See excellent article. The 3A6000 is Loongson’s best and newest and performs between the level of Zen (GloFo 14-12nm) and Zen 2 (TSMC 7nm) but limited to four cores.
A lot depends on the cost to produce that chip, whether Loongson can scale below 14-12nm they’re using currently (EUV without ASML?), and what development resources Loongson can deploy for their next microarchitecture revisions, which aren’t discussed here.
> The 3A6000 is Loongson’s best and newest and performs between the level of Zen (GloFo 14-12nm) and Zen 2 (TSMC 7nm) but limited to four cores.
Which is fine. For most applications, it's the order of magnitude of performance that matters. Not +10..30% or say, Zen 3 vs. Zen 2 (although gamers & AMD/Intel execs will argue otherwise).
Machines with >10y old cpu's are still perfectly useable for (most) everyday tasks, as long as the software support is there. Especially when coupled with enough RAM, decent gpu, SSD etc.
On the way from "pointless museum piece" to "latest & greatest", this cpu is 9/10s there.
> it's the order of magnitude of performance that matters
Finally a sound voice in this merry go round of one-upping. Totally agree. The funny thing is, if people paid half as much attention to optimizing the bloat crap they build, instead of drooling over 10% cpu gains, you’d see a much bigger improvements. Obviously it depends on the domain, but everything I’ve come across (except core tech like SQLite etc) have huge overhead in both memory and cpu that could easily be optimized if someone paid attention.
Like the json parsing issue with gta online, that’s a perfect yet egregious example of software practices today.
I think you missed the "limited to 4 cores" part. Zen 2 has up to 64, which considering it is also slower per core makes it 20 times slower. If like me you count 2x as order of magnitude, that's 4+ orders of magnitude away.
They are certainly accelerating. I think the consensus is that it is within fabrication there is a "chokepoint".
In terms of design, China has world class companies. However in terms of fabrication and especially semiconductor manufacturing equipment, they are still somewhat behind.
Also x86 is very sticky - it is going to take them a decade to get rid of that.
> In terms of design, China has world class companies.
And not just companies. This is currently the world's top open-source RISC-V development https://github.com/OpenXiangShan coming from the Chinese Academy of Science.
(Whether China will remain investing in RISC-V, given that the US government has started to pressure US RISC-V development to limit their involvement with China is another question.)
Interestingly, they've got binary translators for both ARM and x86 which were demoed in Geekerwan's video. Apparently, they're apparently donating a compilation farm for a Debian repository.
Definetly interesting to see how Lithography will develop independently in China, and if it will ever pass the EUV chokepoint.
See excellent article. The 3A6000 is Loongson’s best and newest and performs between the level of Zen (GloFo 14-12nm) and Zen 2 (TSMC 7nm) but limited to four cores.
A lot depends on the cost to produce that chip, whether Loongson can scale below 14-12nm they’re using currently (EUV without ASML?), and what development resources Loongson can deploy for their next microarchitecture revisions, which aren’t discussed here.
> The 3A6000 is Loongson’s best and newest and performs between the level of Zen (GloFo 14-12nm) and Zen 2 (TSMC 7nm) but limited to four cores.
Which is fine. For most applications, it's the order of magnitude of performance that matters. Not +10..30% or say, Zen 3 vs. Zen 2 (although gamers & AMD/Intel execs will argue otherwise).
Machines with >10y old cpu's are still perfectly useable for (most) everyday tasks, as long as the software support is there. Especially when coupled with enough RAM, decent gpu, SSD etc.
On the way from "pointless museum piece" to "latest & greatest", this cpu is 9/10s there.
> it's the order of magnitude of performance that matters
Finally a sound voice in this merry go round of one-upping. Totally agree. The funny thing is, if people paid half as much attention to optimizing the bloat crap they build, instead of drooling over 10% cpu gains, you’d see a much bigger improvements. Obviously it depends on the domain, but everything I’ve come across (except core tech like SQLite etc) have huge overhead in both memory and cpu that could easily be optimized if someone paid attention.
Like the json parsing issue with gta online, that’s a perfect yet egregious example of software practices today.
I think you missed the "limited to 4 cores" part. Zen 2 has up to 64, which considering it is also slower per core makes it 20 times slower. If like me you count 2x as order of magnitude, that's 4+ orders of magnitude away.
2 replies →
They are certainly accelerating. I think the consensus is that it is within fabrication there is a "chokepoint".
In terms of design, China has world class companies. However in terms of fabrication and especially semiconductor manufacturing equipment, they are still somewhat behind.
Also x86 is very sticky - it is going to take them a decade to get rid of that.
> In terms of design, China has world class companies.
And not just companies. This is currently the world's top open-source RISC-V development https://github.com/OpenXiangShan coming from the Chinese Academy of Science.
(Whether China will remain investing in RISC-V, given that the US government has started to pressure US RISC-V development to limit their involvement with China is another question.)
Interestingly, they've got binary translators for both ARM and x86 which were demoed in Geekerwan's video. Apparently, they're apparently donating a compilation farm for a Debian repository. Definetly interesting to see how Lithography will develop independently in China, and if it will ever pass the EUV chokepoint.
A decade or two isn't a long time, politically.