Huawei's New CPU Matches Zen 3 In Single-Core Performance (tomshardware.com) 77
Long-time Slashdot reader AmiMoJo quotes Tom's Hardware: A Geekbench 6 result features what is likely the first-ever look at the single-core performance of the Taishan V120, developed by Huawei's HiSilicon subsidiary (via @Olrak29_ on X). The single-core score indicates that Taishan V120 cores are roughly on par with AMD's Zen 3 cores from late 2020, which could mean Huawei's technology isn't that far behind cutting-edge Western chip designers.
The Taishan V120 core was first spotted in Huawei's Kirin 9000s smartphone chip, which uses four of the cores alongside two efficiency-focused Arm Cortex A510 cores. Since Kirin 9000s chips are produced using SMIC's second-generation 7nm node (which may make it illegal to sell internationally according to U.S. lawmakers), it would also seem likely that the Taishan V120 core tested in Geekbench 6 is also made on the second-generation 7nm node.
The benchmark result doesn't really say much about what the actual CPU is, with the only hint being 'Huawei Cloud OpenStack Nova.' This implies it's a Kunpeng server CPU, which may either be the Kunpeng 916, 920, or 930. While we can only guess which one it is, it's almost certain to be the 930 given the high single-core performance shown in the result. By contrast, the few Geekbench 5 results for the Kunpeng 920 show it performing well behind AMD's first-generation Epyc Naples from 2017.
The Taishan V120 core was first spotted in Huawei's Kirin 9000s smartphone chip, which uses four of the cores alongside two efficiency-focused Arm Cortex A510 cores. Since Kirin 9000s chips are produced using SMIC's second-generation 7nm node (which may make it illegal to sell internationally according to U.S. lawmakers), it would also seem likely that the Taishan V120 core tested in Geekbench 6 is also made on the second-generation 7nm node.
The benchmark result doesn't really say much about what the actual CPU is, with the only hint being 'Huawei Cloud OpenStack Nova.' This implies it's a Kunpeng server CPU, which may either be the Kunpeng 916, 920, or 930. While we can only guess which one it is, it's almost certain to be the 930 given the high single-core performance shown in the result. By contrast, the few Geekbench 5 results for the Kunpeng 920 show it performing well behind AMD's first-generation Epyc Naples from 2017.
Performance is Relative.. (Score:1)
..to expected results driven by stock price.
Let me know when Greed N. Corruption starts bullshitting otherwise.
To infinity and beyond! (Score:2)
CPU speeds aren't that critical for daily use and haven't been for a long time.
My iPhone 6 ran as well as my iPhone 13 in terms of -perceived- performance for common daily tasks.
This matters for science, engineering, higher end gaming and some other niche areas but there's nothing wrong with a 2017 cpu for a phone today.
Re: (Score:3)
Performance and efficiency go hand in hand (with the exception of a couple generations of Intel CPU that just ran extra hot instead.
You may not need the high performance but you might need long battery life - performance per watt goes both ways.
Re: (Score:3)
Those hot couple being gen6,gen7,gen8,gen9,gen10,gen11,gen12,gen13 and gen14.
Re: (Score:2)
No way before that. In terms of heat and temperature around about Gen 6 was the point where we stopped statically clocking / dumb boosting CPUs and started to dynamically control their frequency based on available power envelope. AMD does the same thing. A hot CPU is a hard working CPU, you want it to be hot (when not idle) or you're just artificially gimping your own performance. It has little to do with efficiency figures.
Re: (Score:2)
And why do you think we can still get get decent performance with it clocked down? Because the performance per watt is higher than previous generations.
Re: (Score:1)
It's surprising how fast they are catching up though. We might have a third competitor in the high performance x86 space before too long.
Re: (Score:3)
How are they a "third competitor" in the x86 space? While not specifically mentioned, these Taishan cores are more than likely ARM based cores considering the rest of the SOC are ARM based efficiency cores and ARM based GPU. Also how are they "catching" up? The M1 released in 2020 beat the Epyc 7413 in single core performance in the same benchmark. The Qualcomm Snapdragon 8+ Gen 1 beats the Epyc 7413 starting in 2022.
The Epyc 7413 is a server based CPU that is not designed for great single core performance
Re: (Score:2)
Huawei Socs are indeed Arm-based.
Re: (Score:2)
That would definitely be a good thing.
Re: (Score:2)
Let me know when they have comparative benchmarks in programs that mean anything. Geekbench is worthless for predicting performance in anything but applications which don't need performance.
Re: (Score:1, Offtopic)
It's 2024, we Americans are about to wrapped up in 7 months of nonstop electioneering where we the citizenry get to decide on the makeup of our two main governmental bodies.
Is that happening in Russia? China? No? It's a pretty important distinction if you ask me...
Re: (Score:3)
Re: (Score:1, Offtopic)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2, Insightful)
Biden should issue an executive order to have him executed
The case against Trump is the expectation that he may drag the US into an authoritarian regime. You propose that Biden execute him to avert the danger. A President plotting the assassination of a political opponent does not avert authoritarianism, it implements it.
Re: (Score:3, Interesting)
His point was that granting Trump immunity for life means that Biden can then execute anyone he pleases and face no consequences.
Re: (Score:2)
Re: (Score:2)
I understood the main point regarding immunity. I did not comment on it because I agreed, though I think a better analogy for your argument would be: Biden goes to the street corner and hires a hitman (or convinces one of his followers).
What I understand from the accusations against Trump are things done aside of due process (e.g. making phone calls, trying to influence others), while your original example of Executive Order is a formal use of powers, for which assassinations are not excluded.
"The question
Re: (Score:2)
Re: (Score:2)
how stupid it would be in a democracy to offer a president unlimited immunity, even for executive actions.
I don't enjoy providing a bad example, but the constitution of France does exactly that, except for deportation to the International Criminal Court in The Hague (high-body-count war crime scenarios).
This is specifically designed to provide immunity for the official acts, and not to common criminality such as beating the wife. There is an implicit requirement of the official act to be legal -- such as an Executive Order taken after consulting the relevant agencies and determining that the President has autho
Re: (Score:2)
And yet you know all this becuase we have a free press in the US and all this is taking place in the US justice system where previous court orders were "slapped" down (you like that part of the justice system I notice?) and Fani Willis makes the news and is all taking place in courts and in the public square.
Just because you don't like some of the process or outcomes doesn't make them anywhere, remotely close to equivalent to China or Russia where these circumstances are remarkably different.
Doesn't matter (Score:2)
The US government mandates that vendors sign a disclosure if they use any Huawei stuff (among others) in products.
Yes, but still a dead end (Score:3)
Re: (Score:2)
Re: (Score:2)
And then what happens if they manage to leapfrog ASML and the west becomes dependent on Chinese litho?
There is no serious analyst on earth who believes this. Not even in China. You might as well ask "what if the moon was made of cheese?"
Re: (Score:3)
EUV is a specific technology, not a node size. Yes ASML is the only company with EUV chasing that node size. But in the same node size they have competition from Canon using a pressed mask rather than EUV process.
Re: (Score:2)
But in the same node size they have competition from Canon using a pressed mask rather than EUV process.
That is IF it works. I have not read anyone that has purchased a Canon NIL machine and actually used it to make products. I have doubts that it can achieve the same results as EUV as it requires the mask to physically touch the silicon. Besides all the problems with that approach, it would seem that masks would have to be replaced at a higher frequency than a photomask
Re: (Score:3)
I also never said that EUV was a specific node size. What I'm suggesting is that you will not get smaller than ~7nm (depending on the definition used on what "node" size even is, because it varies) using DUV equipment.
Re: (Score:2)
You are making the same mistake, again. Every time China develops something, there is some excuse why it is a one-off, won't progress any further, the sanctions will work this time.
And every time you are proven wrong. We have got to get a grip and start competing. Even though ASML is European, I'd love to see more competition in that space. I hope Canon can get their system into fabs as soon as possible, and keep improving it. We need to keep pushing ahead, and not relying on sanctions and dubious assumptio
Re: (Score:2)
Well played China.
Re: (Score:2)
Do not think that last thing is going to happen, but 7nm is impressive. Also, I assume China can get EUV a bit faster than it took ASML. If you have a (vague) idea what works, the research goes a lot faster. Also sure that China has other "creative" means to go ahead. This is happening. Well played China.
With lithography it goes beyond just research. If that were the case, Intel would not have been stuck on 10nm for 5 years. There are probably a lot of specific knowledge that ASML is not sharing with outsiders. At this point, China has to invent a new technique as sourcing the parts as ASML's EUV is all but impossible.
Re: Yes, but still a dead end (Score:2)
Re: (Score:2)
I think you vastly underestimate the Chinese ability to make their own EUV equipment. Sanctions work for awhile. And then what happens if they manage to leapfrog ASML and the west becomes dependent on Chinese litho?
Please name the Chinese company that can make EUV equipment. Only ASML makes that machine right now and entire countries that have made lithography equipment in the past like the US and Japan do not make EUV machines. The Japan and US are not under sanctions so they should have zero obstacles, right?
Re: (Score:2)
None, but since you're asking you completely misunderstood the point. The point isn't that a Chinese company can make EUV equipment. The point is they will. The Chinese have proven very adapt at espionage too so you can count on the development of the technology for them not taking as long as it did for western companies.
Point is this brick wall is like Trump's wall, it'll be defeated with a ladder.
Re:Yes, but still a dead end (Score:4, Informative)
None, but since you're asking you completely misunderstood the point. The point isn't that a Chinese company can make EUV equipment. The point is they will
How will a Chinese company make EUV equipment under sanctions when entire countries not under sanctions cannot make EUV machines? You might as well ask for unicorns while you are at it.
The Chinese have proven very adapt at espionage too so you can count on the development of the technology for them not taking as long as it did for western companies.
Espionage does little if Chinese companies cannot make the subcomponents, if they cannot buy the parts, and if they cannot make the parts for the subcomponents. For example, where will China get the lenses used in ASML machine? Only one company (Carl Zeiss SMT) can make them and they are partially owned by ASML. All China has to do is develop multiple industries from scratch to be on par with the best in the world when the rest of world cannot do it when not limited. Does that make any sense?
Re: (Score:3)
Re: (Score:2)
This would be the same aircraft carrier that is complete crap because of shitty design right? Including among other problems a design flaw that makes it impossible for both runways to be used at once, massive issues with the EM cats themselves, and the ski jump style takeoff causing problems with the large engine craft? Oh, and lets not forget that it takes over 10 hours to get underway. An improvement to the 48 hours of the previous two, but still crap.
Re: (Score:2)
Re: (Score:2)
No one said that China couldn't build planes or aircraft carriers.
Aircraft carriers are 1920s technology, they're also not particularly advanced. They are a large ship (which china can absolutely definitely build) with a funky deck. What they are is horribly expensive and you need a crew for which you need a trained crew to train the crew and they are logistically complicated and since no one will tell you about that kind of thing, you generally need to build a bunch in series to figure it out.
The question
Re: (Score:2)
"How will a Chinese company make EUV equipment under sanctions when entire countries not under sanctions cannot make EUV machines?"
They 100% have the industrial capability to make the stuff themselves. Literal districts are dedicated to specific technologies, there's even one dedicated directly to semiconductors (as in literally 'Semiconductor District' is in the physical address.) They simply lack the knowledge to build the experimentation and expertise levels up. That's all they need to steal. Meanwhile,
Re: (Score:2)
They 100% have the industrial capability to make the stuff themselves.
Please tell me which company can make the lenses that EUV machines use. Which company makes the equipment to fire molten tin vaporized by laser into a plasma for the light source? Those are just two of the components.
Literal districts are dedicated to specific technologies, there's even one dedicated directly to semiconductors (as in literally 'Semiconductor District' is in the physical address.) They simply lack the knowledge to build the experimentation and expertise levels up. That's all they need to steal. Meanwhile, most other countries both lack an industrial base and an academic base, so they're non-starters from the get-go.
No one has said China cannot make semiconductors. No one. What is said is they cannot make the same cutting edge chips as TSMC as China does not have the technology. Specifically they cannot buy the EUV machines they need. They cannot make the EUV machines. They cannot make the parts to build t
Re: (Score:2)
China is not good at inventing. They are good at copying.
I remember when the same was said of the Japanese.
They have the resources, including a lot of smart people, the motivation, and know it can be done, to eventually do it, though likely not in a reasonable time frame.
Re: (Score:2)
I remember when the same was said of the Japanese. They have the resources, including a lot of smart people, the motivation, and know it can be done, to eventually do it, though likely not in a reasonable time frame.
The Japanese have the resources, the people, and no sanctions. In fact Canon and Nikon are Japanese companies considered two of the top lithography machine manufacturers before EUV. Yet the Japanese cannot make EUV machines.
Re: (Score:2)
Not the same motivation though. If something has been done, it can be done again, eventually.
Re: (Score:2)
The US specifically denied Canon and Nikon the opportunity to license the technology (they tried).
Nobody in the western world can legally make EUV without a license from the US.
Re: (Score:2)
"Please tell me which company can make the lenses that EUV machines use"
None, because EUV gets absorbed by pretty much everything, so they use mirrors in a vacuum.
"They cannot make the EUV machines. They cannot make the parts to build the EUV machines."
I've been there, you've obviously not. They have anything and everything they need, and if they do not, stealing it and copying it and understanding it and improving upon it is right around the corner - espionage runs rampant now days.
Re: (Score:2)
How will a Chinese company make EUV equipment under sanctions when entire countries not under sanctions cannot make EUV machines? You might as well ask for unicorns while you are at it.
Because they have reason to, and the countries not under sanctions don't.
Other countries probably could if they had the motivation too, but it's extremely expensive so it's much cheaper to buy ready made machines from someone else if you can.
Re: (Score:2)
Because they have reason to, and the countries not under sanctions don't.
Is your point about will? How does the will produce parts that China cannot produce at the moment. Some of the parts of EUV machines are single sourced meaning one company in the world makes the part. China has to create many, many industries and make them the best in the world just to assemble a EUV machine. Or China has to figure out a different methodology. They have been trying to use DUV but can only produce comparitively simple CPUs.
Other countries probably could if they had the motivation too, but it's extremely expensive so it's much cheaper to buy ready made machines from someone else if you can.
I am not sure the Japanese need much motivation. After all, Canon and
Re: (Score:2)
They also have a reason to not have missile systems fueled with water instead of rocket fuel yet here we are.
Re: (Score:2)
Re: (Score:2)
The Chinese have several parallel projects to develop EUV technologies. The Chinese made their own space station. And you seem to think making a mere EUV machine is too complicated for them to do. Good luck with that.
Japan did make their own EUV light source. They just haven't commercialized a machine because the market isn't large enough to fund development of a second vendor i.e. everyone who wanted to buy an EUV machine had already backed ASML's EUV machine.
Thanks to the US forcing the Netherlands to sto
Re: (Score:2)
"Impenetrable" only for the stupid. They will just develop EUV themselves. Protectionism has never worked, but it has often made things worse for those engaging in it.
Re: (Score:2)
It's also the same basic core they used in a cellphone SoC already observed last year. They used it in a Kunpeng server CPU this time around, with no comprehensive benchmarks.
It's not terrible but it's not competitive either.
Re: (Score:2)
Also the mutli-core score of this supposed CPU is abysmal [huaweicentral.com]. It scored a 2806 while AMD Zen 3 7763 scored a 12324. Literally more than four times faster. And we don't even know the power draw of this supposed Huawei server CPU. My guess it is dramatically worse per watt.
Re: (Score:2)
Meaningless benchmark and meaningless result (Score:5, Informative)
These benchmarks were not done nor witnessed by Tom's Hardware. Rather they got the results from a Tweet from Huawei. So no one knows the exact test conditions and no one can verify if they are true or the benchmarks were not rigged. After all Intel has been caught rigging benchmarks. There are no other benchmarks shown; the only benchmark was a single core test.
What I find interesting is that Huawei is not comparing this smartphone CPU with other ARM based smartphone CPU. Instead they are comparing them to x86 server/workstation CPUs? Why? More than likely comparing the Taishan V120 to other smartphone CPUs like Apple A series or Qualcomm's Snapdragon would show how far behind they are. This processor is behind Apple's A13 (2019) [geekbench.com]. It is behind Qualcomm ARM CPUs [geekbench.com] made in 2020.
The comparison to server/workstation CPUs serves one purpose; to misdirect. Server/workstation CPUs are designed to handle multicore workloads. Comparing a single core test is like comparing a passenger car on a circular track with a semi-truck pulling a trailer. Yes the passenger beats the truck but the purpose of the truck is not be the fastest vehicle around a track.
Re: (Score:2)
It's also Grekbench which is less-than-useful, especially for a server CPU.
Re: (Score:2)
Possibly cos they can't get it as efficient / cool running as a mobile device CPU, so they compare it to CPUs which are not as heat / power constrained - hence server / workstation CPUs.
Re: (Score:2)
Huawei has a massive server business. They are selling these CPUs in servers as well as phones, so the benchmarks are relevant. They also use them in network equipment, including 5.5G and their passive optical network gear.
Re: (Score:2)
Huawei has a massive server business. They are selling these CPUs in servers as well as phones, so the benchmarks are relevant. They also use them in network equipment, including 5.5G and their passive optical network gear.
Then why didn't they publish multicore results? Because it would have been crushed. So Huawei compared them to other network equipment in GeekBench 6 results. No. Because no one does a Geekbench comparison that for network equipment. People buying that equipment is not concerned how fast the CPU runs a single benchmark as how much traffic the equipment handles overall.
Re: (Score:2)
Yes, the same reason why Apple focuses on very specific benchmarks to hide the true performance of their CPUs. And Intel, for that matter. And probably AMD, even though they don't need to.
Re: (Score:2)
Yes, the same reason why Apple focuses on very specific benchmarks to hide the true performance of their CPUs. And Intel, for that matter. And probably AMD, even though they don't need to.
Apple does not publish only one benchmark and compares only workstation/server CPUs to their smartphone CPUs. At this point you are desperate to justify your tribalism.
So as fast as apple silicon (Score:2)
I assume the problem will be efficiency and getting more cores on the chip as they have a much worse manufacturing node.
Re: (Score:2)