Four Core Processor to Bring Tera Ops

panhandler writes "As reported at CNet and the Austin American Statesman, researchers at UT are working with IBM on a new CPU architecture called TRIPS (Tera-op Reliable Intelligently adaptive Processing System). According to IBM, 'at the heart of the TRIPS architecture is a new concept called 'block-oriented execution"' which will result in a processor capable of executing more than 1 trillion operations per second."

Comments

Researchers ...

[makapuf]

at Unreal Tournament ? Why, some have cool jobs.

one trillion ops/second

[Adolf Hitroll]

this way you, yankees, can count every dollar of your actual external debt in a little more than a second !

Great....

[innosent]

Great... Just what we need, processors that can perform an instruction, then wait 40000 cycles for the next instruction to be read from memory. I wish we could see some memory improvements to go along with these.

Seriously, though, this will help break all the clustering records, provided we can come up with faster interconnects by then.

Re:Great....

[n3rd]

Great... Just what we need, processors that can perform an instruction, then wait 40000 cycles for the next instruction to be read from memory. I wish we could see some memory improvements to go along with these.

Sun is working on something along those lines, check it out

Re:Great....

[innosent]

That's throughput they're working on, which is great, but not the problem. Latency is the problem, not throughput. Try having large programs with lots of branches and/or syscalls: If the code is large enough, you'll spend more time bringing pages in from memory than actually executing your code, especially since you can forget about pipelining benefits...

Personally, I wish a company would throw out every idea from current memory, put a GB of cache on a chip, and get memory access times down to about 3 picoseconds. But memory doesn't have the marketing appeal that processors do, so we're screwed.

Re:Great....

[asavage]

Personally, I wish a company would throw out every idea from current memory, put a GB of cache on a chip, and get memory access times down to about 3 picoseconds. But memory doesn't have the marketing appeal that processors do, so we're screwed.

The problem is the larger the cache size, the slower the access time. It is a trade off.

Re:Great....

[AlecC]

True, for general purpose computing, which is probably what most /.ers probably do. But this sort of massive processing power is really only needed for the simulation people, who do large amounts of contoguous number cruncong, such as matrix multiplies. That sort of thing will be speeded up enormously by this sort of architecture.

As a concept, this is hardly new. There have been all sorts of different parallel processing architectures over the years - SIMD, MIMD, strings, arrays, etc. Each has performed well on one particular class of problem, but generally crashed out as soon as you move away from its sweet spot. For example, changing gear from array mode (lots of big matrix ops) to linear mode (making decisions on the results) tends to lose most of the time gained by the array processor. OTOH, clusters of general purpose processors tend to wast time on memory or object contention unless they are working on highly decomposable problems.

Maybe this architecture is a better compromise that its many predecessors. (It ought to be - IBM built enough of them). But don't expect to use them youself, unless you work for an advanced engineering company, a weather forecaster or the DoD.

Re:Great....

[drinkypoo]

Maybe someone should just get SRAM working at higher speeds and densities without making it bloody expensive as it has been, thus eliminating the need to do refreshes, and enabling asynchronous reads from memory at synchronous speeds but without the need to be synchronous.

Or maybe IBM's MRAM will do this for us.

And Windows 2005...

[Anonymous Coward]

Will still take five minutes to boot into a login prompt. Some things never change.

Re:And Windows 2005...

[slittle]

??

Things change plenty. Windows' boot times have been improving in recent years, esp. compared to the Win9x days.

I think you meant to make a crack about Doom III or something...

Re:And Windows 2005...

[julesh]

Come on, funny as the line might be, timing from power-on to having a working desktop, my systems come in like this:

Windows 2000 - 45 seconds
Windows 98 - 1 minute
Linux + KDE 2 - 1 minute 10 seconds

(Admittedly linux + console is about 20 seconds, but that's not really a fair comparison - Windows 98 'text mode only', i.e. DOS, is only about 2 seconds).

Also, boot up time is largely IO bound. Improving your processor speed will make comparatively little difference (I think doubling speed might shave 10% off these figures, possibly more for the Linux one because the KDE issue is dynamic linking related which is a CPU problem).

Better link

[Textbook Error]

A somewhat more informative link for more info. Would it really kill submitters to put a link to the actual project in their submission...

Thank God

[(outer-limits)]

EPIC is clearly dead in the water. Intel didn't learn from the 432.

See The Project Yourself

[robbyjo]

Here

SCO's response:

[10Ghz]

"This is yet another breach of our IP! Our fine researcher came up with this technology over 10 years ago, we have just ket it hidden for all this time. Unfortunately we wrote the patent-applications with invisible ninja-ink and they are being kept in a vault in our Fortress of Doom (tm), so we can't show them to anyone.

We expect IBM to pay us 5 billion dollars plus 4 x $699 for each CPU sold"

Fabrication

[Anonymous Coward]

Does anyone remember the Pentium Pro? It was an extremely expensive processor. This was because of its strange system of connecting the CPU core with a massive amount of cache ram; production yields were very low, so fabrication costs were very high.

Imagine how high the failure rate would be with fabricating a CPU with four cores... I don't see how it would be practical unless it was with an extremely-high yield design such as the StrongARM.

Re:Fabrication

[Nazmun]

what if they could fabricate each core separately and then somehow connect the cpu's. Shouldn't be too hard to do in the factory. It wouldn't be as fast as a single core cpu's internal bus but it would be a heck of a lot better then mult cpu's in standard mobos now (like xeon's etc.).

Re:Fabrication

[ottawanker]

Imagine how high the failure rate would be with fabricating a CPU with four cores... I don't see how it would be practical unless it was with an extremely-high yield design such as the StrongARM.

Naw, that doesn't seem like too big a problem. All they have to do is check to see how many cores are working, and then sell the chips like that. Something like this (assuming you pay a premium for more cores, relative to the lower yields):

$500 for 1 core
$1200 for 2 cores
$1800 for 3 cores
$2500 for 4 cores

Only 32 Billion Now

[Nazmun]

The four cores add up to only 32 billion operations right now according to the CNet article. They project that they won't reach 1 trillion until 2010.

Oooh, can't wait for the G6!

[Capt'n Hector]

But of course, these processors will require entire software rewrites.

But this reminds me of a growing trend, and that is that as soon as large infrastructures are finally completed (be it the transition to OS X or 802.11b) the technology becomes obsolete. However, the entire infrastructure must be replaced. I don't care how many gazillion flops this or any other processor can pull. They need to easily scale so that the entire infrastructure does not need replacing.

We'll need a lot better compilers

[kramer2718]

If each chip is basically four processors each of which can execute 16 operations simultaneously, it will be difficult for compilers to find 64 independent instructions to execute each cycle.

I guess one possibilty could be to execute instructions from four different processes simultaneously, thus reducing the probability that the instuctions will interfere.

Re:We'll need a lot better compilers

[boopus]

Exactly. The IA64/itanic/itanium instruction set provides for executing multiple instructions "simultaneously" (aka: pipelined with no interference) but the intel guy I heard from said it so far doesn't provide anything close to the improvements they hoped the feature might. Scaling it up to 64 instructions per clock is only going to help tasks which IBM supercomputers have already lost to beowolf clusters.

Re:We'll need a lot better compilers

[Alain Williams]

• it will be difficult for compilers to find 64
• independent instructions to execute each cycle

The problem is that the word independent is the wrong one.

It depends on what sort of work you choose to do on this sort of beast, finite element work (simulations, etc) involves the same operation on lots of values over and over. This is how Cray made his money years ago.

This is not a desktop machine for you to do office automation on, quake maybe, but not word smithing.

Just a warning

[Kizzle]

If anyone in any way shape or form mentions the word beowulf, expect a swift kick in the nuts by your's truly.

That is all

Re:Just a warning

[iapetus]

But just imagine a BeowAAAAAAAAARRRRGGHH!

Ooh, now I can sing soprano again...

Let me guess.

[JanusFury]

capable of executing more than 1 trillion operations per second."

Let me guess...

NOP
NOP
NOP
NOP
NOP
NOP
NOP
...

Cell

[Jagunco]

Wasn't the PS3 "Cell" chip made by IBM and Sony supposed to deliver 1 teraflop too?

Memory

[Detritus]

It's easy to throw 8 processors on a motherboard. The hard part is designing a memory subsystem that can supply the bandwidth for 8 processors and any other bus masters. Plus, you have to provide cache coherency for all of those processors.

Re:Memory

[White Rabbitt]

Wasn't one of the main premises of the TRIPS system that each processor is more or less independent of its neighbors? They used the term "network" to describe the processors' interrelationship. With operating systems that commonly are processing 20 threads when no apps are running (*cough* XP), what would be the advantage of increased "networking" of the processors?

It appears that the only issue that would be solved is there would be less lag between processors--but at the speeds they're talking about, the memory supply and caching will present new issues anyhow, ne?

Sorry about my complete disregard for proper terminology...

Lemme Guess...

[SharpFang]

TRIPS. Lemme guess. The name says all about reliability of the system.

But...

[Anonymous Coward]

A) Will it run linux
B) Run Quake 3 at an acceptable FPS
C) Take a slashdoting
D) Make my Coffee
E) Run Linux
F) Where is my flying car!?

I just wanted to say...

[boola-boola]

....that I've had Doug Burger and Steve Keckler as professors here at UT, and not only do they know their stuff, but they're great professors as well, and they really seem to intimately care about the technology. They have a great sense of humor too (such as Dr. Burger complaining that he doesn't even have root access to his own machines :-P)

Unfair comparison

[axxackall]

W2k keeps loading its services even AFTER I login. I can change the boot sequence order to Linux time for X11 Login prompt at least half.

Well, I don;t need postfix, Apache, Zope, MySQL, PostgreSQL and many other services at the moment of login. So, Win2k designers has recognized the it and optimized the boot sequence being oriented for a desktop user. In Linux we still keep a server-oriented mentality, that's why XDM/GDM/KDM/EDM is always the last thing to start.

Besides, Win2 boots some services in parallel, while in Linux we still boot all of them sequentially, waiting for [OK] string before starting the next one. The only way to paralelize the sequence is to track dependencies between services. In Gentoo there are some efforts to do the parallel boot.

But as for now, Linux is (by dfault) is oriented for servers, and GUI login is the last (ltterally last) thing you need on your server.

Re:Unfair comparison

[julesh]

Actually, I have substantially optimised my Linux startup times to get it down to that. I've removed a load of non-essential services (I'm not running a mail server or web server at all now, I only really have stuff that runs from inetd and mysql running other than the absolute essentials) and moved the X startup so that it happens before a lot of other stuff has loaded.

OK, I'll admit that I haven't parallelised it beyond this, but I wouldn't expect to see a huge amount of improvement from that. Besides, most unix daemons fork and terminate the parent process before doing very much, in much the same way that most WinNT services just call StartServiceControlHandler (or whatever its called) first thing as they get into their WinMain()... there's not a lot to gain by parellelising that.

IBM can work with the Japanese

[Mostly a lurker]

Have you noticed that big Japanese companies seem comfortable working with IBM? I find it difficult to think of any other large US corporation about which we can say the same. IMHO, it is because (while a hard nosed competitor) they deal in a straight fashion with partners. They are seen as trustworthy.

How is Gentoo doing it?

[roystgnr]

Besides, Win2 boots some services in parallel, while in Linux we still boot all of them sequentially, waiting for [OK] string before starting the next one. The only way to paralelize the sequence is to track dependencies between services. In Gentoo there are some efforts to do the parallel boot.

How are they doing it?

I've often thought that we should be booting up our computers with a parallel invocation of "make". Then when adding a new service you would have none of this "what number between 0 and 100 should I assign?" foolishness: just write a three line makefile that includes all the dependencies that your service has on others.

Re:How is Gentoo doing it?

[mkldev]

You mean kind of like Mac OS X does? From the docs on OpenDarwin:

The Property List

Each startup item bundle contains a property list file at the root level named StartupParameters.plist. The property list is an XML or NeXT-style text file that describes the contents of the bundle. It enumerates the services the bundle prov ides, the services the bundle requires, and other information useful for determining the proper order of execution of the bundles.

The property list contains the following attributes:

OPEN CURLY BRACE
Description = "My Startup Item";
Provides = "MyService";
Requires = ("AnotherService", "Network", ...);
Uses = ("YetAnotherService, ...);
OrderPreferece = "time";
Messages EQUALS OPEN CURLY BRACE
start = "Starting My Item.";
stop = "Stopping My Item.";
restart = "Restarting My Item.";
CLOSE CURLY BRACE
CLOSE CURLY BRACE

Apologies for the EQUALS, OPEN CURLY BRACE, and CLOSE CURLY BRACE, but Slashdot considers them to be 'junk'. Oddly enough, it also thinks double quotes are junk. Talk about encouraging plagiarism.

Here's a modest proposal: if somebody has a Karma bonus, it should be clear that the person doesn't post intentional trolls or other useless crap. Don't subject those of us who actually try to consistently post useful information to these sorts of stupid filters. It only ends up preventing us from being helpful and informative and leads to the decline of the signal-to-noise ratio that it was designed to improve.