sepheronx wrote:kvs wrote:sepheronx wrote:
That aside, after looking at the only review of the Elbrus 4C through CNews.ru, I can safely say that the processor shows a lot of potential. The fact that it came close to competing against a Core i3 and i5, with having little to no real CISC processor development experience before, besides working on the Elbrus E2K core, has shown Russia is more than capable of producing a chip that can easily compete with the big players.
Baikal processors through Baikal electronics, subsidiary of T-Platforms may have a higher chance of being a direct competitor (and it will be good for MCST as they will have internal competition), due to the fact that T-Platforms (a major HPC making company) will be using these processors in the development of their cluster systems, which they seem to have a bigger access to the market than MCST as they have super computers built even in the USA. So in this case, T-Platforms Baikal processors (whenever they will end up showing up for sale), could very well be the basic market consumer processor, especially since ARM processors are making a huge push into the consumer market (good for a RISC processor).
The Elbrus CPU line is a competitor for Xeons and Opterons and should be deployed in Russia for scientific and engineering computing. T-Platforms should adopt this CPU instead of relying on Intel or AMD parts. Unfortunately, the Elbrus CPUs take a large performance hit when doing 64 bit floating point operations compared to 32 bit operations. This is the price of having a low power design with fewer transistors. MCST should release a variant that does not have this limitation and this does not require a ground-up redesign.
The Elbrus design has demonstrated that VLIW can work just fine and Russia has managed this feat. Intel tried VLIW with the Itanium line but has failed.
Exactly. Problem with 64bit is that VLIW isn't native x86 thus it has to have a separate instruction for it. And thus, that is where the performance hit is (theoretically) compared to lets say Intel or AMD's since both of those are CISC (I had it wrong the whole time, thought Elbrus was CISC and not VLIW). That being said, I agree, they can indeed increase transistor counts significantly. I guess we will not know fully until when Elbrus 8C is shown. That said, am looking forward to see what it will be, as well, what the future holds after that. I know they want to move as well onto APU's. Now Baikal electronics is moving into RISC with its future ARM processor and any other processor afterwards.
I forgot to clarify the point about software. For science and engineering what one needs is a good Fortran 90 compiler and there are not that many useful off the shelf software packages. There are more commercial packages for engineering, but even there it is typical to resort to custom code. So the CISC vs. VLIW architecture differences are not a problem. Also, these days you cannot get far without a large amount of CPUs operating in parallel via MPI and to a lesser extend OpenMP. Russia needs to develop good clusters based on the Elbrus.
The Baikal CPU is based on the ARM architecture and is good for integer heavy business computing.
It was good to hear that Mikron has started producing the Elbrus-2S in Russia and has upgraded their equipment to produce 65 nm parts. The fabrication equipment was rated for 90 nm production initially. So Russia is developing, slowly, a proper microelectronics industry.