Slashdot Mirror
More Juicy Dual-Processor Goodness
ronmon writes: "I was cruising around bp6.com (I still want one) and happened to see a link to some pics of dual Socket A motherboards. It's in Japanese, so I can't read most of it, but this particular board caught my eye. It's a SuperMicro sporting two sockets and five DIMM slots, plus four drive connectors (IDE RAID?). Yummy!"
And credulous reader Jim writes: "This one gave me a start when I woke up this morning. 2CPU.com has a screenshot of sisoft sandra scores from a Dual 1.53GHz (11.5x133)T-bird box. Apparently from an anonymous email. The scores are nothing short of amazing. Check it out." Grain of salt, remember.
I know there's a lot of hype behind supermicro, I know they have really cool feature lists.
But I know too many people who have paid a premium to own a supermicro product only to find that they were slightly flaky, that the super nifty features didn't work properly, and that the boards were quickly orphaned (support discontinued) when newer boards were released.
I won't buy 'em, and don't recommend 'em.
This is just like television, only you can see much further.
I found this over at amdzone I agree:
2CPU has what they are calling Dual 1.53GHz Athlon scores on a Tyan board. Well, there are a couple of problems with that which make me very, very unsure that these are legit. First of the Tyan 760MP board does not have overclocking features. I know, I held one in my hands at Comdex. So you think that they might add clock multiplier features in the meantime? Nope, what they had was the final revision of the board which only needs a final chipset from AMD to be complete. Second, it is very difficult to get a 1.2GHz Athlon to even 1.4GHz, much less 1.53GHz, and then you are telling me they got two to go that high? Lastly there are no details about this system at all except that it is using the Tyan board. Who knows, maybe it could be right, and I'm not saying 2CPU is making it up, but there are not enough details and not enough evidence, and there is too much logic keeping me from believing it. And for the fools that will say that I am jealous because I don't have dual scores to post here, don't even bother e-mailing me. That is ridiculous.
Pretty much says it all regarding the benchmarks we "saw".
Try to hack my 31337 firewall!
Comment removed based on user account deletion
________
Does anyone actually have a Java program designed to control air traffic, or for the operation of a nuclear facility?
- Installing the latest RPM or DEB packages, doing a bit of config work via rsh, and being able to scale to 64 way by putting 64 boxes in the rackmount unit, or
- Designing a motherboard that integrates together 64 CPUs?
SMP certainly provides the benefit of very fast communication between CPUs, so if processing is strongly dependent on that, SMP wins.But it is most certainly a lot more expensive to scale it up. The vendors may sell 2-way and 4-way SMP motherboards for not overly princely sums, but moving on to higher multiples guarantees pretty monstrous prices, because you're simultaneously mandating:
- Small production runs, compared to the teeming hordes of single CPU motherboards, and
- Sitting at the bleeding edge, because nobody wants a souped up version of a CPU that was "state of the art," LAST YEAR.
Kernel compiles aren't liable to benefit all that much from parallelism, moving forward; consider that with a Pentium III or Athlon, it doesn't take long to recompile Linux even from scratch. It's just getting to be less and less an issue.As for benefiting from other forms of parallelism, it is entirely likely that the toolsets surrounding Beowulf and PVM will improve over time to make it easier to manage doing "clustered tasks" in much the same way that we have progressed from having rather primitive "package management" tools to having stuff like AutoRPM, apt-get , and BSD Ports.
If you're not part of the solution, you're part of the precipitate.
The 5 DIMM slots are almost certainly a combination of 2/3 with 2 being one type, and 3 of another (between SDRAM and DDR-SDRAM).
But what is more apperent now is that SMP is truly a dead end when it comes to multi processor systems.
// EvilJohn
// Java Geek
I disagree. Clustering does NOT solve the issues solved by SMP. Many applications simply do not function that well clustered (i.e. databases). Intel simply removed SMP from the P4 chip due to cost, not due to the lack effectiveness of SMP.
If Intel had given up on it, why is a major focus of the Itanium multiprocessor operations? The flaws of P3 in reference to the poor scalability beyond four (or even two) CPUs are due primarily to the bus design of the processors, not due to SMP.
The EVE6 bus, much like the Itanium bus, should really begin to approach the Linear scalibility we're all looking for out of properly written applications.
Less Talk, More Beer.
The 760MP has been more than a passing interest to me lately, and I've been digging up information/rumors about it daily for the past week. This is what is sounds like:
One EV6 bus requires a hefty chunk of PCB. Two will require even more (of course). This means that DP Athlon boards will require as many as 8 layers to fit into an ATX form factor. IIRC, the industry standard is 4 or 6, so this would be a new (and likely expensive) manufacturing process that may require new tooling to produce in bulk.
Athlon boards are already more expensive than P3 boards, and I think the overall DP price/performance comparison isn't going to be that bad for Intel when/if 760MP ships.
--
Maybe peyote juice?
But then again we must remember that this story has been promoted to front page material by the same group that brought us nanopants.
That's not a grain of salt the editor mentions, that's a rock of crack ...
But I digress ...
Why do I strongly suspect this is non-authentic? Does anyone else remember the photograph of the modified Duron that was supposedly being produced by AMD to thwart overclockers. A pin was physically "removed" from the pin interface.
The picture was posted all over the place. Everyone was all up in arms. It was the end of all things.
And then someone noticed that this "modified" processor had the same serial number as an unaltered promotional processor photo from another web site.
NEW FLASH: Overzealous Tech Sites Taken in by Paintshop Pro Forgery ...
Many Japanese tech sites are notorious for posting outrageously altered faux benchmark screens. This would appear to be another one of those posts ...
Multiprocessor Thunderbirds will rock, I am certain. But this is as ridiculous as the recently reported SMP Thunderbird Linux kernel compile that supposedly demonstrated a greater than two times increase in speed between one processor and two ... Yup.
Why couldn't I find people who believe things like this when I was selling electronics? I'd have made a fortune in commission. :)
Get off my virtual lawn, you damned virtual kids!