kvs wrote:From what I understand, chip production is an art. TSMC would produce the Elbrus using library component layouts. Intel spends a great deal of effort to hand optimize these transistor layouts before etching the silicon. The performance gain is substantial. One of the reasons the Elbrus as been stuck at a small clock frequency at each process resolution is because TSMC does not conduct such hand optimization. And MCST has had little chance to develop such optimization capacity. So we see clock speed gain essentially from process shrinkage and likely some attempt at MCST to make better use of the library layouts. I guess TSMC is also increasing its own capacity to work with customers to squeeze extra performance out. But that is not their focus. Russia needs an Intel type development and production.
Clock frequency is not the golden grail from the early 2000 era. Multiple cores clocked at a medium frequency provide much better performance. Architecture is the key. Compare the 3.0GHz Sandy Bridge CPUs against the Phenom II clocked at 3.8GHz. A Pentium 4 clocked at 5GHz won't come anywhere near a Core2Duo clocked at 2.8GHz, not even to mention the Sandy Bridge architecture.
This is towards 'kvs', 'sepheronx', and anyone else interested. As far as Russia's domestic electronics industry is concerned, KRET has made a serious step forward in the field of 'photonics', with incredibly promising early developments!
kvs wrote:From what I understand, chip production is an art. TSMC would produce the Elbrus using library component layouts. Intel spends a great deal of effort to hand optimize these transistor layouts before etching the silicon. The performance gain is substantial. One of the reasons the Elbrus as been stuck at a small clock frequency at each process resolution is because TSMC does not conduct such hand optimization. And MCST has had little chance to develop such optimization capacity. So we see clock speed gain essentially from process shrinkage and likely some attempt at MCST to make better use of the library layouts. I guess TSMC is also increasing its own capacity to work with customers to squeeze extra performance out. But that is not their focus. Russia needs an Intel type development and production.
Clock frequency is not the golden grail from the early 2000 era. Multiple cores clocked at a medium frequency provide much better performance. Architecture is the key. Compare the 3.0GHz Sandy Bridge CPUs against the Phenom II clocked at 3.8GHz. A Pentium 4 clocked at 5GHz won't come anywhere near a Core2Duo clocked at 2.8GHz, not even to mention the Sandy Bridge architecture.
Comparing the Pentium 4 with its ridiculously long narrow pipeline that was afflicted with branch bubbles with what I am discussing is not quite valid. The Elbrus is a VLIW and thus wide pipe design and cannot have the same clock frequencies as designs with many small pipeline stages (e.g. Pentium 4). My point is that for the specific Elbrus architecture, the performance takes a major hit from lack of in house capacity MCST to optimize the layout. When it was at 500 MHz it could have been at 1 GHz. That is a factor of two performance hit.
I run long duration geophysical simulations and a factor of two means months of time savings. I am using dozens of cores and shared memory MPI and get nice linear scaling. Others are not so lucky and doubling the number of CPUs does not imply 100% speedup. Some models don't scale effectively past 24 CPUs. Leaving the system and having to use an interconnect introduces a penalty even with the latest Infiniband adapters. The performance per core matters. Then there is the cost factor. It's not like people swim in CPU cores and the Elbrus is only going to have 8 cores per processor die only in the future. AMD sells 16 core chips now. There are systems out now based on 4 core Elbrus chips. That is only 16 CPUs per 1U rack unit. Throw in a 50% discount on the performance due to lack of layout optimization and that is basically an 8 CPU unit.
For the Russian military the Elbrus as it is on a 180 nm process is good enough. But Russia needs to have access to a modern, powerful CPU for its science and engineering. NATO already imposed sanctions limiting access to Intel and AMD CPUs by Russian system vendors. This BS cannot be tolerated. Russia does not need NATO and its products.
kvs wrote:From what I understand, chip production is an art. TSMC would produce the Elbrus using library component layouts. Intel spends a great deal of effort to hand optimize these transistor layouts before etching the silicon. The performance gain is substantial. One of the reasons the Elbrus as been stuck at a small clock frequency at each process resolution is because TSMC does not conduct such hand optimization. And MCST has had little chance to develop such optimization capacity. So we see clock speed gain essentially from process shrinkage and likely some attempt at MCST to make better use of the library layouts. I guess TSMC is also increasing its own capacity to work with customers to squeeze extra performance out. But that is not their focus. Russia needs an Intel type development and production.
Clock frequency is not the golden grail from the early 2000 era. Multiple cores clocked at a medium frequency provide much better performance. Architecture is the key. Compare the 3.0GHz Sandy Bridge CPUs against the Phenom II clocked at 3.8GHz. A Pentium 4 clocked at 5GHz won't come anywhere near a Core2Duo clocked at 2.8GHz, not even to mention the Sandy Bridge architecture.
Comparing the Pentium 4 with its ridiculously long narrow pipeline that was afflicted with branch bubbles with what I am discussing is not quite valid. The Elbrus is a VLIW and thus wide pipe design and cannot have the same clock frequencies as designs with many small pipeline stages (e.g. Pentium 4). My point is that for the specific Elbrus architecture, the performance takes a major hit from lack of in house capacity MCST to optimize the layout. When it was at 500 MHz it could have been at 1 GHz. That is a factor of two performance hit.
I run long duration geophysical simulations and a factor of two means months of time savings. I am using dozens of cores and shared memory MPI and get nice linear scaling. Others are not so lucky and doubling the number of CPUs does not imply 100% speedup. Some models don't scale effectively past 24 CPUs. Leaving the system and having to use an interconnect introduces a penalty even with the latest Infiniband adapters. The performance per core matters. Then there is the cost factor. It's not like people swim in CPU cores and the Elbrus is only going to have 8 cores per processor die only in the future. AMD sells 16 core chips now. There are systems out now based on 4 core Elbrus chips. That is only 16 CPUs per 1U rack unit. Throw in a 50% discount on the performance due to lack of layout optimization and that is basically an 8 CPU unit.
For the Russian military the Elbrus as it is on a 180 nm process is good enough. But Russia needs to have access to a modern, powerful CPU for its science and engineering. NATO already imposed sanctions limiting access to Intel and AMD CPUs by Russian system vendors. This BS cannot be tolerated. Russia does not need NATO and its products.
No arguing there. You seem to be very updated on the Russian CPU market situation. My knowledge is pretty amateurish as I'm reading the sheet and don't run computer simulations. However, they do need to start somewhere and it's necessary to get off that Western dependence. Russia has world class mathematicians and engineers, time to put them to good use.
kvs wrote:From what I understand, chip production is an art. TSMC would produce the Elbrus using library component layouts. Intel spends a great deal of effort to hand optimize these transistor layouts before etching the silicon. The performance gain is substantial. One of the reasons the Elbrus as been stuck at a small clock frequency at each process resolution is because TSMC does not conduct such hand optimization. And MCST has had little chance to develop such optimization capacity. So we see clock speed gain essentially from process shrinkage and likely some attempt at MCST to make better use of the library layouts. I guess TSMC is also increasing its own capacity to work with customers to squeeze extra performance out. But that is not their focus. Russia needs an Intel type development and production.
Clock frequency is not the golden grail from the early 2000 era. Multiple cores clocked at a medium frequency provide much better performance. Architecture is the key. Compare the 3.0GHz Sandy Bridge CPUs against the Phenom II clocked at 3.8GHz. A Pentium 4 clocked at 5GHz won't come anywhere near a Core2Duo clocked at 2.8GHz, not even to mention the Sandy Bridge architecture.
Comparing the Pentium 4 with its ridiculously long narrow pipeline that was afflicted with branch bubbles with what I am discussing is not quite valid. The Elbrus is a VLIW and thus wide pipe design and cannot have the same clock frequencies as designs with many small pipeline stages (e.g. Pentium 4). My point is that for the specific Elbrus architecture, the performance takes a major hit from lack of in house capacity MCST to optimize the layout. When it was at 500 MHz it could have been at 1 GHz. That is a factor of two performance hit.
I run long duration geophysical simulations and a factor of two means months of time savings. I am using dozens of cores and shared memory MPI and get nice linear scaling. Others are not so lucky and doubling the number of CPUs does not imply 100% speedup. Some models don't scale effectively past 24 CPUs. Leaving the system and having to use an interconnect introduces a penalty even with the latest Infiniband adapters. The performance per core matters. Then there is the cost factor. It's not like people swim in CPU cores and the Elbrus is only going to have 8 cores per processor die only in the future. AMD sells 16 core chips now. There are systems out now based on 4 core Elbrus chips. That is only 16 CPUs per 1U rack unit. Throw in a 50% discount on the performance due to lack of layout optimization and that is basically an 8 CPU unit.
For the Russian military the Elbrus as it is on a 180 nm process is good enough. But Russia needs to have access to a modern, powerful CPU for its science and engineering. NATO already imposed sanctions limiting access to Intel and AMD CPUs by Russian system vendors. This BS cannot be tolerated. Russia does not need NATO and its products.
This is particularly true, as there are major jumps in performance of Elbrus 2C+ and Elbrus 4, with both being the same architecture. somehow, it is limited by design on clock performance but let us wait and see. Elbrus 4C in various tests at 700mhz was very good performers and if clocked to same speed as the Core i7 920, it would have closer to same performance as predicted. So I would say that isn't so bad for a 65nm processor. Elbrus 8C rumored to be postponed a little to have an DDR4 memory controller.
sepheronx wrote:Elbrus 8C rumored to be postponed a little to have an DDR4 memory controller.
That's amazing news. Feature wise it will make it on par with Skylake.
Right now rumored, but first samples may be DDR3, dunno. But like Elbrus 2C+ being moved to Elbrus 2SM, Elbrus 8C would eventually get it. Apparently they are working or have completed the new chipset too for Elbrus 8C. 8C will be their performance line I imagine.
sepheronx wrote:Elbrus 8C rumored to be postponed a little to have an DDR4 memory controller.
That's amazing news. Feature wise it will make it on par with Skylake.
Right now rumored, but first samples may be DDR3, dunno. But like Elbrus 2C+ being moved to Elbrus 2SM, Elbrus 8C would eventually get it. Apparently they are working or have completed the new chipset too for Elbrus 8C. 8C will be their performance line I imagine.
Do you get information from insiders or do you dig deep on the internet? Any extra info on Baikal's status? Baikal is supposed to be a mainstream CPU for consumers by the sounds of it.
sepheronx wrote:Elbrus 8C rumored to be postponed a little to have an DDR4 memory controller.
That's amazing news. Feature wise it will make it on par with Skylake.
Right now rumored, but first samples may be DDR3, dunno. But like Elbrus 2C+ being moved to Elbrus 2SM, Elbrus 8C would eventually get it. Apparently they are working or have completed the new chipset too for Elbrus 8C. 8C will be their performance line I imagine.
Do you get information from insiders or do you dig deep on the internet? Any extra info on Baikal's status? Baikal is supposed to be a mainstream CPU for consumers by the sounds of it.
Most of it comes from reading around, mostly on sdelanounas.ru and comments section/discussion board and cnews.ru is for technology. I heard it on sdelanounas.
I am more interested in mikroclet processor. Last I heard of Baikal ARM processor is about month or two ago that it is still being worked on and planned release here this year. It is T-Platforms subsiduary, Baikal electronics who is designing it, but will be manufactured in Taiwan. I think I posted news on it under semiconductor thread in this section of the forums.
sepheronx wrote:Elbrus 8C rumored to be postponed a little to have an DDR4 memory controller.
That's amazing news. Feature wise it will make it on par with Skylake.
Right now rumored, but first samples may be DDR3, dunno. But like Elbrus 2C+ being moved to Elbrus 2SM, Elbrus 8C would eventually get it. Apparently they are working or have completed the new chipset too for Elbrus 8C. 8C will be their performance line I imagine.
Do you get information from insiders or do you dig deep on the internet? Any extra info on Baikal's status? Baikal is supposed to be a mainstream CPU for consumers by the sounds of it.
Most of it comes from reading around, mostly on sdelanounas.ru and comments section/discussion board and cnews.ru is for technology. I heard it on sdelanounas.
I am more interested in mikroclet processor. Last I heard of Baikal ARM processor is about month or two ago that it is still being worked on and planned release here this year. It is T-Platforms subsiduary, Baikal electronics who is designing it, but will be manufactured in Taiwan. I think I posted news on it under semiconductor thread in this section of the forums.
Although, technically MCST is first in semiconductor R&D, dunno how they determined Baikal electronics in it, unless for pure civil tech then maybe.
Interesting thanks. What's so special about the mikroklet CPU?
sepheronx wrote:Elbrus 8C rumored to be postponed a little to have an DDR4 memory controller.
That's amazing news. Feature wise it will make it on par with Skylake.
Right now rumored, but first samples may be DDR3, dunno. But like Elbrus 2C+ being moved to Elbrus 2SM, Elbrus 8C would eventually get it. Apparently they are working or have completed the new chipset too for Elbrus 8C. 8C will be their performance line I imagine.
Do you get information from insiders or do you dig deep on the internet? Any extra info on Baikal's status? Baikal is supposed to be a mainstream CPU for consumers by the sounds of it.
Most of it comes from reading around, mostly on sdelanounas.ru and comments section/discussion board and cnews.ru is for technology. I heard it on sdelanounas.
I am more interested in mikroclet processor. Last I heard of Baikal ARM processor is about month or two ago that it is still being worked on and planned release here this year. It is T-Platforms subsiduary, Baikal electronics who is designing it, but will be manufactured in Taiwan. I think I posted news on it under semiconductor thread in this section of the forums.
Although, technically MCST is first in semiconductor R&D, dunno how they determined Baikal electronics in it, unless for pure civil tech then maybe.
Interesting thanks. What's so special about the mikroklet CPU?
Sorry, I misspelt it, it is Multiclet. I somehow merged Mikron and Multiclet together
Multiclet blog on sdelanounas: http://sdelanounas.ru/blog/MultiClet/
http://www.multiclet.com/
MultiClet is a technical term used to define processor core, processor with multicellular architecture, and multicellular processor based device. For this reason the term was chosen to be the company's name.
Main advantages of multicellular architecture
It enhances performance and at the same time reduces energy consumption There is no need for compliated controlling units such as brunch prediction, out-of-sequence command execution etc., therefore it reduces considerably the size of the chip.
Liveness Fault Tolerance (сontinuous program running in case of hardware environment break-down - cells failure) Dynamic reconfiguration (in the process of task performance 1,2,3 or 4 cells can be involved, while uninvolved cells can perform other tasks) Super security of information Natural virus immunity Cryptographic capabilities Anti-hacking features
Multicellular architecture features
"Natural" implementation of parallelism (without solving the problem of parallelizing) Program execution without recompilation for any quantity of cells Effective settlement of any class of problems (commutation environment doesn't limit intercellular data exchange) Significant reduction of memory accesses during execution of task
The multicellular architecture differs from the von Neumann model by the direct indication of informational connections between operations and consequently any requirement for the ordered arrangement of the operation description in the program is annulled. It is different from the well known non-von Neumann architectures by means of sequential fetching which realizes imperative programming languages as well as by dynamically generated tags, but not instruction addresses of indicating informational connections. Any instruction is executed at the “data readiness” and “its output users’ readiness”.
Started sales of the first computers with the Russian processor
Russian developer of computer technology MCST announced the beginning of reception of orders for computers on the basis of domestic microprocessor "Elbrus-4C". Sales are open only to legal entities, the company said on its website. Personal computer "AWP Elbrus-401" is built on a single Quad-core microprocessor "Elbrus-4C" and is proposed for workplace equipment, information terminals and microservers. It is noted that this computer as its server side modification, suitable for use in enterprises with high demands on information security. In the base rack server "Elbrus-4.4" lies four microprocessor "Elbrus-4C", and one or two controller peripheral interfaces for connecting external devices. The server supports up to 384 gigabytes of RAM, making it suitable for the organization's web servers, database servers, data storage systems, servers, remote desktop and HPC clusters. Both models are running the operating system "Elbrus", developed on the basis of kernel Linux version 2.6.33. Processor "Elbrus-4C", based on 65-nanometer technology, is currently the most productive Russian processor and its speed is comparable with Intel Core i3 and Intel Core i5 processors in the Ivy Bridge architecture. Its development was completed in April 2014, and serial production began in the fall of 2014.
Russia Unveils Homegrown PC Microprocessor Chips Moscow Center of SPARC Technologies begins taking orders for Russian-made computer chips, but one expert warns the technology lags five years behind that of western companies
Moscow Center of SPARC Technologies (MCST) has announced it’s now taking orders for its Russian-made microprocessors from domestic computer and server manufacturers. The chip, called Elbrus-4C, was fully designed and developed in MCST’s Moscow labs. It’s claimed to be the most high-tech processor ever built in Russia, and is comparable with Intel Corp’s Core i3 and Intel Core i5 processors.
Besides the chips, MCST unveiled a new PC, the Elbrus ARM-401 which is powered by the Elbrus-4C chip and runs its own Linux-based Elbrus operating system. MCST said that other operating systems, including Microsoft’s Windows and other Linux distributions, can be installed on the Elbrus ARM-401. Finally, the company has built its own data center server rack, the Elbrus-4.4, which is powered by four Elbrus-4C microprocessors and supports up to 384 GB’s of RAM. MCST said the Elbrus-4.4 is suitable for Web servers, database servers, storage systems, servers, remote desktops and high-performance clusters
However not everyone was convinced Elbrus-4C was up to scratch with its U.S. made competitors. Sergei Viljanen, editor in chief of the Russian-language PCWorld website, told Kommersant the design was inferior to foreign chips.
“Russian processor technology is still about five years behind the west,” Viljanen said. “Intel’s chips come with a 14 nanometer design, whereas the Elbrus is 65 nanometers, which means they have a much higher energy consumption.”
So Russia design a brand new quadcore processor comparable in performance with intermediate level Personal PCs in the west .. this is something never done before by any other nation with 365 Gigabytes ram support.. and the best thing that Moron can say is that oh .. is not better than the thing we use from the west.. ignore it.
Is a 100% spy free processor without US government spy tools ,with a 100% NSA spy free Operating System ,for 100% secure data storage of your company.. Are you kidding? The guy above is a real Imbecile , the HUGE benefits this processor will have for security with an ok performance and a huge memory capabilities 365 GB is like being the second nation to travel to space as america is.. Things can only get better from here.. and i will love to buy one PC of those.. for peace of mind with a safe internet computer..
It people needs windows aplications..microsoft windows can also be installed ,perhaps works in emulation mode? it can also be installed.. so is a plus one . i bet the price performance is very good too.. I will love to see Russia catching Intel and that idiot saying.. oh no.. is not Made by Intel so is not better.
How much i will have liked Russia to jump in the semiconductor race as they did in the space race.. but anyway never is too late .
By the way curious about how much will cost in dollars a PC with Elbrus 4C. in any case this is a major world achievement for Russia and you can bet it will not be mentioned anywhere in western media.. that is on sale.. Russia needs now to advertise it .. you can have the best product in the world.. and if you do not advertise it propertly it will not sell.
here elbrush 4c with PC games..
Is a bit slow but for sure the operator did not configure correctly the graphics settings.. in any case ,there is a lot of potential in Elbrus proccessors ,advertising ,advertising and advertising is the way.. for business security and for personal use for secure internet navigation and personal data.. is priceless. Is like heaven . in just 5 years if Russia continues pushing it will be at the same level in performance of the west and steal a HUGE semiconductor market and finally we can get rid of US monopoly in computers..
"Lags 5 year behind Western analogues". So what? People with broad computer knowledge say AMD lacks behind Intel 2-3 years in IPC performance and energy efficiency with their CPUs. Does that make the Russian chips lag less behind AMD? The most crucial point has been reached: there is now a solid basis for domestic chip design and production on Russian soil. As long as they keep attracting students, specialists and investments this so-called "lag" is a matter of time.
Neutrality wrote:"Lags 5 year behind Western analogues". So what? People with broad computer knowledge say AMD lacks behind Intel 2-3 years in IPC performance and energy efficiency with their CPUs. Does that make the Russian chips lag less behind AMD? The most crucial point has been reached: there is now a solid basis for domestic chip design and production on Russian soil. As long as they keep attracting students, specialists and investments this so-called "lag" is a matter of time.
The big deal also is that this one is entirely Russian. Does it perform? Yes. In case of sanctions this is critical.
Anyone heard about "Aquarius"? Well I just did after googling a few things about Russian CPU market and I stumbled on these guys. Looks like a solid company to me with a broad supply of products.
Also, I just read more on the upcoming Baikal processor. Well it seems it's not entirely Russian. Baikal Electronics and an American company called "Synopsys" teamed up and the latter company's portfolio chip architecture design was chosen as a basis. Good start regardless. TSMC already has shipped working prototypes to the company for evaluation and somewhere in mid-May the mass production model should be printed. If everything goes well, they'll show it at the economic forum in Saint-Petersburg.
Neutrality wrote:Anyone heard about "Aquarius"? Well I just did after googling a few things about Russian CPU market and I stumbled on these guys. Looks like a solid company to me with a broad supply of products.
Also, I just read more on the upcoming Baikal processor. Well it seems it's not entirely Russian. Baikal Electronics and an American company called "Synopsys" teamed up and the latter company's portfolio chip architecture design was chosen as a basis. Good start regardless. TSMC already has shipped working prototypes to the company for evaluation and somewhere in mid-May the mass production model should be printed. If everything goes well, they'll show it at the economic forum in Saint-Petersburg.
There is no point in such mutual American -Russian agreements in semiconductor. the major goal is to design Computers and software that is 100% safe from US Government dirty hands which is the major issue Russia needs to deal.
All American companies can be forced at gun point if needed by its corrupt government and its secrets revealed to them.. So you cannot depend of any american made technology as long they live in a totalitarian fascist state. Because it will be technology that will have US government spies if Russia participate.
This is serious security issue in using american technology in for example Russia Space program and military.. because American technology could come with trojan horse surprises ,that disable the hardware.. and for example makes a space rocket or a system of defense like S-500 stop working if they were fool enough to use american technology in them.. The west could also sabotage a Nuclear Reactor if had American computers . So Russia designing their own computers is critical for their nation survival and modernization.
Russia is under obligation for security of their country to completely design cutting edge technology ,specially computers that are 100% design by them. for its nation security. They already had them.. but now it needs to take things to the average non government based..business industry , and entertainment too.. thats were the big money is.. and a new Internet too.. BRICS and LAtin America with will be an Excellent platform to launch a new Internet..that could cover the entire world. advertised as free of radical extremism and free of NSA spying and ideal for family. Not neccessarily free of Porn because Internet will not work without it..lol but very easy for family to keep it away of children.
it will become a major success. Internet Next.. or Internet 2.. can be called. and will promote Business in emerging economies. like it is Asia and Latin America.
As it stand now American companies have the monopoly on internet content and its Windows based Computers.. specially its Operating System , like Windows OS.. Is specifically made to made it easier for the US government and Private american companies and any of its allies to spy on you.. The place you visit and what say and the information on your PC.
For example there is absolutely NO reason ,for the way it is now .. that a simple visit to an internet website is enough for your COmputer to grab a Virus that open your PC to anyone who wants to visit it and steal your personal information.. Microsoft Windows is designed for US government easy spying on everyone.. and that bullshit needs to stop. Is not only to track "criminals" or "Terrorist" as they will like you to believe , but for stealing of Intellectual property or simply for spying on their competition what they know and do..
Im sure with the proper advertising and presentation ,aimed at security and peace of mind,the Elbrus processors will be a major success. specially in small and big business ,who needs to secure their information.. I really hope Russia continues pushing in the semiconductor and never quits..The world need it.
Last edited by Vann7 on Sat May 09, 2015 12:43 am; edited 1 time in total
Vann7 wrote: There is no point in such mutal American -Russian agreements in semiconductor. the major goal is to design Computers and software that is 100% safe from US Government dirty hands which is the major issue Russia needs to deal.
All American companies can be controlled by its corrupt government and its secrets revealed to them.. So you cannot depend of any american made technology. Because it will be technology that will have US government spies if Russia participate.
This is serious security issue in using american technology in for example Russia Space program and military.. because American technology could come with trojan horse surprises ,that disable the hardware.. and for example makes a space rocket or a system of defense like S-500 stop working if they were fool enough to use american technology in them.. The west could also sabotage a Nuclear Reactor if had American computers .
They already have that covered and it's called Elbrus. As I understand, Elbrus is going to be the basis of the Russian MIC. Meanwhile the Baikal will be produced for less crucial government structures and legal entities.
and how much bits is the new Elbrus processor? Will be nice to see one day 128 bit processors. For very deep graphics and precision. and of course with really good speed.
and how much bits is the new Elbrus processor? Will be nice to see one day 128 bit processors. For very deep graphics and precision. and of course with really good speed.
64bit? Why would you want to see 128 bit processors? 64bit processors can theoretically support 16EB (exabytes) of RAM. That's 1 billion gigabytes or 1 million terabytes. However, you need software to adress this amount of memory and the most recent version of Windows (W8) can "only" support 512GB of RAM.
"For very deep graphics and precision". That's what GPU cards are made for. They are alot faster than CPUs in terms of flops. For example, a GTX980 is capable to produce 4,5Teraflops while the i7 5960X does 384Gigaflops. This is also why you need a good dedicated graphics card to play the latest games without any hiccups. Games are very complex calculations and GPUs are capable of delivering that at much faster rates than a CPU. Brute force attacks (for cracking passwords) is also done by mostly using GPUs for the same reason I previously explained.
A minus of the Elbrus architecture, in its current form implemented on silicon, is that it does not do 64 bit floating point without a penalty. The Intel and AMD server chips and their consumer variants do 64 bit FP without any penalty. The Elbrus flop value drops by over a factor of two every time the FP bit count is doubled.
I can see this having some value since the chip real estate is not packed with 64 bit wide interconnects and this saves power and reduces operating temperature. But it is rather annoying since double precision math is actually necessary in many instances since the single precision math (32 bit) exhibits too much machine error intrinsic to this amount of precision.
MCST should aim at producing a proper Elbrus server chip that has 64 bit FP built in from scratch. Some variant of SSE3 would be especially worthwhile since having 128 bit deep FPUs and vectorization can really pay off in terms of performance.
sepheronx wrote:Vliw isnt cisc process. It is RISK, thus defeats the purpose if sse instruction is added. It emulates x86, be it 32bit or 64.
VLIW has no relevance to my point. VLIW is a short pipeline, complex pipeline stage design that in the case of the Elbrus happens not to vectorize FP math. SSE introduces 128 bit FPUs that can execute two 64 bit or four 32 bit FP instructions at once and this has exactly zero to do with it being used in RISC processors. Before the SSE FPUs there was no capacity to vectorize FP math on the Intel CPUs even though they could issue several instructions in a single clock cycle. The Elbrus cannot execute four 32 FP instructions at once even though it is a VLIW design:
http://mcst.ru/mikroprocessor-elbrus4s
50 gigaflops for 32 bit floating point 25 gigaflops for 64 bit floating point
Clearly there is a penalty for doing double precision floating point math on the Elbrus. That was my point and you threw in some nonsense about VLIW. The Elbrus VLIW does not vectorize FP operations and does not even give you double precision FP for free. The Elbrus VLIW is about how you handle all the other tasks that CPU performs. It can issue 23 operations in one clock cycle, but this does not include four 32 bit FP operations. They need to modify the design to increase FP performance. But that costs a lot of chip real estate and increase heat output.
sepheronx wrote:Vliw isnt cisc process. It is RISK, thus defeats the purpose if sse instruction is added. It emulates x86, be it 32bit or 64.
VLIW has no relevance to my point. VLIW is a short pipeline, complex pipeline stage design that in the case of the Elbrus happens not to vectorize FP math. SSE introduces 128 bit FPUs that can execute two 64 bit or four 32 bit FP instructions at once and this has exactly zero to do with it being used in RISC processors. Before the SSE FPUs there was no capacity to vectorize FP math on the Intel CPUs even though they could issue several instructions in a single clock cycle. The Elbrus cannot execute four 32 FP instructions at once even though it is a VLIW design:
http://mcst.ru/mikroprocessor-elbrus4s
50 gigaflops for 32 bit floating point 25 gigaflops for 64 bit floating point
Clearly there is a penalty for doing double precision floating point math on the Elbrus. That was my point and you threw in some nonsense about VLIW. The Elbrus VLIW does not vectorize FP operations and does not even give you double precision FP for free. The Elbrus VLIW is about how you handle all the other tasks that CPU performs. It can issue 23 operations in one clock cycle, but this does not include four 32 bit FP operations. They need to modify the design to increase FP performance. But that costs a lot of chip real estate and increase heat output.
Intel's i7 5960x (the most powerful Intel CPU available for consumers right now) does 817.5GFLOPS in single precision and 408.6GFLOPS in double precision. I have no idea what "penalties" you're talking about.
sepheronx wrote:Vliw isnt cisc process. It is RISK, thus defeats the purpose if sse instruction is added. It emulates x86, be it 32bit or 64.
VLIW has no relevance to my point. VLIW is a short pipeline, complex pipeline stage design that in the case of the Elbrus happens not to vectorize FP math. SSE introduces 128 bit FPUs that can execute two 64 bit or four 32 bit FP instructions at once and this has exactly zero to do with it being used in RISC processors. Before the SSE FPUs there was no capacity to vectorize FP math on the Intel CPUs even though they could issue several instructions in a single clock cycle. The Elbrus cannot execute four 32 FP instructions at once even though it is a VLIW design:
http://mcst.ru/mikroprocessor-elbrus4s
50 gigaflops for 32 bit floating point 25 gigaflops for 64 bit floating point
Clearly there is a penalty for doing double precision floating point math on the Elbrus. That was my point and you threw in some nonsense about VLIW. The Elbrus VLIW does not vectorize FP operations and does not even give you double precision FP for free. The Elbrus VLIW is about how you handle all the other tasks that CPU performs. It can issue 23 operations in one clock cycle, but this does not include four 32 bit FP operations. They need to modify the design to increase FP performance. But that costs a lot of chip real estate and increase heat output.
OK, I see what you are saying. But Elbrus direct competitor is the Intel Itanium line of processors and I cannot really find much on if it actually does use SSE instructions.
You can say it is nonsense, but VLIW is RISC by nature and RISC stands for reduced instruction set computing, and SSE introduces 70 instructions......... Not really reducing it, now is it? SSE was introduced to compete against AMD's instruction 3DNow! They could introduce it like you said, of course, all depends on what they are going to end up with in the end as a processor. If this is the case, they should have gone to x86 processing development instead of VLIW if they need to add plenty of instructions (since Intel has abandoned the project, but this may be the reason why as they could be the only developer in VLIW processing next to Nvidia and AMD (ATI)).
I am excited to see what they will come up in the future, after 8C. Because at this point, plenty of money should be going through the system at MCST, and they may come up with something even more interesting, unless it will all be simply more cores and higher speeds.
Maybe a second development, move away from their SPARC line and add in another line of something else, like an x86 processor.
I say if they are worried about using licensed tech unlicensed, why would they care? Especially today's sanctions.
Home
Latest images
Register
Log in
kvs
Are you kidding?
I will love to see Russia catching Intel and that idiot saying.. oh no.. is not 
