I had a few classes in Kane Hall at the UW, seats 720 people. The nice thing is you can sit in the back and tell jokes to your friends or sit in front and learn depending on your mood.
"Write a function to sum all the numbers from 0 to 100"
Don't be so hard on them, I don't know how to do that either due to the fact that the set of numbers from 0 to 100 is infinite. However, if you meant integers, here it is: x=(100*100)/2
The System38 and follow on machine AS400 have had capability based security since 1978 (everything is accessed through 128 bit pointers which contain a 64 bit capability/security key).
Not sure when KeyKOS patented it but the System 38 (introduced in 1978) and it's successor the AS400 use capability based security (every 128 bit pointer has a 64 bit security key embedded).
Was your original comment a joke? If it was and I had realized it, I would have laughed. I see lots of dogmatic statements like that where the person is serious.
The indian thing was a commercial when I was growing up, usually played during sat morn cartoons.
Its only idiots and newbies who use the buttons anyway. Experienced users are already using the keyboard
Sometimes I see comments like this and I'm just stunned. I just want to cry, like that indian in that commercial looking at the litter on the side of the road.
I looked at all of my options about 6 months ago and if I could have bought a 16 core machine without paying for server licenses/hardware and/or moving to linux (my app needs to run under windows right now), then I probably would have done that, but for the price, the gtx280 is working well.
You don't seem to understand the Cell's architecture at all.
The cell has 7 SPEs, but one is reserved for the Hypervisor.
The cell has 8 spe's not 7, the reservation for the hypervisor has nothing to do with "cell's architecture", that has to do with the implementation of software on the PS3 by Sony. Although that is a valid point regarding the number of available SPE's.
Second, each SPE can perform two 128-bit vector operations per-cycle, so long as you do not use double-precision. That's EIGHT 32-bit float operations per SPE, per cycle, excluding the overhead of store operations
You will notice that I said "FP" operations and you will also notice that the FP ALU is on the even pipeline, not both the even and the odd pipeline, that means you can perform 4 single precision FP's per cycle.
Here are a couple tips:
1) In the future, when you correct someone, make sure you are right first
2) Helpful corrections can be good, statements like "don't understand the architecture at all" aren't really valuable
By "silliness" do you mean the fact that my first pass (simple and unoptimized) at running my calcs on GPU gave me a 10x speed up (including memory xfer)? I can get through thousands of generations of my simulation in hours instead of days.
If intel brings out their 80 core proc I will be one of the first to compare to gpu to see which I should continue to use, until then I will use my gpu for real performance gains.
For me it was the opposite, I started with the cell but because the SIMD requires the operands to be adjacent to really gain something (I could have loaded from multiple locations and then moved them all into the 128 bit reg but by then gains would be lost), I switched to cuda and it's working although I do have to jump through hoops on the conditionals to make sure the threads are staying in sync with their instructions.
I don't think anyone who actually works with CUDA refers to individual thread processors as "GPU processors."
Well I actually work with CUDA and I just used that term, so that makes at least 1 person.
The term "GPU processor" was merely a shorthand method of stating that the number 163,200 related to circuitry that performs calculations but without as much flexibility as a core on a traditional CPU. They do work, but groups of them share the same instruction. The term "core" would have seemed inaccurate also, maybe I should have said "streaming processor cores"??
On reading the article, the box has 30 thousand cores, of much the vast majority are AMD Opterons in Sun boxes. No mention of how/in what you'd program this to actually put the GPUs to good use
You may want to read the article again, if not here's a recap:
655 Sun Boxes each with 16 AMD cores=10,480 CPU cores
680 Tesla Cards each with 240 processors=163,2000 GPU processors
As for how to use the GPU's, I use my GTX280 (almost same thing as Tesla) to crunch through lots of numeric calculations in parallel. I'm sure these guys are doing the same thing as that is the strength of the GPU. NVIDIA has made it easier to access the processing power of the GPU with CUDA. You create a program in C that gets loaded on the GPU and when you launch it you can tell it how many copies to run at one time, each one typically operates on a different portion of the data. Because you can launch more threads than there are processors, the GPU can be reading data in from global vid mem while other threads are performing calculations.
I can tell about it's use in the ERP world:
Prior to ERP we had MRP (Manufacturing Resource Planning) software which started out in the Inventory, Sales Orders, Purchase Orders, Work Orders, etc. part of distribution/manufacturing companies. At the same time there was financial and HR software coming from the accounting side of the business. Everyone (software companies) was scrambling to have complete suites of software that covered every department in the enterprise by building missing modules or buying other software companies.
In typical fashion, prior to achieving the goal of having complete sets of software, all of the large vendors changed the name of their software to Enterprise Resource Planning, and voila, now they were selling Enterprise software, not just manufacturing or accounting software.
xTuple instead of Quickbooks (great enterprise-class accounting/sales/CRM/inventory software that can truly rival the "polished quality" of Quickbooks with pretty much the same features)
While xTuple looks interesting, I wouldn't classify it as "enterprise-class", more like entry level/small business (Quickbooks isn't enterprise either, even if they sell something they call "enterprise").
I don't disagree. It seems that we can create something mathematically equivalent without jumping through the exact same hoops as nature.
Regarding your point about text without context: I completely agree, what we write and speak is really a translation/communication of an inner model based on interaction with our environment. Programming in a bunch of text with rules won't cut it.
NVIDIA has done a good job of making the processing power accessible to programmers that are not GPU coding experts. In addition, they have made hardware changes to better support the type of scientific computation being done on these devices.
So, while in theory you could put together some Radeon's, work with their API and achieve the same thing, NVIDIA has significantly reduced the level of effort to make it happen.
Slashdot Mirror
I had a few classes in Kane Hall at the UW, seats 720 people. The nice thing is you can sit in the back and tell jokes to your friends or sit in front and learn depending on your mood.
This thing some people call a "chair", worst invention ever. Now people sit around instead of getting stuff done.
I liked the 6809 (16 bit operations, multiply, good addressing). My friends using the 6502 had to jump through a lot of hoops.
I see other posters have n*(n+1)/2. Looks like my (100*100)/2 has a bug (how fitting that I posted it in this thread about bugs).
Don't be so hard on them, I don't know how to do that either due to the fact that the set of numbers from 0 to 100 is infinite. However, if you meant integers, here it is: x=(100*100)/2
The System38 and follow on machine AS400 have had capability based security since 1978 (everything is accessed through 128 bit pointers which contain a 64 bit capability/security key).
The System38 and follow on machine as400 both use capability based security (every 128 bit pointer contains a 64 bit security/capability key).
S/38 was introduced in 1978
Not sure when KeyKOS patented it but the System 38 (introduced in 1978) and it's successor the AS400 use capability based security (every 128 bit pointer has a 64 bit security key embedded).
Was your original comment a joke? If it was and I had realized it, I would have laughed. I see lots of dogmatic statements like that where the person is serious.
The indian thing was a commercial when I was growing up, usually played during sat morn cartoons.
Sometimes I see comments like this and I'm just stunned. I just want to cry, like that indian in that commercial looking at the litter on the side of the road.
I saw this on TV also, but the thing was attached to a silver car that could travel back and forth through time
How about letting us pump our own gas first, then work on this high-tech stuff.
We added 4 more cores to perform this "thinking" about which core the process should run on, we should be able to get back that 10% we lost, right?
I looked at all of my options about 6 months ago and if I could have bought a 16 core machine without paying for server licenses/hardware and/or moving to linux (my app needs to run under windows right now), then I probably would have done that, but for the price, the gtx280 is working well.
The cell has 8 spe's not 7, the reservation for the hypervisor has nothing to do with "cell's architecture", that has to do with the implementation of software on the PS3 by Sony. Although that is a valid point regarding the number of available SPE's.
You will notice that I said "FP" operations and you will also notice that the FP ALU is on the even pipeline, not both the even and the odd pipeline, that means you can perform 4 single precision FP's per cycle.
Here are a couple tips:
1) In the future, when you correct someone, make sure you are right first
2) Helpful corrections can be good, statements like "don't understand the architecture at all" aren't really valuable
By "silliness" do you mean the fact that my first pass (simple and unoptimized) at running my calcs on GPU gave me a 10x speed up (including memory xfer)? I can get through thousands of generations of my simulation in hours instead of days.
If intel brings out their 80 core proc I will be one of the first to compare to gpu to see which I should continue to use, until then I will use my gpu for real performance gains.
For me it was the opposite, I started with the cell but because the SIMD requires the operands to be adjacent to really gain something (I could have loaded from multiple locations and then moved them all into the 128 bit reg but by then gains would be lost), I switched to cuda and it's working although I do have to jump through hoops on the conditionals to make sure the threads are staying in sync with their instructions.
This is why:
8 PS3's = 8 cells
8 cells X 7 available SPE's per cell = 56 SPE's
56 SPE's X 4 simultaneous FP calcs = 224 FP calcs per cycle
You would need to get quite a few of those x86 dual core kits to match that performance
Well I actually work with CUDA and I just used that term, so that makes at least 1 person.
The term "GPU processor" was merely a shorthand method of stating that the number 163,200 related to circuitry that performs calculations but without as much flexibility as a core on a traditional CPU. They do work, but groups of them share the same instruction. The term "core" would have seemed inaccurate also, maybe I should have said "streaming processor cores"??
You may want to read the article again, if not here's a recap:
655 Sun Boxes each with 16 AMD cores=10,480 CPU cores
680 Tesla Cards each with 240 processors=163,2000 GPU processors
As for how to use the GPU's, I use my GTX280 (almost same thing as Tesla) to crunch through lots of numeric calculations in parallel. I'm sure these guys are doing the same thing as that is the strength of the GPU. NVIDIA has made it easier to access the processing power of the GPU with CUDA. You create a program in C that gets loaded on the GPU and when you launch it you can tell it how many copies to run at one time, each one typically operates on a different portion of the data. Because you can launch more threads than there are processors, the GPU can be reading data in from global vid mem while other threads are performing calculations.
I can tell about it's use in the ERP world:
Prior to ERP we had MRP (Manufacturing Resource Planning) software which started out in the Inventory, Sales Orders, Purchase Orders, Work Orders, etc. part of distribution/manufacturing companies. At the same time there was financial and HR software coming from the accounting side of the business. Everyone (software companies) was scrambling to have complete suites of software that covered every department in the enterprise by building missing modules or buying other software companies.
In typical fashion, prior to achieving the goal of having complete sets of software, all of the large vendors changed the name of their software to Enterprise Resource Planning, and voila, now they were selling Enterprise software, not just manufacturing or accounting software.
While xTuple looks interesting, I wouldn't classify it as "enterprise-class", more like entry level/small business (Quickbooks isn't enterprise either, even if they sell something they call "enterprise").
Did any parts of sandwich drop onto floor?
If yes, have fewer than 5 seconds elapsed?
If yes, pick up and continue with procedure...
I don't disagree. It seems that we can create something mathematically equivalent without jumping through the exact same hoops as nature.
Regarding your point about text without context: I completely agree, what we write and speak is really a translation/communication of an inner model based on interaction with our environment. Programming in a bunch of text with rules won't cut it.
NVIDIA has done a good job of making the processing power accessible to programmers that are not GPU coding experts. In addition, they have made hardware changes to better support the type of scientific computation being done on these devices.
So, while in theory you could put together some Radeon's, work with their API and achieve the same thing, NVIDIA has significantly reduced the level of effort to make it happen.