DailyTech - Microsoft and Intel Join Forces to Spur Multicore Research

Submit News

IT Microsoft and Intel Join Forces to Spur Multicore Research
Gabriel Ikram (Blog) - March 20, 2008 4:02 PM

E-mail

del.icio.us

29 comment(s) - last by mcmilljb.. on Mar 21 at 5:13 PM

Microsoft and Intel provide joint grants to optimize multicore software

Microsoft and Intel joined together to build two research centers focused on parallel computing and multicore programming research. The companies have agreed to offer a combined $20 million to two universities to fund the centers.

Microsoft and Intel plant to establish Universal Parallel Computing Research Centers at two universities, UC-Berkely and the University of Illinois at Urbana-Champaign. Both companies have agreed to invest the grant over a period of five years.

"This is a once-in-a-career opportunity to recast the foundations of information technology and influence the entire IT industry for decades to come," said David Patterson, UC-Berkeley professor of computer sciences, in an announcement. "We are excited and proud to be a part of this ambitious effort."

The centers will promote industry/university collaboration in solving problems that face the IT industry. The centers will be focused on parallel computing and multicore programming research and will help make new software technologies available to mainstream users. According to Microsoft and Intel, software developed by the centers will be made available to the vast technology community for further refinement.

According to Microsoft and Intel, the shift to multicore processors has made it imperative to further refine our current methods of programming as well as creating new methods. Software optimization is necessary in order to take advantage of the multiple cores found on modern CPUs.

Microsoft and Intel won't be the only ones providing for the research centers. UC-Berkely researchers are also applying for a UC Discovery Grant, which matches industry grants through state and university funding. The University of Illinois committed another $8 million to the project.

Microsoft says the two universities were chosen out of a pool of 25 high-tech and leading computer science universities.

Comments

Threshold

Username
Password
remember me

This article is over a month old, voting and posting comments is disabled

Linux already did it.

By greylica on 3/20/2008 4:26:20 PM , Rating: 5

They have 5 years to study linux...

RE: Linux already did it.

By nerdboy on 3/20/2008 4:56:39 PM , Rating: 5

Simpsons Did it!!!!!

Parent

RE: Linux already did it.

By Mitch101 on 3/20/2008 5:46:41 PM , Rating: 2

I guess someone didn't see that episode of south park. I thought it was funny.

Parent

RE: Linux already did it.

By TomZ on 3/20/2008 5:47:59 PM , Rating: 2

quote:
They have 5 years to study linux...

That's an idiotic comment. Every modern OS does, and has for years, supported multi-core and multi-threaded. But the research problem to study is how to design application software that scales well as the number of cores increases. And by that I don't mean 1->2 cores; I mean 4, 8, 16, 32, etc. That's the challenge for the future.

Parent

RE: Linux already did it.

By Tsuwamono on 3/20/2008 7:15:21 PM , Rating: 3

Again.. already done. Industrial computers have been running multithreaded aps for years. We build boards at work with 32 processors.. not 32 cores.. 32 processors. Some boards are dual core processors aswell and they all run multi-threaded applications. Of course they don't run on Windows but again. already been done.

I am glad they are starting to work on the problem for PCs though. Quite annoying being at work and seeing 64 cores working together to solve something then i come home and the only use my dual core has is playing music while i play video games lol.

Parent

RE: Linux already did it.

By greylica on 3/20/2008 8:28:13 PM , Rating: 3

Some calls me idiot, but all of my comments make sense, even with my pretty bad english.
Well, when we think multicores or multithreaded process, we are thinking in an overhead to adress with many (above 8)
Risc X Cisc Isa proved that the overhead only increases when we use full cores of Cisc. It´s a Kernel Issue solved By another ISA, like Itanium ISA or T1000, or Cell, for example.
Like yellow dog Linux on Cell PS3, the problem, either treating Cisc ISA like Risc ISA, is already solved, and Linux kernel let you choose between preemptive or superescalar thread processing, or both of them.
There is no faster way to control the overhead with our CISC ISAs than Linux development already did. This will only change with another ISA.

As I stated, Linux solved the problem.

But may be they are trying to reinvent the wheel, or ask for Intel to make an exclusive ISA, in order to make it closed source and try to say someday that their kernel or apps are more fast, but then they will put a bunch of DRM lines (the overhead...) in everything and Windows (plus pack ultimate 10) is hungry again for 512 cores.

Parent

RE: Linux already did it.

By mcmilljb on 3/21/08, Rating: -1

RE: Linux already did it.

By greylica on 3/21/2008 9:55:15 AM , Rating: 3

Linux have it´s own development tools. The most used is C++ and now Python. Java makes it´s party too.
But in the core codes, C++ already have it´s own tools to spread process into threads or cores or whatever.
The fact:
If the ISA is open standard, we can compete in equal positions. Microsoft investing with Intel sounds to us as a manner to create a closed trick into the processor, that make a kind of obscured advantage.
Only time will tell, but the C++ compiler for Linux already solved the majority of problems of thread distribution.
Linux kernel and the companion compilers already runs into super multi processor environment.

Parent

RE: Linux already did it.

By Steve Guilliot on 3/21/2008 3:26:22 PM , Rating: 4

One of two things is true here:

1. Leading researchers at MS, Intel, and 25 universities don't understand the state of multicore support in application development. These same legions of experts are also unfamiliar with Linux. As a result, they are trying to invent something that is already present in Linux. Praise be to Linus.

or

2. You don't understand the areas of multicore support that still need development. Thus you don't understand the imperative of the multicore programming research project.

Anyone taking bets?

Parent

RE: Linux already did it.

By mcmilljb on 3/21/2008 5:11:24 PM , Rating: 2

Linux doesn't have ITS OWN development tools. The most common is the GCC(GNU Compiler Collection), and it's not owned by Linux. You can get it for any operating system, and it was created in 1987(well before Linux was created). Plus Linux is written in GNU's C not C++. The Linux kernel is programmed to do handle threads just like any other operating system. Like anything else, it needs to keep being improved. Microsoft won't be the only one to benefit from the research. I am sure any new algorithms discovered will be shared with the open source community.
The ISA standard you speak of (x86) is owned by Intel. They created it back in 1978. Granted it's considered "open", Intel can break it's pact with AMD, and everyone would follow Intel. If AMD was leading the processor designs instead of Intel, Microsoft would have teamed up with them. The advantage is Microsoft has the inside track on x86 designs. It's not super multi processor environment, it's symmetric multiprocessing environment. You really don't know much about Linux.

Parent

Great investment

By Highbuzz on 3/20/2008 4:19:51 PM , Rating: 4

As the years pass and pass, more CPUs with even more cores will come into the market. I remember when I bought my dual-cored AMD CPU and everyone said "Yea, but it won't take advantage of ____ program!" and it turned out true. Now some applications do, but some still lack that ability... Now more and more CPUs with more and more cores are coming out and will come out... Quad-Core is definitely here and more will come. The IT industry will needed to whip itself into shape and learn to optimize for multi-cored CPUs. This is a great first step, props to Intel and MS for heeding the call.

RE: Great investment

By Xodus Maximus on 3/20/2008 4:54:12 PM , Rating: 2

Yeah but instead of pumping money into unknowns, I wish they would improve integration and implementations of OpenMP, but that is not likely to happen with MS or Intel anytime soon.

Parent

RE: Great investment

By amanojaku on 3/20/2008 5:00:49 PM , Rating: 3

Something to consider: performance gains from multiple cores aren't always related to multi-threaded apps. Simply spreading the processes over more cores lowers the amount of context switching per core. Context switching reduction would improve performance significantly. The only downside is the increase in interprocess communication, but that would be mitigated by cores sharing dies.

And lastly, core speed still matters. Not every operation can be split into simultaneous tasks (i.e. [2+3]*5)

Parent

RE: Great investment

By MandrakeQ on 3/20/2008 5:04:16 PM , Rating: 2

Multi-core is useful for desktop apps up to a certain point, but afterwards parallelism is best extracted by machines that are inherently parallel like vector processors and gpus. Once we start getting SOC multi-purpose processors with lots of application specific accelerators covering physics, graphics, AI, DSP, pretty much anything we can think of, that's when we will start seeing some awesome performance.

Of course, servers are another ball game, and the more independent concurrent threads can be handled the better.

Parent

RE: Great investment

By BruceLeet on 3/20/2008 8:20:15 PM , Rating: 2

I concur, I have a dual-core Intel and it will be good for another 4 years. The software front isn't developing as fast as the hardware front.

IMO?

Parent

RE: Great investment

By Jedi2155 on 3/20/2008 11:26:42 PM , Rating: 2

I believe thats because its easier to design the hardware than design the software usually.

There are just so many additional considerations when designing software that don't need to be considered when making the hardware.

Parent

How come I can't

By download7 on 3/20/2008 8:24:24 PM , Rating: 5

How come I can't put a CD in my drive without my whole computer coming to a hault just to spin the CD up and read it? I do have a quad core cpu OCed to 3.2ghz.

How come I can't look at a huge picture size (at full size) and SMOOTHLY scroll around to look at all 4 corners? I do have a 640mb video card and 2gig of ram.

RE: How come I can't

By greylica on 3/20/2008 11:49:22 PM , Rating: 2

The problem is the ISA that is made for PCs, and to some other ISAs too. The first processor in the majority of the cases handles all IRQs, and subsequent processors have it´s own IRQs.
When you put your CD, your controller sometimes request the attention of the first processor, and the first handles all of the IRQs of the system, causing this problem.
But there is workarounds in modern systems to solve this problem.

Parent

RE: How come I can't

By encryptkeeper on 3/21/2008 9:01:32 AM , Rating: 2

Would you be so kind as to name some?

Parent

RE: How come I can't

By mcmilljb on 3/21/2008 5:13:04 PM , Rating: 2

Turn off autorun?

Parent

RE: How come I can't

By ineedaname on 3/21/2008 11:31:01 AM , Rating: 2

neither of those have anything to do with multi-core.

Its just because of the way that the software was designed. Your cd drive uses a timer which causes windows to wait for it to load certain things. Even though you have a good video card your picture is being viewed in 2d mode which doesn't utilize your video card in the same way.

Parent

Doing it the hard way

By bebesito21 on 3/21/2008 12:00:43 PM , Rating: 3

I think this is kinda backwards. Seems to me that instead of thinking of a software/programming way to solve the problem, they need to have a hardware solution. The cell processor in the ps3 has a dedicated core that instructs the other cores to keep them busy. With all the cores Intel can fit on a chip, they should dedicate one to order routing. Then developers are free to code away without worring about optimizing threads and cores.

Got to speed it up

By crystal clear on 3/21/2008 10:42:33 AM , Rating: 2

quote:
Microsoft and Intel joined together to build two research centers focused on parallel computing and multicore programming research.

Its a step in the right direction but rather late.

This should have started when quad cores were undergoing testing before their final launch.

Intel/M.S should invest more in START UPS who deliver at a faster rate.

We need the software fast if not there is no point adding more & more cores.

Hardware & Software go hand in hand,right now Software lags far behind the hardware.

Multi Core Let Down

By ineedaname on 3/21/2008 11:44:36 AM , Rating: 2

I switched from single to dual to quad thinking i would get a performance boost only to realize that nothing really can fully utilize my quad. I've played some multi threaded games but nothing really ever goes above 35% usage. Even done some multi threaded encoding but that too only used about up to 35%. At 35% a dual would've done me just as good or even better than a quad since wolfdale's OC so high. I'm sure there are programs out there that CAN utilize all of a quad but the majority of everything I use can't.

So I can't wait for some parallel coding 2 come out since procs are multiplying with cores. With Nahalem coming out not 2 long from now bringing out 6 cores or 8 cores, I doubt parallel processing will b close 2 perfected yet. Its about time they start dumping more research into this.

new name for the article

By fezzik1620 on 3/20/08, Rating: -1

RE: new name for the article

By Master Kenobi

(blog) on 3/20/2008 4:21:43 PM , Rating: 3

Hmmm, an opportunity to teach these kids right the first time. That parallel computing is now, and developers need to keep that in mind. I don't see a problem with this.

Parent

RE: new name for the article

By TomZ on 3/20/2008 5:44:31 PM , Rating: 2

Well, these are research grants, and relatively small ones at that, so the reach to students will be quite small. It will just give some exposure to handful of PhD candidates writing obscure papers that 99.9% of practicing software developers will never even hear about.

Basic research is good, don't get me wrong. But it's going to take a lot more than $20M of grant money to push multi-threaded programming into mainstream widespread use.

Parent

RE: new name for the article

By trexpesto on 3/21/2008 1:25:46 AM , Rating: 2

quote:
... it's going to take a lot more than $20M of grant money to push multi-threaded programming into mainstream widespread use.

I agree, $20M profit would probably see more results. Or contest, like NASA's

Parent

RE: new name for the article

By mcmilljb on 3/21/2008 8:15:09 AM , Rating: 1

This is not for mainstream. This is just paying for some research to be done. I am sure they know exactly what actual projects are being funded by this money right now. Will this chnage they way we program in the short-term? No. Will it change the way we program in the future? Possibly. This is just a getting the ball rolling kind of grant.

If they wanted to be mainstream, they would have a public product launch with a blue screen of death.

Parent