New CPU from China's sole x86 chipmaker grapples with AMD's Bulldozer in Geekbench results — lower-end model isn't the fastest, but it gives Zhaoxin a full roster of CPUs for use in the country [Updated]

News
By Matthew Connatser
published

The chip is slow, to say the least.

Zhaoxin
(Image credit: Zhaoxin)

Update 12/24/23: Clarified details of this new 'G2' variant. 

A Geekbench 6 result for Zhaoxin's x86 KX-6640MA G2 CPU has been uncovered, and the performance is about on par with an early 2010s AMD Bulldozer CPU (via @BenchLeaks). The new chip falls into the KX-6000G series. The KX-6000 series came out in 2019, but the KX-6640MA G2 is a relatively recent model that arrived in late 2021, and this Geekbench 6 result is perhaps the first time it has ever been publicly tested. 

On Geekbench 6, the 6640MA G2 scored 386 points in the single-core test and 1,110 points in the multi-core test. For reference, that's just about on par with AMD's Bulldozer generation FX-4100 from 2011, which Geekbench 6 rates at 414 and 1,103 points in single- and multi-core performance, respectively.

Swipe to scroll horizontally
KX-6000 Specs
Row 0 - Cell 0 KX-6640MAKX-6640AKX-U6880AAMD FX-4100
Cores4484
Threads4484
Base Frequency2.2 GHz2.6 GHz3 GHz3.6 GHz
Boost Frequency2.6 GHzN/AN/A3.9 GHz
Cache4MB L24MB L28MB L24MB L2 + 8MB L3

The KX-6640MA G2 is the lowest-end member of the KX-6000 family, with just four cores, 4MB of L2 cache (it has no L3 cache), and a boost clock of 2.6 GHz. That it has boost technology at all is notable, as no other KX-6000 CPU does. Given that the 6640MA G2 is otherwise identical to the 6640A, that boost feature might be its reason for existing, and it might have come in handy in this mini-PC featuring the 6640MA. We've also learned that the 'G' suffix denotes an updated GPU core, to revision c-1080, that has higher clock speeds and support for DX12.

Obviously, being a tad slower than a 2011 CPU that was already regarded as pretty slow isn't ideal, but there are some silver linings to the 6640MA G2. Although Zhaoxin's low-end CPU has two-thirds the frequency and one-third the cache of the FX-4100, it's just about as fast, which indicates a much more robust architecture. Not to mention, the 6640MA G2 has PCIe 3.0 support instead of just 2.0, DDR4 memory, and integrated graphics.

Zhaoxin is also making progress in delivering its next-generation KX-7000 series. These new chips come with even higher clock speeds, up to 32MB of L3 cache, and support for DDR5 memory and PCIe 4.0. The Chinese CPU designer has only shared the specifications for its top-end KX-7000 CPU and hasn't detailed what low-end models will look like, so we can only assume that such a low-end model will generally have better specifications, just not by how much.

Matthew Connatser
Matthew Connatser

Matthew Connatser is a freelancing writer for Tom's Hardware US. He writes articles about CPUs, GPUs, SSDs, and computers in general.

See more CPUs News
15 Comments Comment from the forums
  • artk2219
    Admin said:
    Zhaoxin's KX6640MA x86 processor has popped up on Geekbench 6, and it's about as good as AMD's old Bulldozer chips.

    New CPU from China's sole x86 chipmaker grapples with AMD's Bulldozer in Geekbench results — lower-end model isn't the fastest, but it gives Zhaoxi... : Read more
    That puts it around Nehalem IPC, thats pretty decent honestly, that definitely gets it into usable territory provided they can get some higher clocks.
    Reply
  • hotaru251
    the 1 amd name you don't want to be compared to >_<!

    tbh im shocked china didn't try to push for arm/riscv advancement over x86.
    Reply
  • ivan_vy
    hotaru251 said:
    the 1 amd name you don't want to be compared to >_<!

    tbh im shocked china didn't try to push for arm/riscv advancement over x86.
    I would think for HW and SW compatibility/ecosystem.
    Reply
  • ivan_vy
    artk2219 said:
    That puts it around Nehalem IPC, thats pretty decent honestly, that definitely gets it into usable territory provided they can get some higher clocks.
    the article failed to mention the KX-6640MA has 25W TDP while the FX-4100 has 95W TDP.
    Reply
  • ingtar33
    artk2219 said:
    That puts it around Nehalem IPC, thats pretty decent honestly, that definitely gets it into usable territory provided they can get some higher clocks.
    no it does not. the 8130 was bulldozer, which was decidedly slower then nehalem. that was why it was called faildozer. the late gen phenomII cpus were about on par with stock Nehalem when overclocked, but bulldozer was a significant step BACKWARDS in performance from the Phenom II, hence "Faildozer"

    Piledriver would be about on par if not slightly faster in some workloads with Nehalem (with both at stock), however when Piledriver launched Intel had already launched Sandybridge and Ivybridge was about to launch too.... and piledriver was not competitive with either.
    Reply
  • artk2219
    ingtar33 said:
    no it does not. the 8130 was bulldozer, which was decidedly slower then nehalem. that was why it was called faildozer. the late gen phenomII cpus were about on par with stock Nehalem when overclocked, but bulldozer was a significant step BACKWARDS in performance from the Phenom II, hence "Faildozer"

    Piledriver would be about on par if not slightly faster in some workloads with Nehalem (with both at stock), however when Piledriver launched Intel had already launched Sandybridge and Ivybridge was about to launch too.... and piledriver was not competitive with either.
    Correct, but its 2.2 - 2.6 Ghz vs 3.6 - 4Ghz for the 4100 which it is matching with a much lower clock speed, making its IPC much higher, placing maybe between Nehalem and Sandy Bridge in terms of IPC. Other chips in this performance range would be the Core 2 Quad Q9650 at 3Ghz, I5 - 760 clocking in at 2.8 - 3.3 Ghz, and Phenom II X4 965 at 3.4 Ghz. Thats a pretty huge improvement in IPC compared to their previous offerings, and would make that a generally useful web surfing chip.
    Reply
  • DavidC1
    That sounds nice until you realize that Tremont performs at the Ivy Bridge level and Gracemont is Skylake level.

    According to Geekbench 6 results, Pentium Silver N5000 scores about 360 in ST and 1000 in MT at 2.7GHz clocks. Meaning Zhaoxin only about 10% faster than that chip. The N5000 performed around 45nm Penryn level so that makes sense.

    N5000 used Intel's 14nm process while this chip uses "16nm" so TSMC. N5000 had only 6W TDP though.

    10% makes it somewhere in the Nehalem range for ST as Nehalem was only a small gain for it if you normalize it(disable SMT and boost).
    Reply
  • George³
    Why does the title claim this is a new processor. This is not true, it is misleading and raises doubts that the article was released for political propaganda purposes. Although it is cleverly disguised with technical data and does not have characteristic political cliché expressions. And yes I read that in text is mentioned the real new series Zhaoxin KX-7000. In one sentence.
    Reply
  • frogr
    ivan_vy said:
    the article failed to mention the KX-6640MA has 25W TDP while the FX-4100 has 95W TDP.
    FX-4100 is 32 nm. KX-6640MA is 16nm, thus the power difference.
    Reply
  • DavidC1
    frogr said:
    FX-4100 is 32 nm. KX-6640MA is 16nm, thus the power difference.
    So what? I just pointed out the 6W N5000 is Intel 14nm.

    The Via chips used to be popular in low power devices. They aren't used anymore, because they aren't competitive. 4x power difference with mere 1 generation gap(in the best case scenario) is absolutely enormous. They would need to jump to N5 to have any chance of competing with the Intel 14nm N5000.

    Bulldozer level of perf/clock is useless when people complain about Gracemont, the chip that's at least as good as Skylake!

    George³ said:
    And yes I read that in text is mentioned the real new series Zhaoxin KX-7000. In one sentence.
    The 7000 series is faster but not enough:
    https://browser.geekbench.com/v5/cpu/compare/526995?baseline=458824
    The N100 Gracemont at 3.4GHz scores 1200, meaning it's 50% faster per clock. So the previous generation Tremont is 12% faster than Zhaoxin 7000 per clock, not to mention the incomparable power advantage. You'd need to go down to Meteorlake's LP-E cores(which sip power) for somewhat ballpark performance. I hear even the LP-E cores clock in the 2.xGHz range meaning even that's significantly faster.
    Reply