AMD nearly doubles Top500 supercomputer hardware share

In a sign of how meteoric AMD's resurgence in high performance computing has become, the latest list of the world's 500 fastest publicly known supercomputers shows the chip designer has become a darling among organizations deploying x86-based HPC clusters. The most eye-catching bit of AMD news among the supercomputing set is …

[bazza]

AMD have been doing really well in recent years, and kept it going too. OK, I know Intel have properly screwed up, but AMD have had to get things consistently right to exploit that weakness. One can't really ask more than that.

[Randy Hudson]

x86?

Are there a lot of 32-bit supercomputers these days??

[UBfusion]

Re: x86?

x86 in this context is a shorthand for the Intel architecture (as opposed to AMD's or other CPU manufacturers'), the more technically accurate term would of course be x86-64

[CFtheNonPartisan]

Linpack is techno-porn on its best day. Even Jack Dongarra was and might still be fascinated by the attention it has attracted over decades. He was able to make a side career of it. Last I met Jack his reply was, to paraphrase, 'At least the computer has to stay up long enough to run it' as a defense of its usefulness. That was still in the times when uptime could be in hours or a day or two, not with modern resilient hardware.

[StargateSg7]

Linpack and its derivative ALL suck the big one when it comes to supercomputing performance tests!

What you NEED is a suite of tests that HAMMER the 1st, 2nd, 3rd level caches, ALL of the internal CPU/GPU/DSP registers, all of local system DRAM and shared global DRAM plus any local SSD-based virtual memory, all off-chip I/O buffers and all the network I/O channels to any local crossbars or local routers and switches and network-attached SSD disk arrays ALL AT THE SAME TIME AT MAX bandwidth!

(p.s. we use multi-beam Dense Wave Multiplexing over single-mode glass fibre for ALL off-chips communications to attached devices and networks)

THAT type of test is a true metric of overall system performance!

So when I say OUR under-the-radar all-Canadian Aerospace company's custom in-house designed and built "Haida Gwaii" supercomputer system in Vancouver, British Columbia, Canada does 119 ExaFLOPS sustained at 128-bits wide, it means EVERYTHING IS BEING HAMMERED ALL AT ONCE ALL THE TIME for 192 hours straight (8 days) until the control software and data overload finally buckles and gets brought down to it's knees! THAT is a TRUE TEST of supercomputer performance!

Soooooooo, when I say our "ThunderHorse" and "DragonSlayer" ultracomputers located underground in a far remote mountain region of Northern British Columbia does over a YottaFLOP (1,000,000,000 PetaFLOPS or One Billion PetaFLOPS!) in that same 8 day period at 128-bits wide, then it means we have the TOP THREE WORLD'S FASTEST SUPERCOMPUTERS of any kind anywhere in the world bar none !!!!

We Win !!!!

We're Number One! Yay Canada!

V

[HammerOn1024]

Just Remember...

that this is a list of declared computers.

There are faster out there.

[StargateSg7]

Re: Just Remember...

YES there are! Like ours!

The Three World's Fastest Supercomputers of any kind are "ThunderHorse" and "DragonSlayer" at One YottaFLOP each at 128-bits wide as of last weekend's latest upgrades plus "Haida Gwaii" at 119 ExaFLOPS at 128-bits wide.

WE WIN!

V