And you thought dual core was advanced...

[doc_bean]

From Gamespot:

Nicknamed the "Quad-father," AMD's new 4x4 platform will support two dual-core processors on a single motherboard for a total of four processing cores.

[Papewaio]

Quad Xeons have been around awhile.

All this is trickle down of server style technology to the PC level.

Also it really isn't that different in effect to the old co-processor.

[rory_20_uk]

Then there's the Cell, and this

Four? Pah...

[shifty157]

2007 should see the arrival of Quad-core processors from Intel and AMD.

[Lemur]

And just as soon as there's software that will run in Windows that actually takes advantage of all those cores, we'll be dancing the happy dance.

[lars573]

From Gamespot:

Seen them, put it together too. It was both my job and my pleasure. An MSI Opteron MoBo. With 2 Opteron (natch) CPU's in it. Problem is that the chipset was desgined for servers and need ECC (error correction code) RAM. The machine wouldn't even post till we figured that out. It's $150 per stick at 1 gig. The MB was also SLI ready.

[hoom]

I thought 4*4 is supposed to use non ECC RAM?!
[edit]Gah! commonsense decided to appear late:
Chipset has nothing to do with EEC or not since memory controller is on the CPU so Opteron = ECC, A64 or FX CPU = non ECC.

Doesn't really float my boat on account of I'm already not happy about the price I'm going to need to pay for a good dual-core, let alone 2

Especially if Conroe comes out at quite reasonable price & unless AMD really drops prices significantly.

[Beirut]

I read, a few years ago, that the NSA (morning Echelon, now that you're reading this...) used computers that were made of four-thousand AMD 800mzh CPUs strung together.

Anyone else hear of this?

[edyzmedieval]

Technology advances fast, like the time - Modern proverb.

I'll get a Conroe though.Scrap the AMD's.

[lars573]

I thought 4*4 is supposed to use non ECC RAM?!
[edit]Gah! commonsense decided to appear late:
Chipset has nothing to do with EEC or not since memory controller is on the CPU so Opteron = ECC, A64 or FX CPU = non ECC.

Doesn't really float my boat on account of I'm already not happy about the price I'm going to need to pay for a good dual-core, let alone 2

Especially if Conroe comes out at quite reasonable price & unless AMD really drops prices significantly.

I did say that it was a server Mobo, an MSI Master2 FAR, to be exact. Honestly I think who ever ordered the machine read that article (or one like it) and though they were out now. They aren't, what he got was a stop gap Opteron MoBo with SLI.

Also AMD X2's and FX's are priced similarly to a Pentium D. If the Conroe is loads cheaper than the Pentium D then I'm sure AMD will cut their prices.

[drone]

I heard somewhere that having multi-processors requires multiple licences for Windows, whereas with a multi-core processors only 1 is required. Is this the case?

[Lemur]

Pretty sure Microsft is not going that route. There are SW makers, however, who do want to charger per core. Typically database stuff, where the SW is residing on a big server. I'll dig up a link if I've got time. Shouldn't affect desktop users in any way or form, however.

Hmm, looks like a lot of these issues surfaced in '04. Here's an article on Microsoft rejecting per-core pricing, even for server software. Looks like the only large vendor getting serious about charging per core is Oracle:

Some vendors are still sticking to per-core licensing, and corporate IT shops caught in contracts with them could be turned off. "One of the issues with multi-core is scalability," he said. "When you get up to 16 or 32 cores, the licensing would be so much, you wouldn't use Oracle."

Reynolds stressed that the issue is only prevalent in a quickly-expanding server setting; for most typical setups aimed at reliability and stability, Oracle still has an advantage. However, Reynolds warned that even though the multi-core licensing mess will have its greatest impact in five or more years, it requires planning now. "It's one of those things where you have to make the policy now," he said, adding that Oracle may change its own policy on per-core licensing at some point, but that real change will be driven only by customers.

[drone]

Microsoft rejected per-core pricing for per-CPU pricing for its server software, such as SQL Server and BizTalk Server, and other software such as Windows XP. The decision means software prices won't increase based on the number of cores on a chip. For example, licensing SQL Server on a dual-core chip would be the same as the price of a license on a single-core chip.

"If Microsoft had come out like IBM and Oracle, the open source community would have had a field day," says Jonathan Eunice, president of research firm Illuminata. "Microsoft has dodged a competitive bullet there."

While users are happy that the dual-core decision won't affect licensing prices, reaction was muted because many users have lingering issues with CPU-based pricing, which they contend is a penalty for the improved performance of more-sophisticated hardware.

"I think everything is still status quo," says Matthew Bailey, LAN engineer for CSK Auto in Phoenix, which operates Checker, Schucks and Kragen auto parts stores. "The per-processor pricing is a strategy that most people don't like. It penalizes you for buying more processing power."

While Microsoft's licensing obviously disagrees with that assessment, it's not taking the same angle on dual-core technology.

I think this is what I heard. Cores within a processor are free, extra processors means more licenses are necessary. Which means that this 4x4 would require 2 XP licenses.

[Vladimir]

Do you think that four barreled processor is faster than this?

http://www.newscientisttech.com/arti...ng-speeds.html

A silicon-based microprocessor has been accelerated to a record speed of 500 billion operations per second (500 gigahertz) in an experiment that raises hopes for super-fast, yet cost-effective, electronic devices.

[orangat]

And just as soon as there's software that will run in Windows that actually takes advantage of all those cores, we'll be dancing the happy dance.

If the application is multithreaded then quad cores will by definition work better than dual cores.

[Lemur]

Correct. The problem that I was alluding to was how much software is not multithreaded ...

[hoom]

500GHZ of doing nothing useful is not all that impressive.

Lars, it seemed like you were talking about a dual Opteron but then you were calling it a 4*4 so you know

Hmm, rumor has it that AMDs AM2 processors will be able to use 'anti-hyperthreading' to get a single thread to somehow run split over the dual cores.

Quite how this works & whether it will actually produce usable improvements is yet to be seen, a bunch of people seem to think its pure bunkem but others seem excited.

[Sasaki Kojiro]

I read, a few years ago, that the NSA (morning Echelon, now that you're reading this...) used computers that were made of four-thousand AMD 800mzh CPUs strung together.

Anyone else hear of this?

It's a cheap way to make a supercomputer, fairly common in universities as well. Though perhaps not with 4,000.

[lars573]

500GHZ of doing nothing useful is not all that impressive.

Lars, it seemed like you were talking about a dual Opteron but then you were calling it a 4*4 so you know

Hmm, rumor has it that AMDs AM2 processors will be able to use 'anti-hyperthreading' to get a single thread to somehow run split over the dual cores.

Quite how this works & whether it will actually produce usable improvements is yet to be seen, a bunch of people seem to think its pure bunkem but others seem excited.

And the only differences are going to be the socket (939 instead of 940) and the CPU that goes into it (Athlon 64 X2 and FX) and no EEC ram. Really all it's going to be is an affordable version of the Opteron stuff. So I feel comfortable saying I've worked with it.

[hoom]

On the face of it yes the only difference should be the socket & RAM but AMD are saying its not just a dual socket Opteron board with different sockets & RAM...

Quite how it can be significantly different is not clear at all.

[edyzmedieval]

What about Conroe? Any details what motherboards will support it?

[lars573]

What ever MoBo configuration that Intel gives to it's vendors to make them with.

[hoom]

ATI is doing RD600 with 3 PCIE * 16 slots (though only a PCIE *4 connection on the 3rd slot) so they can do crossfire + a 3rd card dedicated as physics chip...

[rory_20_uk]

Is the physics really worth it though? Certainly dedicated chips aren't all that I thought they were going to be.

[Lemur]

Well, I haven't read a sinlge positive review of the PhysX card. I don't know if anybody has tested the GPU versions, though. Last I heard, Nvidia hadn't gotten much past the announcement phase on that tech ...

[hoom]

ATI reckons their X1600XT should be able to easily beat the Ageia card & they are probably right.

I think its important to note the difference between game play physics & effects physics, the ATI solution at least is aimed at effects physics rather than gameplay.

Gameplay is stuff like interactive object collisions, bullet trajectories etc.
Effects is stuff like sparks & flying debris.

If you try to run gameplay physics on a seperate accelerator ie off the CPU, it may actually decelerate due to latency of sending, processing & returning results before you can actually render the next frame.