Wednesday, February 19th 2025
Loongson's Next-gen 3B6600 CPU Reportedly Targeting Intel 12/13th Gen Level Performance
Loongson Technology has announced its next-generation 3B6600 processor—according to recent Chinese press reports, early details were disclosed in "investor relations activity" documentation. Their eight-core design is still in development, with the fabless company (allegedly) setting some lofty performance goals. According to a MyDrivers report (citing Fast Technology), the 3B6600 processor's single-core and multi-core "high level" performance is touted to match that of Core i5 and i7 offerings from Intel's 12th and 13th Generation portfolios. In the past, Loongson has compared its older products to Western equivalents. Last year, their 3A6000 chip design was mentioned in the same sentence as "AMD Zen 4 and Intel Raptor Lake" processor families. Around November 2023, reports had it approximating Team Blue Core i3-10100 "Comet Lake" CPU performance.
The next-gen 3B6600 CPU is supposedly using its predecessor (3A6000) as a springboard; MyDrivers believes that the same in-house LoongArch ISA design (fabricated on 12 nm/14 nm) will be revisited. The publication highlighted a significant area of optimization: "single-core performance of 3B6600 is expected to be in the world's leading ranks...Loongson 3B6600 will continue to use mature technology, and the architecture core will be upgraded to the new LA864. The same frequency performance is greatly improved by about 30%, compared with the Loongson 3A6000's LA664 architecture." Loongson engineers are reportedly targeting a maximum turboboost frequency of 3.0 GHz, but under normal operation the 3B6600 chip is predicted to offer a main frequency of 2.5 GHz. The new design will integrate a new "LG200 GPGPU" graphics computing core—additionally, supported standards include: DDR5 memory, PCIe 4.0 bus, and HDMI 2.1.
Sources:
MyDrivers News, Tom's Hardware
The next-gen 3B6600 CPU is supposedly using its predecessor (3A6000) as a springboard; MyDrivers believes that the same in-house LoongArch ISA design (fabricated on 12 nm/14 nm) will be revisited. The publication highlighted a significant area of optimization: "single-core performance of 3B6600 is expected to be in the world's leading ranks...Loongson 3B6600 will continue to use mature technology, and the architecture core will be upgraded to the new LA864. The same frequency performance is greatly improved by about 30%, compared with the Loongson 3A6000's LA664 architecture." Loongson engineers are reportedly targeting a maximum turboboost frequency of 3.0 GHz, but under normal operation the 3B6600 chip is predicted to offer a main frequency of 2.5 GHz. The new design will integrate a new "LG200 GPGPU" graphics computing core—additionally, supported standards include: DDR5 memory, PCIe 4.0 bus, and HDMI 2.1.
14 Comments on Loongson's Next-gen 3B6600 CPU Reportedly Targeting Intel 12/13th Gen Level Performance
The problem is that it remains full of incompatibility issues;
"CPUs are only as useful as the programs they can run. Loongson managed to make a mess out of the situation at several levels. At the OS level, Loongson’s CPUs have two binary-incompatible Linux distributions. At the application level, Loongson faces software optimization problems. Developers will optimize programs for the most common CPUs like x86 and ARM ones. Web browsers are one example. Loongson ships LBrowser v3, a Chromium based browser. It performs horribly on the Speedometer 3.0 browser benchmark:"
Can China’s Loongson Catch Western Designs? Probably Not.
I remember playing games and doing homework on a 386 laptop, to you know these things will be useful
They run hotter and faster for 20 or 30 seconds.
You can run them in an efficiency mode and use less power.
They certainly cost less.
but unless there is some ground breaking new architecture, material or lithography tech, we are plateaued.
China will catch up.
"What CPU have you got, mate?"
"Erm, wait, let me look it up." *Searches notes*
they should just compare while crosscompiling the compile times. I do not care for clock rates. I care for post times, compile times and fps.
those pcie 5.0 or ddr5 bus rates are just peripherals and not that important.
I want to see how long it takes to cross-compile app-office/libreoffice-24.2.7.2, sys-devel/gcc-14.2.1_p20241221
jup it’s stuck in my head now, even displacing the the lion sleeps to night by the tokens.
Meanwhile Western companies will concentrate on share buy backs, executive renumeration, and keeping investors happy for the next 3 months. Look at what has become of Intel.
Both lagging behind a few gens and both suffering from software incompatibilities.
I'd say they are doing better than Intel, considering GPU hardware directly targets the same DX, Vulkan, SPIRV etc. standards while CPU hardware targets completely different assembly languages (MIPS, superset of RISCV with some simplified instructions) which are then used to make compilers for higher level languages.
A business model that only works with constant growth is not a good business model in my opinion. Slow and steady wins the race as they say.
I want those other architectures succeed with coreboot. most stuff runs anyway in a browser these days.