Discussion of Russian economics (split from TSW)

[phongn]

Since the mid 90s the Chinese state-sponsored processor development effort has consistently trailed Intel / IBM / AMD etc by about eight years.

So China's cloning of Pentium 4 factors in here how?

Do you have a source for their Pentium-4 clone? I'm fairly curious about that. The only Chinese CPU project I've been aware of is the Loongson series, which is a very different design (MIPS-lineage)

[K. A. Pital]

The only Chinese CPU project I've been aware of is the Loongson series, which is a very different design (MIPS-lineage)

Yeah, that. I mistakenly described it as a Pentium-4 clone (I read about it a while ago in the news, in 2006 in fact, and it was described as equal in performance, more or less, to the P4).

P4 with 180nm tech cores was released in 2000, with 90nm tech cores in 2004. China was slightly lagging behind, obviously, but in 2006 it released the Loongson 2E with 90nm tech already. The gap is somewhere between 2 (counting by technology) and 6 (prely by clock rate) years, and in my view, the gap is contracting. For example, right now China is working on below 50nm cores, and the recently launched Core i7 with 32nm tech is supposed to be replicated by the latest Loongson CPU, which is also aiming for 32nm, if I read correctly.

[Vympel]

Slight bump, but:-

The demography issue is what spooks me the most, Stas, in a "general foreboding" kind of way...

I think the worst of that is over. The fall in population slowed significantly recently, and 2009 reported net population growth - meager but there.

BBC news

Last year saw a falling death rate, rising birth rate, and increased migration.

[Starglider]

The gap is somewhere between 2 (counting by technology) and 6 (prely by clock rate) years, and in my view, the gap is contracting.

Clock rate is trivial to achieve in isolation, but it won't do you any good. Similarly, you can get massive IPC using VLIW, but that won't necessarily get you any real performance either (see; Itanium). Intel went from selling 3.8 GHz P4s to 2.2 GHz Core 2s, and the later were 20-40% faster. Having a low half-pitch on your process is also pretty meaningless, if the functional feature size (e.g. gate length) and process characteristics (switching/leakage power, swing rate, stray capacitance, electromigration vulnerability, defect density, lots of others) are not good. Intel, IBM etc regularly demo chips using processes two or three years before they're ready for commercial use; it takes that long to get the mass production costs down (particularly yield) and performance (particularly leakage, recently) up. Meanwhile most chips (e.g. GPUs) are made on processes that are substantially inferior to CPU processes even though they have the same half-pitch; if the current 40nm Nvidia and AMD GPUs were made on the same process as the 45nm Intel i7s, they would be substantially faster and more efficient despite the larger half-pitch.

The relevant way to measure processor performance is benchmarks of products available to customers, and by this metric Chinese processor development has been trailing by around eight years.

[phongn]

The relevant way to measure processor performance is benchmarks of products available to customers, and by this metric Chinese processor development has been trailing by around eight years.

Looking at performance numbers, Godson-3A @ 800MHz has a SPECint2000 score of 345 and SPECfp2000 score of 289[1]. That is roughly equivalent to a P3-800EB[2], released in 2000 or an Atom Z500[3] released in 2008.

[1] Gao X, Chen YJ, Wang HD et al. System architecture of Godson-3 multi-core processors. JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY 25(2): 181–191 Mar. 2010

[2] http://www.spec.org/cpu2000/results/

[3] http://www.tgdaily.com/hardware-feature ... processors