Thats actually what I thought at first, too. Then I looked it up here. The little box with "Clawhammer" in it also says "w/ x86-64 technology", which is their way of saying "it runs 64-bit." It's just made for 1-2 processor configurations, while the Sledgehammer is for bigger SMPs (like up to 8).
Uh...did you actually read the article? The Sun processors aren't for the Supercomputer - they're to control the I/O. They're the "front-end", so to speak. Everything in the SV2 (except memory) and MTA-2 is completely custom. And the architectures are very VERY novel.
Um, you forgot to mention that in your STREAM results, that the 8-year old Cray T932 was actually the closest contender, at about 1/2 to 2/3 of the SX-5. The SV2 is much more related to the T90 series as far as performance levels than the SV1 (read: budget line). The memory bandwidth is not the only performance measure either - how about scalability? Interconnect latency? Cray has a pretty good record in those arenas as well (see T3E).
As to your coffee table...I love it! A guy I know has an old SuperServer that he converted into an end table. Oh, and in one of the old Cray buildings in WI, they use an old Cray-XMP as a waiting room couch. Thats a nice touch.
Yes, the Alpha Clusters are just that - clusters. There is a lot of creativity going on with their new products, however. The SV2 is a completely new architecture with pretty-much everything being custom designed. It combines the benefits of the vector architecture with the scalability of the huge parallel T3E-style machines. And the MTA-2 is an all-new (commercially) architecture - so new that the first people that will probably purchase them will be researchers trying to figure out how to best use them.
So, yeah, Cray is branching into some lower markets, but they've finally escaped the creative dearth (at least on the high-end) that is SGI.
Re:What's it good for if your friends don't have o
on
Update From Cray World
·
· Score: 2
Sorry, big guy. Unfortunately it doesn't work that way.
The distributed/clustered model is great for some problems - lots of completely independent data processing (like SETI, or Google) works great. You start moving into a realm where you're doing scientific simulations where all the calculations are interdependent, then a Beowulf cluster (even with some good interconnect) hold a scalability candle to, oh, say, a Cray T3E. There's a good reason that the Cray T3E and SV1 won the "Co-Supercomputer product of the year award" this year, as handed out by the people who use them.
Similarly, at the University of Michigan, there is a senior level class where students implement a limited version of the Alpha 21264 in verilog. It is pipelined, and even superscalar.
If a company (like a car company) wants to use arm processors in their mass-produced products, they:
1) won't use FPGAs because they're WAY too expensive
2) will find it much easier (and cheaper) to just buy a full version (like the Intel SA-110) instead of contracting out to fab their own.
In the early 90s, they got too big. When they started out, HPC was a big GROWING market. Around that time, it started to flatten out, and investors aren't real thrilled when you anticipate growth, and it doesn't show. The market for their machines was (and is) still there, it just wasn't growing nearly as quickly.
I think the "new" company has much better focus, and knows what its strengths and weaknesses are. Hopefully, with this new Linux/Alpha clustering, they aren't starting to branch out too far again like they did back then.
Good lineage post - I'm going to have to nit-pick, though...
The Origin (MIPS-based) line was never a Cray product. SGI developed it in conjunction with the DASH project at Stanford.
Also, interestingly enough, Cray #4 is actually very close to Cray #1...through the sale and un-sale to SGI, a lot of the extra stuff was stripped back off. Tera bought the name, yes, but they also brought their employee total from ~50 to ~950.
Hang on there, cowboy. Perhaps you should look at some data before pronouncing the Cray-ons fit for burial.
They've won two straight Supercomputing product-of-the-year awards with their SV1 and T3E lines, they have a couple of very highly anticipated (in the HPC community) product releases coming in the next year or two (the MTA-2 and SV2), and, unlike their ex-parent company (SGI), they're actually profitable.
The "dead for almost a decade" you're thinking of probably is related to the fact that they were sucked into SGI for the last 5 years of the 90s. It's hard to hear anything about "Cray", when nobody calls them "Cray" anymore.
Re:pdp-10 ran UNIX therefore...
on
PDP-10 Revival
·
· Score: 1
Ooops:) My bad for skimming. I hope my prof isn't reading this...
Re:pdp-10 ran UNIX therefore...
on
PDP-10 Revival
·
· Score: 2
Um....That's a nope. Thompson and Ritchie wrote the first version of UNIX on a PDP-11. I just read the paper last week.:)
The PDP-11 is much different - it's a "mini" and was a LOT cheaper than the 10. In fact, several of the PDP-10's used 11's as front ends. (from the development site)
To respond to your 1,2,3: (per request:)
1) Vector machines get very little "attention" in the US from the media - it's old news. People still buy them and people still use them, it's just that all the attention is focused on the new and "cool" stuff like the massive ASCI machines that the gov't decided to splurge on.
2) I find it amusing as well, actually. They (the gov't) do this for two reasons - they don't really "trust" overseas SC vendors to put SC in government sites (i.e. they don't want to call a japanese service guy to fix a SC at CIA HQ or Army labs, etc). Also, they want to have some say in the design of the machines - because they help fund the R&D for Cray, they can be real up front about what kind of machine they need, even before it's designed.
3) I think I pointed this out before - the top500 list is not an accurate measure of performance, by any means. I don't know how the application performance of the Hitachi and the Cray might line up, but the reason the Vector machines don't show up very well is that all of these other shared mem/superclustered-type machines are overrated. Companies are still buying Crays. Why? They're not stupid. (well, not all of them!) I'm guessing Ford bought 5 (I checked the numbers:) because their applications just ran fastest on that machine. I think an SV1 set a record recently for NASTRAN.
No offense, but you stated some whoppers there. Let me try and give you some more information.
1) yes, the GFLOPS per "CPU" is fewer for the SV1. That is because they're made to run in highly-coupled groups of 4 called an "MSP". That's why they prefer to say it's really 7.2 GFLOPS as opposed to 1.8. It's not the same as just having 4 CPU's in parallel - it's really more like 4 CPUs as 1 CPU. Let's also remember that this is Cray's "Budget" line. Did you look at the cost on the Hitachi? Ouch.
2) Where the heck do you get 40TB/sec of memory bandwidth?!? Each node-to-node link in the Hitachi is 1 to 1.2 GB/s each way. See their site.
3) Calling the Top500 list "real evidence rather than marketing numbers" is a big joke. One of the biggest discussions at the conference was about a new benchmark suite being created to rank the top500 list. Why? Because Linpack is really a peak-speed measurement. These huge parallel ASCI machines run them at 30-50% of peak. On real applications, they'd be damn lucky to get 1% of peak, if that. In fact, a Cray T3E still holds the world record for sustained performance on a real application. It's interconnect is that good. I'm sure Hitachi is behind the new benchmarks as well (Cray is)...anything which emphasizes real application speed will make any vector machine look MUCH better.
4) Cray does not "mainly sell Alpha-based machines" these days. They sell them once in a while, but if you're counting by number of systems shipped, they sell a lot more SV1s. I think Ford just bought 2 or 3. I think they've maybe sold 4 or 5 T3Es all year.
5) I don't have any numbers on this, but I really really don't think the Hitachi has a MTTI of over a year.
lastly, did you see who won the "Supercomputing product of the year" award at the conference? The SV1 did. For the second straight year.
Now, I'm not knocking the Hitachi - it's a good machine - I'm just pointing out that vector computing in the US is doing just fine, thank you. And it will be doing WAY better when the SV2 is out in 2002...it's going to be a doozie!
You have to watch out for those pesky write in candidates in Minnesota! Ventura was lucky - he had a party at least - the previous Governor, Arne Carlson, didn't get his party's nomination, so he ran as a write in candidate - and won!
Nope - the compatibles companies simply created their own processor that used the same instruction set and interface, which were freely published (so people could write software). The underlying hardware circuits (which is what this reverse-engineering article is all about) were not copied.
"Reverse Engineering" didn't create PC clones. All of the cloning companies simply created a processor that followed the same (freely published) instruction set and rough timings, so that the software still worked. The underlying hardware is irrelevant as long as the interface is the same.
Slashdot Mirror
Thats actually what I thought at first, too. Then I looked it up here. The little box with "Clawhammer" in it also says "w/ x86-64 technology", which is their way of saying "it runs 64-bit." It's just made for 1-2 processor configurations, while the Sledgehammer is for bigger SMPs (like up to 8).
Actually, they still are. (Press release from before they were sold).
And yes, this (the one in the article) is the same Cray that built all the vector machines. Check out the gallery if you don't believe me.
Uh...did you actually read the article? The Sun processors aren't for the Supercomputer - they're to control the I/O. They're the "front-end", so to speak. Everything in the SV2 (except memory) and MTA-2 is completely custom. And the architectures are very VERY novel.
As to your coffee table...I love it! A guy I know has an old SuperServer that he converted into an end table. Oh, and in one of the old Cray buildings in WI, they use an old Cray-XMP as a waiting room couch. Thats a nice touch.
Oh yeah - the "spite" factor. Is there a company policy now against using "the font"?
So, yeah, Cray is branching into some lower markets, but they've finally escaped the creative dearth (at least on the high-end) that is SGI.
The distributed/clustered model is great for some problems - lots of completely independent data processing (like SETI, or Google) works great. You start moving into a realm where you're doing scientific simulations where all the calculations are interdependent, then a Beowulf cluster (even with some good interconnect) hold a scalability candle to, oh, say, a Cray T3E. There's a good reason that the Cray T3E and SV1 won the "Co-Supercomputer product of the year award" this year, as handed out by the people who use them.
If a company (like a car company) wants to use arm processors in their mass-produced products, they:
1) won't use FPGAs because they're WAY too expensive
2) will find it much easier (and cheaper) to just buy a full version (like the Intel SA-110) instead of contracting out to fab their own.
I think the "new" company has much better focus, and knows what its strengths and weaknesses are. Hopefully, with this new Linux/Alpha clustering, they aren't starting to branch out too far again like they did back then.
The Origin (MIPS-based) line was never a Cray product. SGI developed it in conjunction with the DASH project at Stanford.
Also, interestingly enough, Cray #4 is actually very close to Cray #1...through the sale and un-sale to SGI, a lot of the extra stuff was stripped back off. Tera bought the name, yes, but they also brought their employee total from ~50 to ~950.
They've won two straight Supercomputing product-of-the-year awards with their SV1 and T3E lines, they have a couple of very highly anticipated (in the HPC community) product releases coming in the next year or two (the MTA-2 and SV2), and, unlike their ex-parent company (SGI), they're actually profitable.
The "dead for almost a decade" you're thinking of probably is related to the fact that they were sucked into SGI for the last 5 years of the 90s. It's hard to hear anything about "Cray", when nobody calls them "Cray" anymore.
Ooops :) My bad for skimming. I hope my prof isn't reading this...
The PDP-11 is much different - it's a "mini" and was a LOT cheaper than the 10. In fact, several of the PDP-10's used 11's as front ends. (from the development site)
2) I find it amusing as well, actually. They (the gov't) do this for two reasons - they don't really "trust" overseas SC vendors to put SC in government sites (i.e. they don't want to call a japanese service guy to fix a SC at CIA HQ or Army labs, etc). Also, they want to have some say in the design of the machines - because they help fund the R&D for Cray, they can be real up front about what kind of machine they need, even before it's designed.
3) I think I pointed this out before - the top500 list is not an accurate measure of performance, by any means. I don't know how the application performance of the Hitachi and the Cray might line up, but the reason the Vector machines don't show up very well is that all of these other shared mem/superclustered-type machines are overrated. Companies are still buying Crays. Why? They're not stupid. (well, not all of them!) I'm guessing Ford bought 5 (I checked the numbers :) because their applications just ran fastest on that machine. I think an SV1 set a record recently for NASTRAN.
Only the Cray Vector line would officially fall under "SIMD". Methinks he's confusing instruction types with memory systems.
1) yes, the GFLOPS per "CPU" is fewer for the SV1. That is because they're made to run in highly-coupled groups of 4 called an "MSP". That's why they prefer to say it's really 7.2 GFLOPS as opposed to 1.8. It's not the same as just having 4 CPU's in parallel - it's really more like 4 CPUs as 1 CPU. Let's also remember that this is Cray's "Budget" line. Did you look at the cost on the Hitachi? Ouch.
2) Where the heck do you get 40TB/sec of memory bandwidth?!? Each node-to-node link in the Hitachi is 1 to 1.2 GB/s each way. See their site.
3) Calling the Top500 list "real evidence rather than marketing numbers" is a big joke. One of the biggest discussions at the conference was about a new benchmark suite being created to rank the top500 list. Why? Because Linpack is really a peak-speed measurement. These huge parallel ASCI machines run them at 30-50% of peak. On real applications, they'd be damn lucky to get 1% of peak, if that. In fact, a Cray T3E still holds the world record for sustained performance on a real application. It's interconnect is that good. I'm sure Hitachi is behind the new benchmarks as well (Cray is)...anything which emphasizes real application speed will make any vector machine look MUCH better.
4) Cray does not "mainly sell Alpha-based machines" these days. They sell them once in a while, but if you're counting by number of systems shipped, they sell a lot more SV1s. I think Ford just bought 2 or 3. I think they've maybe sold 4 or 5 T3Es all year.
5) I don't have any numbers on this, but I really really don't think the Hitachi has a MTTI of over a year.
lastly, did you see who won the "Supercomputing product of the year" award at the conference? The SV1 did. For the second straight year.
Now, I'm not knocking the Hitachi - it's a good machine - I'm just pointing out that vector computing in the US is doing just fine, thank you. And it will be doing WAY better when the SV2 is out in 2002...it's going to be a doozie!
No kidding! You think they could spare a few processors to help us out...
You have to watch out for those pesky write in candidates in Minnesota! Ventura was lucky - he had a party at least - the previous Governor, Arne Carlson, didn't get his party's nomination, so he ran as a write in candidate - and won!
hmm...maybe next we'll get "RIAA sues nanotech firm for distrubuting music for free using nanobots"
Ooops, no, I'm wrong. Intel Clone PROCESSORS weren't reverse engineered, Clone PC's were. My bad :P
My bad. I was thinking more about the Intel Processor clones. The confusion has now subsided. I feel dumb :)
Nope - the compatibles companies simply created their own processor that used the same instruction set and interface, which were freely published (so people could write software). The underlying hardware circuits (which is what this reverse-engineering article is all about) were not copied.
"Reverse Engineering" didn't create PC clones. All of the cloning companies simply created a processor that followed the same (freely published) instruction set and rough timings, so that the software still worked. The underlying hardware is irrelevant as long as the interface is the same.