China's Loongson reveals new 3A6000 CPU, claims performance on par with 3-year-old Intel chips

[erek]

Pretty cool

“Loongson 3A6000 uses the new LoongArch architecture, a poorly documented ISA which was mostly derived from MIPS. The RISC-based MIPS architecture was developed by the namesake US company, which retired the technology in 2021 and adopted the RISC-V architecture. Loongson continued MIPS development after licensing the MIPS32 and MIPS64 architectures in 2011, and LooongArch is speculated to include design choices derived from both MIPS and RISC-V.


Even if performance claims over CPU hardware are true, Loongson will likely have a more difficult time on the software front. Linux kernel and operating systems still don't provide full support for the LoongArch ISA, and a working Windows version that could run on the new processors is clearly out of the question.”

Source: https://www.techspot.com/news/99718-china-loongson-reveals-new-3a6000-cpu-claims-performance.html

 

[1_rick]

"
Loongson states that these results put the 3A6000 quad-core CPU on par with quad-core x86-64 processors released by Intel three years ago."

That's Comet Lake i3s. Color me skeptical.

 

[scottypippin]

Another interesting point from the article - some data is from a benchmark retired in 2018. I guess somebody is really trying to impress the CESI overlords in China (or else, trying to keep their job) by producing great numbers form out of date testing methods....

That was the "SPEC CPU 2006" benchmark though. I've never used Unixbench - anyone know how good 7400 points with that benchmark it is? I couldn't find an online repository of scores from that benchmark.

 

[stryder2720]

That's pretty sad, but expected from the CCP.

 

[philb2]

That's pretty sad, but expected from the CCP.

Look where that same "lie upwards" mentality has done for Russia in its war with Ukraine. Not to mention the corruption in both countries.

 

[Krenum]

That's pretty sad, but expected from the CCP.

Well, it takes time steal secrets and copy other peoples work!

 

[KD5ZXG]

Poorly documented?
https://loongson.github.io/LoongArch-Documentation/LoongArch-Vol1-EN.html

MIPS for comparison sake.
https://www.cs.cmu.edu/afs/cs/academic/class/15740-f97/public/doc/mips-isa.pdf

Not seeing sufficient similarity to cry wolf here...

 

[michalrz]

I've never used Unixbench - anyone know how good 7400 points with that benchmark it is? I couldn't find an online repository of scores from that benchmark.

I haven't used it myself as well, but I built it and it's running now.
Will update this post.

Edith: results.
First batch of numbers is on 1 core, the other is on all cores.
The system was already loaded with a VM consuming 25% of CPU and RAM, I was watching a movie and dicking around the forums.

It's a Ryzen 5 3600,
/proc/cpuinfo said ~4000 MHz on all cores,
32 gigs of 3200MHz CL16 Ripjaws V (4 x 8)
1050ti

Warning: long-ass output:

------------------------------------------------------------------------
Benchmark Run: wto sie 22 2023 09:29:22 - 09:57:19
12 CPUs in system; running 1 parallel copy of tests

Dhrystone 2 using register variables       51261733.5 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                     9120.7 MWIPS (10.0 s, 7 samples)
Execl Throughput                               5700.8 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks       1493153.3 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          410720.4 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       3502578.7 KBps  (30.0 s, 2 samples)
Pipe Throughput                             2257893.9 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 169914.8 lps   (10.0 s, 7 samples)
Process Creation                               8524.3 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                  16487.6 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                   8872.3 lpm   (60.0 s, 2 samples)
System Call Overhead                        2314943.6 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   51261733.5   4392.6
Double-Precision Whetstone                       55.0       9120.7   1658.3
Execl Throughput                                 43.0       5700.8   1325.8
File Copy 1024 bufsize 2000 maxblocks          3960.0    1493153.3   3770.6
File Copy 256 bufsize 500 maxblocks            1655.0     410720.4   2481.7
File Copy 4096 bufsize 8000 maxblocks          5800.0    3502578.7   6038.9
Pipe Throughput                               12440.0    2257893.9   1815.0
Pipe-based Context Switching                   4000.0     169914.8    424.8
Process Creation                                126.0       8524.3    676.5
Shell Scripts (1 concurrent)                     42.4      16487.6   3888.6
Shell Scripts (8 concurrent)                      6.0       8872.3  14787.1
System Call Overhead                          15000.0    2314943.6   1543.3
                                                                   ========
System Benchmarks Index Score                                        2327.4

------------------------------------------------------------------------
Benchmark Run: wto sie 22 2023 09:57:19 - 10:25:28
12 CPUs in system; running 12 parallel copies of tests

Dhrystone 2 using register variables      397497003.1 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                    91669.2 MWIPS (9.9 s, 7 samples)
Execl Throughput                              31521.0 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks      10698415.6 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks         3093001.1 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       7504485.7 KBps  (30.0 s, 2 samples)
Pipe Throughput                            17929066.6 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                2032153.1 lps   (10.0 s, 7 samples)
Process Creation                              65248.3 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                  85193.5 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                  11523.5 lpm   (60.0 s, 2 samples)
System Call Overhead                       19544849.6 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0  397497003.1  34061.4
Double-Precision Whetstone                       55.0      91669.2  16667.1
Execl Throughput                                 43.0      31521.0   7330.5
File Copy 1024 bufsize 2000 maxblocks          3960.0   10698415.6  27016.2
File Copy 256 bufsize 500 maxblocks            1655.0    3093001.1  18688.8
File Copy 4096 bufsize 8000 maxblocks          5800.0    7504485.7  12938.8
Pipe Throughput                               12440.0   17929066.6  14412.4
Pipe-based Context Switching                   4000.0    2032153.1   5080.4
Process Creation                                126.0      65248.3   5178.4
Shell Scripts (1 concurrent)                     42.4      85193.5  20092.8
Shell Scripts (8 concurrent)                      6.0      11523.5  19205.8
System Call Overhead                          15000.0   19544849.6  13029.9
                                                                   ========
System Benchmarks Index Score                                       13895.8

The benchmark takes a lot of time, so I'll re-run it at a cleaner runlevel (leaner, no other loads) later today.
"System Benchmarks Index Score" is probably the final number.

Edith II: Results from Runlevel 3 (text, multi-user, networking)

Worse for single core, better at all-cores. Sorry it's a photo, I forgot to >

 

[scottypippin]

michalrz thanks, that's mighty kind of you! I don't have a desktop quad core lying around to try and compare core to core, but is this the repo you cloned from? Maybe I'll give it a go:
https://github.com/kdlucas/byte-unixbench

I guess it is "performative"....according to that metric....if they reported the one number correctly and it was under typical user conditions? Guess I'm asking questions we'll never get answered. But, others are right: fat chance they'll be able to get the software support any time soon and let's face it - there are many reasons for that score to be falsified, cherry picked, or else not representative of real world tasks. If AMD and Intel (ordered alphabetically folks...not trying to start a riot ) will do it, I'd bet that a processor company under the CCP (who keeps tauting how advanced they are vs. "the West") is going to do it.

 

[michalrz]

michalrz thanks, that's mighty kind of you! I don't have a desktop quad core lying around to try and compare core to core, but is this the repo you cloned from? Maybe I'll give it a go:
https://github.com/kdlucas/byte-unixbench

I guess it is "performative"....according to that metric....if they reported the one number correctly and it was under typical user conditions? Guess I'm asking questions we'll never get answered. But, others are right: fat chance they'll be able to get the software support any time soon and let's face it - there are many reasons for that score to be falsified, cherry picked, or else not representative of real world tasks. If AMD and Intel (ordered alphabetically folks...not trying to start a riot ) will do it, I'd bet that a processor company under the CCP (who keeps tauting how advanced they are vs. "the West") is going to do it.

Yup, that is the repo.
I bet many distros have this available natively, Mint didn't seem to.
The prerequisite packages mentioned in the apt-get command are likely already on everyone's system.
All I did was:

sudo apt-get install libx11-dev libgl1-mesa-dev libxext-dev perl perl-modules make git   
git clone https://github.com/kdlucas/byte-unixbench.git
cd byte-unixbench/UnixBench/
./Run

The version it spat out in the benchmark header was:

Version 5.1.3                      Based on the Byte Magazine Unix Benchmark
Multi-CPU version                  Version 5 revisions by Ian Smith,
                                               Sunnyvale, CA, USA
January 13, 2011                   johantheghost at yahoo period com

And indeed, we don't know a lot of things.
There's a USAGE file in the directory, and the "how to interpret results" is a wall of text that mentions lots and lots of factors that might influence the result with just the first one being how it was compiled. Heck, there's some moving around of files in that suite, and I used a regular SATA 500GB SSD, not my NVMe.

The article said it was 4-core @ 2.5 GHz achieving 7400 points. So, guessing that was the multi-threaded run. Not a horrible score, I guess.
But, IMHO, as far as UnixBench goes, we should probably ignore this one as a metric to evaluate that chip, unless they provide every bit of info on the environment it was compiled and ran on.

 

[zandor]

A 4 core Intel chip from 3 years ago doesn't have to be a desktop chip, or even an i3. Intel has a few quad core Pentium models and then there are assorted low power mobile CPUs from 3 years ago with a TDPs of like 5-15W and base clocks in the 800-1200MHz range. Or something like that.

I looked up some SPEC 2006 results Intel published, and to put it in terms gamers think in their SPEC numbers look more like a Sandy Bridge i7-2600k (2011) except for multithreaded floating point, which is closer to a Haswell i7-4770 (2013). That may be doing pretty well if it's a low power chip or it could be total trash if they're burning 50W.