8-core monster crunchers

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,199
Credit: 133,823,847
RAC: 13,298

RE: I happen to know where

Message 80755 in response to message 80753

Quote:
I happen to know where there are quite a few 486's. Probably several 100, another industrial control and measurement system that is being re-furbished. I'll talk to my friend to see what the score is, before they throw them out. He said about two weeks ago, at the moment, the computers are been kept as spares until re-furbishment is completed.


Several 100?? :-)
Well I'm fairly certain NASA would love to hear from you. They are ( quite rightly ) preferring maintenance of tested systems rather than retrofitting some new paradigm.

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter. Blaise Pascal

peanut
peanut
Joined: 4 May 07
Posts: 162
Credit: 9,644,812
RAC: 0

RE: RE: Presumably the

Message 80756 in response to message 80752

Quote:
Quote:

Presumably the smaller resolution of more modern components make it easier to knock them out with the odd high energy cosmic ray etc ...

Cheers, Mike.


I jokingly have used the stray cosmic ray toggling a bit excuse when equipment did things it should not have done. What I really thought, and am pretty sure was the truth, was that an operator did something wrong. Programs usually mess up in bad ways not in controlled fowl ups. To err is human but to really mess things up takes a computer.
DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,343
Credit: 2,066,546,946
RAC: 1,266,276

RE: Well I'm fairly

Message 80757 in response to message 80755

Quote:


Well I'm fairly certain NASA would love to hear from you. They are ( quite rightly ) preferring maintenance of tested systems rather than retrofitting some new paradigm.

Cheers, Mike.

ARe you sure you're not misremebering? Nasa's most recent salvage CPU expedition was for 8086's to replace ailing hardware on the shuttle.

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,199
Credit: 133,823,847
RAC: 13,298

RE: ARe you sure you're not

Message 80758 in response to message 80757

Quote:
ARe you sure you're not misremebering? Nasa's most recent salvage CPU expedition was for 8086's to replace ailing hardware on the shuttle.


I think that's correct, but the history with NASA is much longer and wider, as they've included much of the '86 line ( 8088 thru 80486 ) in their 'going-out-of-the-atmosphere' hardware ranging from :

- balloon borne experiments
- SR71 Blackbird laden with X-33 'aerospike' research at high altitude
- FA/18 research into 'laminar flow' on wing surfaces
- space probes like Galileo, Cassini-Huygens, Solar-B & I think Magellan and the Mars Polar Lander ( ooops .... ) as well
- Hubble, 286 originally then to 486 on the 'big fix' mission
- Many Shuttle turnouts. For instance there were IBM ThinkPads on Columbia - 486 @ 75MHz.
- the ISS has various '86 line types in orbit as we speak and not just 'spacecraft' stuff but onboard experiments ( example )
- for that matter the GPS satellite web has radiation hardened 386's
- and ground based matched/mirrored simulators for the above

Even future use still seems to be mooted [ see the NASA Electronic Parts & Packaging Program, say for instance here for a relatively recent discussion of the sorts of environmental rigours that space work leads stuff into ]

I guess the whole '86 crew is so well understood and physically suitable. Because of it's more general popularity then human expertise, experience, software tools etc abound. Like yourself (?) I initially underestimated the usage, but I think NASA was/still-is a big client of Intel ... I even vaguely remember special versions being turned out for them only .... would that be right Peter?

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter. Blaise Pascal

archae86
archae86
Joined: 6 Dec 05
Posts: 2,994
Credit: 4,545,902,936
RAC: 4,418,376

RE: I even vaguely

Message 80759 in response to message 80758

Quote:
I even vaguely remember special versions being turned out for them only .... would that be right Peter?


We ran the 80386 wafers for the ISS in Fab 9 here just outside Albuquerque New Mexico quite a few years ago.

However there is nothing special about them except for extra inspections and measurements taken during production to satisfy the mostly bureaucratic requirements NASA imposes in hopes of getting reliability. The 80386s were stored for quite a while as the space station had various delays. They are flying in it today, at the heart of the station's control systems.

Of course more modern processors travel to the station--some embedded in equipment, and more in laptops used for various purposes by the astronauts.

I'm not aware of any functionally distinct designs done by Intel for NASA. I think NASA has used genuinely rad-hard functional equivalents to one or more Intel processor design, processed on a tiny boutique line maintained here in Albuquerque at Sandia Labs for that purpose. Those, however, are rather substantially different in design and manufacturing than Intel's own parts. Intel's main involvement has been to provide some design information and to refrain from excessive license fee requests (they gave a free license to a Pentium design in 1998 according to a Sandia Labs press release--don't know whether that one got finished or not).

As to reasons why, I think it wise not to underestimate the value of the available development environments. I happened to chat briefly with a Lockheed Missiles and Space guy about processor choice on their birds sometime in the 1980s, and he made it clear that having to start over on his development environment would be a killer objection to the advantages a new architecture processor choice might have.

dulcimoo
dulcimoo
Joined: 3 May 07
Posts: 2
Credit: 1,136,912
RAC: 0

RE: Awhile back there were

Quote:
Awhile back there were a few articles written about 8-core systems that should be ready soon for home PCs. Does anyone here know when they will be ready for us crunchers to start upgrading to these new monster crunchers?

You can already get on from Apple. 2799 MSRP for dual 2.8 GHz 4 core Intel XENON mPUs.

Only 2 GB of RAM and 1 HD for that price however.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,516
Credit: 463,939,239
RAC: 7,934

While not being an 8-core

While not being an 8-core chip, the upcoming Nehalem which is previewed here has 8 logical cores (hyperthreading), so I guess this thread is adequate :-).

The preliminary performance figures from the article look very promising indeed. Looks like Nehalem will give a very solid 20...50% percent (depending on benchmark) compared to Intels best 45nm Penryns.

I have to do a bit of googling to find some test samples doing BOINC, the CPUID should be "Family 6 Model 26 Stepping 2".

I can't wait to see these chips getting affordable :-)

CU
Bikeman

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,343
Credit: 2,066,546,946
RAC: 1,266,276

RE: The preliminary

Message 80762 in response to message 80761

Quote:

The preliminary performance figures from the article look very promising indeed. Looks like Nehalem will give a very solid 20...50% percent (depending on benchmark) compared to Intels best 45nm Penryns.

I have to do a bit of googling to find some test samples doing BOINC, the CPUID should be "Family 6 Model 26 Stepping 2".

I can't wait to see these chips getting affordable :-)

It shouldn't be too long depending on your definition of affordable. Intel's planning to release them initially in $300/500/1000 dollar parts (2.6, 2.9, 3.2gig). The X58 chipset is intended to cover the upper part of the P45 market as well as the X48 feature range, so there should be $200ish mobos from the start. 3 gigs of DDR3-1600 is currently ~$200 or 6gigs for $300. That's $700-800 for the core system, depending on what else you wanted to do besides crunching on it a full system could be build for under $1k.

Donald A. Tevault
Donald A. Tevault
Joined: 17 Feb 06
Posts: 439
Credit: 73,516,529
RAC: 0

RE: RE: The preliminary

Message 80763 in response to message 80762

Quote:
Quote:

The preliminary performance figures from the article look very promising indeed. Looks like Nehalem will give a very solid 20...50% percent (depending on benchmark) compared to Intels best 45nm Penryns.

I have to do a bit of googling to find some test samples doing BOINC, the CPUID should be "Family 6 Model 26 Stepping 2".

I can't wait to see these chips getting affordable :-)

It shouldn't be too long depending on your definition of affordable. Intel's planning to release them initially in $300/500/1000 dollar parts (2.6, 2.9, 3.2gig). The X58 chipset is intended to cover the upper part of the P45 market as well as the X48 feature range, so there should be $200ish mobos from the start. 3 gigs of DDR3-1600 is currently ~$200 or 6gigs for $300. That's $700-800 for the core system, depending on what else you wanted to do besides crunching on it a full system could be build for under $1k.

On the other hand. . .

AMD doesn't have anything at all that can compete at the top end. So, I doubt that Intel will be in any real hurry to cut prices.

archae86
archae86
Joined: 6 Dec 05
Posts: 2,994
Credit: 4,545,902,936
RAC: 4,418,376

RE: AMD doesn't have

Message 80764 in response to message 80763

Quote:
AMD doesn't have anything at all that can compete at the top end. So, I doubt that Intel will be in any real hurry to cut prices.


Without a doubt competition matters, but it is not at all the whole story.

There is a pretty simple rule which usually works: Intel will jigger the product mix and pricing to sell out the available fab capacity at any given time to maximum revenue. (not quite, as they will sometimes pursue longer term strategic goals--for example the summer 2007 amazing Q6600 price drop appears to have been a deliberate attempt to condition the mass market toward greater than two cores)

As the folks willing to pay $2000 for a CPU are very, very limited, that means that if capacity comes online at anything like the rate I expect, that Atom/Pennryn/Nehalem prices will come down considerably to sell the avalanche of capacity coming on line.

Mind you, I'm not predicting really cheap Nehalem real soon now, but unless there is a product problem (of a sort I've not detected on the jungle tom-tom's yet) I think you'll find some interesting spec points priced where folks here would be willing to buy them pretty soon after introduction.

After I wrote that I googled a little, and noticed that today's Wikipedia entry for Nehalem has some pricing entries, the lowest of which is referred to this story which claims a $284 price in 1000-unit quantities for a 2.66 GHz Bloomfield (4 cores, 8 threads). Not top-of-the-line to be sure, but not a price reserved to servers and royalty, either.

Caveat: I have zero inside information on pricing--though I think this sort of web rumor has turned out true more often than not, and the 1000-unit price list price has often turned out pretty close to the "careful shopper" price one can find for onesies on the web. I do have a little inside information, and so far as it goes the Nehalem hype meter is still green internally as well as externally.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.