the wafers. Only packaging was handled by a Chinese company. I'm not sure how much of the architecture was actually available to the Chinese side.

Does that mean Intel will also regret licensing x86 to AMD?

Or AMD licensing amd64 to Intel?

 

[evernessince]

AMD will one day come to regret the day it sold the early Zen architecture to the Chinese.

Big difference between a Chinese company licensing a custom design based off Zen IP and buying the IP. AMD did the former, not the latter.

 

[NeuralNexus]

AMD will one day come to regret the day it sold the early Zen architecture to the Chinese.

I doubt it given this is an ARM architecture