Time to ask for another computer!
-
<boss-mode>Now is the time to dazzle us with your optimization techniques! That is, if you want a new computer again. Ever.</boss-mode>
-- Kein Mitleid Für Die Mehrheit
Joergen Sigvardsson wrote:
Now is the time to dazzle us with your optimization techniques! That is, if you want a new computer again. Ever.
< boss response mode >My test proved the new design is scaleable to an n-processor model of unknown upper limit. Using four threads and gradually increasing data-flow we processed approximately 10 times the amount of data we processed under the previous design. That is effectively 50 times the last scaling model used in 1990's on the SGI/Cray Origin computers. We exceeded our goals by a whopping 2+ multiple through the use of more efficient processes and a redesigned core based entirely for a massively parallel and auto-scaling system. We respond to data as a company does to work, the threading model starts off respectively small and efficient, when work exceeds the ability to handle the load, more workers are hired automatically scaling the system. More than just a thread pool, this system is designed for massive data flow without synchronization blocking allowing free-running threads to "clear the work" faster than the competitor's products by at least 9 times. In short, four processors should have gained us 3.5 multiple in a decent scaling model. Through effcient design we stretched that number 10 times increase in data flow at the full limit of those 4 processors. Where that upper limit is we will have to find out. You could help us find that out as we shatter the data flow records again putting the company as a leader in parallel technology for ... industry.< /boss response mode > -- modified at 19:07 Wednesday 26th September, 2007
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
PIEBALDconsult wrote:
but I don't have The Grand Negus' compiler.
That's okay, his doesn't do threading. and all that atomic stuff just overcomplicates everything. ;P
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
Plus you use all those pesky floats and doubles.
This blanket smells like ham
-
Joergen Sigvardsson wrote:
Now is the time to dazzle us with your optimization techniques! That is, if you want a new computer again. Ever.
< boss response mode >My test proved the new design is scaleable to an n-processor model of unknown upper limit. Using four threads and gradually increasing data-flow we processed approximately 10 times the amount of data we processed under the previous design. That is effectively 50 times the last scaling model used in 1990's on the SGI/Cray Origin computers. We exceeded our goals by a whopping 2+ multiple through the use of more efficient processes and a redesigned core based entirely for a massively parallel and auto-scaling system. We respond to data as a company does to work, the threading model starts off respectively small and efficient, when work exceeds the ability to handle the load, more workers are hired automatically scaling the system. More than just a thread pool, this system is designed for massive data flow without synchronization blocking allowing free-running threads to "clear the work" faster than the competitor's products by at least 9 times. In short, four processors should have gained us 3.5 multiple in a decent scaling model. Through effcient design we stretched that number 10 times increase in data flow at the full limit of those 4 processors. Where that upper limit is we will have to find out. You could help us find that out as we shatter the data flow records again putting the company as a leader in parallel technology for ... industry.< /boss response mode > -- modified at 19:07 Wednesday 26th September, 2007
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
Plus you use all those pesky floats and doubles.
This blanket smells like ham
Andy Brummer wrote:
you use all those pesky floats and doubles.
and half-floats and long double floats, if it floats, I use it. I think that qualifies me as evil incarnate to PEC.
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
Andy Brummer wrote:
Sounds like you need to put in a request for one of these[^]. If you can max out 16 cores without busting out a ridiculous number of threads, you really know your stuff.
Well, I floored it with data, it is simply designed to queue waiting data and workers never halt if there is data on the queue, they clear it. Without blocking to halt it, it basically plows through numbers in massive quantities. You have to feed it a lot and fast to get that far. That is about 20 times the data flow that any one customer has ever sent us, basically equivalent to all the data from all our customers sent simultaneously to the same machine. Fairly unlikely, but I need to double that to meet an outstanding challenge from one customer. Right now I am neck and neck with a 64 node processor for this same problem double my cores and I could shatter the record they took from us, on a machine that costs a small fraction of the costs. :-D
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
I just beat this one up. Yup, got my auto-scaling (based on data, not processors) threading model running well, and knocked out all four cores with a large dataset. That justifies a new computer right? Or rather, I just have to limit my scaling to the appropriate number of cores.... darn....
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
Or rather, I just have to limit my scaling to the appropriate number of cores.... darn.... I was going to say that.
We are a big screwed up dysfunctional psychotic happy family - some more screwed up, others more happy, but everybody's psychotic joint venture definition of CP
My first real C# project | Linkify!|FoldWithUs! | sighist -
El Corazon wrote:
and given a month that it takes to order anything around here, my timing isn't as far off as you might think....
hmmm So your purchase request is going to say "..The new nVidia card coming out next month that I can't put the name of here because everyone who knows is gagged by an NDA. Quantity: However many can be SLIed together. Price: TBD". Getting that approved sounds like it's going to be an 'interesting' experience. :rolleyes:
-- If you view money as inherently evil, I view it as my duty to assist in making you more virtuous.
dan neely wrote:
hmmm So your purchase request is going to say "..The new nVidia card coming out next month that I can't put the name of here because everyone who knows is gagged by an NDA. Quantity: However many can be SLIed together. Price: TBD". Getting that approved sounds like it's going to be an 'interesting' experience.
Now see, I just got my first call in.... customers already drooling and going to set aside FY08 (starting October) money for us. Given time it takes to transfer that kind of stuff, the money will be here when the products arrive on the shelf, we'll order them and still wait 30 days.... Unfortunately, I probably won't be keeping any myself.... :sigh:
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
Or rather, I just have to limit my scaling to the appropriate number of cores.... darn.... I was going to say that.
We are a big screwed up dysfunctional psychotic happy family - some more screwed up, others more happy, but everybody's psychotic joint venture definition of CP
My first real C# project | Linkify!|FoldWithUs! | sighistpeterchen wrote:
I was going to say that.
:laugh: ya caught that did you? :-D I had to find out where the top end was before I capped it. But it made a good joke about getting a new machine since I did order an oct-core (dual quad) for a customer. but I won't even see it, he'll get it and put it together because of location. Just ordered a quad-core for one of my team to replace a failed dual processor and save money. Hoping next year could mean some new hardware. One project in particular could use an nVidia Tesla. Next couple of months will tell. In the mean time, I am ready for FY08 funding with a new design finished. :)
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
PIEBALDconsult wrote:
but I don't have The Grand Negus' compiler.
That's okay, his doesn't do threading. and all that atomic stuff just overcomplicates everything. ;P
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
And nested conditionals.
Visit http://www.readytogiveup.com/[^] and do something special today.
Trollslayer wrote:
And nested conditionals.
Those are just silly after all. Everyone knows RPG II was ahead of its time, and only needed an english front end to be useful! :)
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
I just beat this one up. Yup, got my auto-scaling (based on data, not processors) threading model running well, and knocked out all four cores with a large dataset. That justifies a new computer right? Or rather, I just have to limit my scaling to the appropriate number of cores.... darn....
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
Never mind a new computer - just turn the computers on the LAN into your own private compute cluster, MWAHAHAHA!!!! I did that about 13 or 14 years ago on our VAXcluster - spawned jobs on remote nodes to parallel compute Mandelbrot sets, displaying the results on the local node. Wish I could remember what VMS system services I did it with.
-
Never mind a new computer - just turn the computers on the LAN into your own private compute cluster, MWAHAHAHA!!!! I did that about 13 or 14 years ago on our VAXcluster - spawned jobs on remote nodes to parallel compute Mandelbrot sets, displaying the results on the local node. Wish I could remember what VMS system services I did it with.
Stuart Dootson wrote:
just turn the computers on the LAN into your own private compute cluster, MWAHAHAHA!!!!
considering I have 2 dual/dual machines at my desk, and the software group has on the order of about 25+ cores with the 5 of us, I haven't counted recently, but we have considered it on more than one occasion and pitched it as a possibility before. So far few have been interested because I tend to solve the problems on one computer as the computers grow. Time will tell if we jump back into something that needs grid computation. But if it does, I am prepared. :) (but I would still rather have a Tesla to add to my systems)
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
I just beat this one up. Yup, got my auto-scaling (based on data, not processors) threading model running well, and knocked out all four cores with a large dataset. That justifies a new computer right? Or rather, I just have to limit my scaling to the appropriate number of cores.... darn....
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
Yeppers time to get a new computer, however management got a good deal on on of those AMD tri core machines, you know the AMD Quad Core chip with one dead core. You can rewrite the code right to work in 1,2, 3 or 4 cores.., oh it also has to be Window ME compatatble. RC
-
I just beat this one up. Yup, got my auto-scaling (based on data, not processors) threading model running well, and knocked out all four cores with a large dataset. That justifies a new computer right? Or rather, I just have to limit my scaling to the appropriate number of cores.... darn....
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
Heck, I've got my IT people trained so well that they just give me additional new computers before I even think to ask now. Of course my cores are spread out and not as handy as your current 4 core machine.
-
Stuart Dootson wrote:
just turn the computers on the LAN into your own private compute cluster, MWAHAHAHA!!!!
considering I have 2 dual/dual machines at my desk, and the software group has on the order of about 25+ cores with the 5 of us, I haven't counted recently, but we have considered it on more than one occasion and pitched it as a possibility before. So far few have been interested because I tend to solve the problems on one computer as the computers grow. Time will tell if we jump back into something that needs grid computation. But if it does, I am prepared. :) (but I would still rather have a Tesla to add to my systems)
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
El Corazon wrote:
(but I would still rather have a Tesla to add to my systems)
What the heck is a Tesla? Google didn't give me any good responses.
-
El Corazon wrote:
(but I would still rather have a Tesla to add to my systems)
What the heck is a Tesla? Google didn't give me any good responses.
firegryphon wrote:
What the heck is a Tesla?
mmmmmmmmmmmmmm....[^] aaaaaaaaahhhhhhhhhh![^]
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
Andy Brummer wrote:
you use all those pesky floats and doubles.
and half-floats and long double floats, if it floats, I use it. I think that qualifies me as evil incarnate to PEC.
_________________________ Asu no koto o ieba, tenjo de nezumi ga warau. Talk about things of tomorrow and the mice in the ceiling laugh. (Japanese Proverb)
-
Never mind a new computer - just turn the computers on the LAN into your own private compute cluster, MWAHAHAHA!!!! I did that about 13 or 14 years ago on our VAXcluster - spawned jobs on remote nodes to parallel compute Mandelbrot sets, displaying the results on the local node. Wish I could remember what VMS system services I did it with.
Yar, VMS! I recently bought a used DS10L to run OpenVMS. It's just one unit from one of these beauties: http://h18002.www1.hp.com/alphaserver/download/ds10l_67_ts.pdf[^] :cool:
-
Yar, VMS! I recently bought a used DS10L to run OpenVMS. It's just one unit from one of these beauties: http://h18002.www1.hp.com/alphaserver/download/ds10l_67_ts.pdf[^] :cool:
The quickest VMS machine I ever used was one of these[^] - not entirely quick :-) We had (still have - I don't use them, though) 10-20 VAXStations, of varying performance, and supported up to 8 users on each workstation using X11 (i.e. graphical) terminals. And we still use them for various applications (e.g. a 680x0 Ada cross-compiler, used on legacy projects) that haven't been /can't be migrated to Windows.
-
The quickest VMS machine I ever used was one of these[^] - not entirely quick :-) We had (still have - I don't use them, though) 10-20 VAXStations, of varying performance, and supported up to 8 users on each workstation using X11 (i.e. graphical) terminals. And we still use them for various applications (e.g. a 680x0 Ada cross-compiler, used on legacy projects) that haven't been /can't be migrated to Windows.
Stuart Dootson wrote:
not entirely quick
Well, it should do at least 32 ft/sec/sec, right? :-D