This article is rated Start-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects:

Computing Low‑importance This article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.ComputingWikipedia:WikiProject ComputingTemplate:WikiProject ComputingComputing Low This article has been rated as Low-importance on the project's importance scale. This article is supported by Computer hardware task force (assessed as Mid-importance).

Look: I don't know all of whom has contributed to this well meaning page, but it has some just completely wrong impressions. I've stood next to this EPFL machine in the photos (not the best representative photos) as well as 1 of the 2 Cray-2s the NSA purchased. The X-MP wasn't a better crypto machine. The NSA uses whatever it can get, and I know they were happy with their 2 2s for that time (as well as they could be [it didn't run their own in-house OS, but it was one they were familiar with]). You only had to be at LANL to miss out on big memories. They had to wait for a big Y-MP years later.

The comment about the X1 is strange. I have no idea what that person can be thinking, because I have limited familiarity with the X1. I am familiar with Cray 3 and Cray 4 architecture design differences and the agony CCC went through. I do know what I would ask to disambiguate that person's knowledge about the X1.

That coment should either be removed/retracted or refined/elaborated.

The article vaguely mentions that Cray-2 machines "found their way" to several places, but that section should state some sales data and revenue data, especially if it can be compared to Cray-1 and Cray XMP sales. Tempshill 16:46, 26 October 2007 (UTC)[reply]

Just under 30 multiprocessors were sold in 4-, 2-, and a single 8-processor with corresponding memory configurations (256 MW, 512 MW, 128 MW). This is about ball-park, if maybe arguably low, for supercomputers fewer than 100 Cray-1s were made and ditto most of Cray's other machines like the 7600. Don't expect to ever see a public accounting of every machine; don't trust lists, because client information like that is highly sensitive. 143.232.210.46 (talk) 23:17, 5 September 2008 (UTC)[reply]

Reading this article I can't figure out how fast the Cray-2 was. While parallel and single-CPU machines can't be easily compared, some benchmark or number, like floating point operations or working memory ought to be available.EverGreg (talk) 14:37, 31 March 2008 (UTC)[reply]

Thanks to the one adding the GFLOPS benchmark. It's tempting to draw comparision in this article with modern PCs, which according to the FLOPS article can muster up to 51 GFLOPS in their quad core processors or 576 GFLOPS of specialized processing power in the graphics cards, but comparing machines of different architectures is a minefield :-) EverGreg (talk) 12:33, 16 July 2008 (UTC)[reply]

Just from memory, a vector scale ran at 100Mflops (1.6 Gbyte/sec) on the C-2 and 30Mflops on the C-1. A more complicated DAXPY ran at 289Mflops (6.9 Gbyte/sec) on the YMP. Chen's baby totally killed off the C-2 in price, performance, and sales.220.240.225.135 (talk) 22:50, 8 September 2016 (UTC)[reply]

This relates the Speed? above. You will never really have a full appreciation how great the Cray 2 was for its time without having run real supercomputer codes on any real supercomputer. At the time, if you had a Cray (1 or X-MP), and you had a program which required 9 MW of memory, you had to write all kinds of code to manage overlays, time I/O, and all kinds of stuff. Your dimensionality of your data structures, and hence your science, had to remain low (1-2 dimensions or small orders of 3-D). And that I/O, still faster than all the other mainframes of the time, slows your program down. Cray never believed in virtual memory addressing unlike the CDC 205. You might think that stupid now, but you likely never founded a company like Cray Research. You can't begin to appreciate how nice it was to write DIMENSION X(90000000) which, BTW, you could not do on the ETA-10 which was limited on any single dimension to indexes of 65K elements in FTN200.

The Cyber-205 was an absolute *DOG* re performance. I prefered to run overnight on a big VAX with LPS memory. Just one VM page load on a 205 and your job was stuck in VM Purgatory forever. The YMP did have a thing called an SSD for effective VMing at near vector speeds. It was a 500MB of LPS memory, but looked like an HDD to the programmer. The programmer had to do his own mem management. The SSD was a very effective memory increasing device. Scientific projects that had been developed for earlier machines often already included such code to utilize the best HDDs of the time, which performed little better than 5MB/s. What constitutes a Super is the same now as it was 40yrs ago. A Super is not a general purpose machine. It is specialized in some way, vector or parallel, and runs such jobs at much higher speeds. CDC had adapted general purpose machines to vector systems, much like the IBM 3090 "Sierra". The inclusion of VMM was I'd suspect just a sales gimmick. Time is the destroyer of worlds, of course. A Pentium of 2006 vintage gave equal performance on vector code as a one-CPU YMP.220.240.225.135 (talk) 23:42, 8 September 2016 (UTC)[reply]

I commented this statement on the page, since it seems false : One could use a massively parallel supercomputer to find the minimal Hamiltonian cycle of a given (Hamiltonian) graph (Travelling Salesman Problem) by assigning to each node a set of cycles to construct and try. However, branch-and-bound is much more efficient (in time), yet uses more memory. And -as far as I know- very large instances of the Travelling Salesman belong to the realm of supercomputing. —Preceding unsigned comment added by 82.249.246.194 (talk) 17:50, 11 August 2010 (UTC)[reply]

"spiritual descendant"? I doubt that. The person who wrote that must have been associated with MIPS Computer because Seymour didn't like non-uniform memory. I can image who wrote that. Enough said. 66.122.34.11 (talk) 17:03, 6 August 2011 (UTC)[reply]

The Cray-2 is NOT a spiritual descendant of the Cray X-MP. Indeed, Steve Chen (the X-MP project manager and an astute corporate politician) was worried about embarrassing Seymour and kept many design and performance aspects secret until the machine went into production. The X-MP team was created as a hedge against anything happening to Seymour. Cray, the company, had grown large enough that it was worried about losing its position as the premier super-computer company if it should lose its only designer. BJ Creighton (talk) 18:02, 14 May 2024 (UTC)[reply]

¿Cuánto cuesta el Cray-2? — Preceding unsigned comment added by 176.7.49.68 (talk) 14:54, 20 June 2016 (UTC)[reply]

My university was quoted $3.2M AUD for a 1-CPU YMP in 1991. The C2s and XMPs were long gone by then. Cray were pretty flexible on price, and you could essentially short lease one by just by paying the fairly expensive maintenance bill. At this time Cray had lost a few vector sales to NEC and Fujitsu. The late 80s--early 90s NEC machines were the fastest and the most expensive by a large margin.220.240.225.135 (talk) 23:00, 8 September 2016 (UTC)[reply]

2s were about $2xM a piece depending on the number of DD-29 and later DD-49 drives (none of these cheap) you got. Basically, if you have to ask, you can't afford. 2603:3024:182B:2240:C97C:C1F2:8984:3A4B (talk) 18:51, 2 February 2018 (UTC)[reply]

References have been removed. Important sections on software have been removed, to make the machine sound bad. That the article is so hardware slanted the important historic points and differences are obscured. 2603:3024:182B:2240:C97C:C1F2:8984:3A4B (talk) 18:47, 2 February 2018 (UTC)[reply]