Slashdot Mirror
AMD to Demo '8-socket' Dual-Core Opteron System
flynn_nrg writes "AMD will make the first public demonstration of a system built out of its dual-core processors today, the result of a strategy first made public almost a year ago. Two-core Opteron chips aren't due to ship until the middle of 2005, but AMD will have four of parts running inside an HP ProLiant DL585 server at its Austin plant later today."
No, 4-sockets, each with dual core CPU.
that would be a pretty cripling blow, seeing as how HP is pretty much the biggest supporter of itaniums
- tristan
4 procs, dual cores? Kickass. A short read on implementation differences between AMD64 & Intel's 64. http://www.theinquirer.net/?article=17906
...and you've eaten your pen. simply stunning.
..remember, this may be a Server processor, but it is also the close cousin of the PC dual core offering, see http://www.anandtech.com/cpuchipsets/showdoc.aspx? i=2178&p=2 . So, we will be able to see if it is worthwhile to spend bucketloads of money on a 939 socket athlon motherboard today.
"If a boss demands loyalty, give him integrity. But if he demands integrity, give him loyalty." (John Boyd, 1927-1997)
Also, remember that these are 90nm while all shipping AMD64 chips are 130nm. AMD uses silicon on insulator which I believe means they will have much less problems with leakage at 90nm compared to intel...
In general, power dissapation scales in frequency with n^2, in multiple cores with n. So for the power of a processor 2x as fast, you could probably deliver 2^2=4x with 4 cores.
Granted, this is only true if the task is parallellizable, but with todays multi-tasking computers I could at least use two cores. (If main task is blocked, there's probably a dozen other background processes who'd like a few cycles).
Kjella
Live today, because you never know what tomorrow brings
Yeah, ask Oakley Sunglasses. They seem to have discovered it...
Rule #1 -- Politics always trumps technology.
They are probably two independant cores manufactured seperately, they are just combined in the die. So if any given core fails along the way, you aren't discarding a second perfectly good core. Unless they fail when manufacturing adds them to the die.
If you'd read the article you would know that a stock 4-socket Opteron mainboard used, turned into a 8 CPU system. Such simple (cost effective?) upgradability is a very good reason in my optinion.
And if you take into account that both Intel and AMD decided to go for dual-core, it might be the most logical way to improve the performance of the chips.
Not really. Opteron is part of HP's strategy for 64-bit computing everywhere.
The roadmap looks something like this:
- Tandem (NSK) will eventually turn to Itanium (as soon as lockstep is deployed and working fine);
- Alpha and PA-RISC will evolve into Integrity (Itanium2);
- Proliants (IA-32) will evolve into Opteron.
It's just that 32-bit computing is taking its last breath, and it's time to move on.
Now that looks like a smart move.
HP has several (6 actually) server product lines. They will probably use opterons in their high-volume/lower-profit proliant server line. However they have firmly commited to ditching pa-risc, mips, and alpha for their other 5 server platforms. The high-end/high-profit/low-volume systems are largely independant of the proliant group.
HP doesn't view itanium and opteron as an either/or proposition. Given their product porfolio, it's quite reasonable to use both. Itanium is fast and expensive, a good fit for a 128-way superdome. Opteron is pretty-fast and inexpensive, a good fit for a 4-way proliant.
Socket implies the physical chip. An 8-socket system using a dual core chip implies 16 processors. The poster really meant a 4-socket dual core system.
Dual core makes economic sense. There is a sweet spot for die size (around 100mm squared), below which the production costs start becoming negligible. But with technology improvements, 100mm allow you to fit more and more transistors, and dual core is what gets you the maximum processing power (and therefore money) out of those extra transistors.
CEO lives in Austin.
I have been building systems for some time, I have made many nForce3 systems and have never experienced any hdd or usb noise. Then again my company only sells performance cooled computers, could make a diffrence for sound seepage. I would recommend only using a 10K Sata Raptor drive, since they have been around it's all I use. The real world performance gain only relates to about 2K 3dmarks (3dmark2k1), however overall system speed and performance is greatly enhanced, especially if you are going to skimp and not put 1GB or more of RAM, that faster swap file is quite apparent. If your worried about losing HDD space get a DVD-RW drive at less then 70 bucks for a nice one, it only makes sense to store your stuff on a DVD anyway, for the performance boost.
On another note I should mention if you are into gaming the amd64 core does far more than a high end video card. Obviously the fusion of the two is stellar, however if you have to choose get the cheaper $100 video card and focus on the amd64 core (for all you skimpers out there) with at least 1GB of RAM, you will be very happy with the result.
Weren't there rumors AMD's 90nm Opteron was 105 Watts peak? Make no mistake about it. 90nm is NOT a "cool" process regardless of who makes it.
Most of the losses for chips like these are dynamic - i.e., caused by switching capacitive loads. A 90nm chip has features with half the area that a 130 nm chip has. Even with thinner layers for some features, this results in lower capacitance, and so less heat generation for the same clock rate.
The key words are "for the same clock rate". These chips are hot because they are run faster, not because of feature size.
Itanium is completely different from the x86 line, and its FPU unit absolutely crushes the P4 one.
I wonder if somebody could explain why dual-core CPUs are a good idea. If it's a pair of cores on a single piece of silicon, it seems it would take the same silicon as two separate cpus, so where's the benefit?
Less packaging overhead, and faster communication between cores (on-die bandwidth and latency are far, far better than any motherboard's crossbar's bandwidth and latency).
You also have less contention over memory, for single-chip systems with multiple cores vs. multi-chip systems. Instead of having to muck about with cache coherence across a bus, the chip looks like a single processor as far as the memory subsystem is concerned, with coherence operations only involving the first one or two cache levels on-die.
yield decreases roughly exponentially with die size, which argues for 2 separate cpus.
Processes are optimized so that you can build a chip with 1-2 square centimetres of area with reasonable yield (as this is what chip manufacturers demand). This has been pretty constant (or if anything, has been increasing). However, with each design generation, the number of transistors available in this area has doubled. We're now at the point where we can get high yields on chips with enough transistors that multi-core designs make sense.
A chip with N cores also doesn't take N times as much area as a single-core chip, as the lowest levels of cache aren't duplicated (just L1 and usually now L2). So overhead is reasonable, and the real estate is there. It makes a lot of sense to use it.
Look for a report this afternoon on AMDZone.com.
ignorance is bliss. googlefiberatx.com
The 1xx has everything the 8xx has, except it lacks two hypertransport links. The 1xx series has only a single link, and this must be used for system I/O, which is why the 1xx is exclusively uniprocessor. The 2xx series have two HT links, so they can form multiprocessor systems, and the 8xx, with three, can form efficient systems up to 8-way.
You have almost no idea what you're talking about. It might be a good idea for you to simply accept your fate as a mediocre sysadmin, a profession any number of teenagers and grease monkeys could have just as easily. I will repeat: you do not know what you are talking about. Shut the fuck up, Gelinas, your stupid pulp novels about economic history don't even relate to the shit you (try to) explain. I almost want to fucking puke every time I see you post this stupid crap, stroking your goatee and twiddling your pony tail, because deep down inside you desire to have sex with the men you pretend to intellectualise with. Unfortunately for yourself, I'm addicted to hating you, J. Maynard Gelinas of MIT's Nuclear Laboratory system administration department (lol, what a phoney title, ape).
http://www.digit-life.com/articles2/amd-hammer-fam ily/
Basically, 2.0 ghz Opteron SPECfp peak 1170
1 ghz Itanium 2 SPECfp peak 1356.
Think for yourself, destroy your television.