Slashdot Mirror
First Intel Quad Core Ready Desktop Mobo Spotted
MojoDog writes "Today Universal Abit launched their AW9D and
AW9D-MAX motherboards based on the Intel 975X chipset. There has been much
anticipation in the industry for this series and as far as looks go, these
boards are built to please. One interesting bullet point in the spec list
is that these boards are "Quad Core Ready", in line with a possible year-end
release of
Intel's Quad-Core Kentsfield CPU perhaps? Time will tell!"
no, these motherboards are for the LGA775 socket, not the LGA771 socket used by the Xeons in the Mac Pro or the Socket 479 used by the iMac/Mac notebooks/Mac mini.
"Clovertown" may be a quad-core version of the "Woodcrest" Xeon chips used in the Mac Pro, although I couldn't find anything definitive.
I'm not sure if Intel has any plans for quad-core mobile chips (the Core Duo used in the iMac/Mac Mini is the mobile-oriented chip, but has shown up in smaller desktop computers incl. the Macs).
How can a motherboard have all this stuff and leave out ECC? I would never buy a motherboard without ECC. Don't people want their machines to stay up more than a week at a time???
I've run multiple systems with non-ECC memmory. Uptime was originaly limited by time between brownouts/blackouts (~3 months). Then I got a UPS and uptimes have only gone up. If you need ECC to keep your system up for more than a week, you've got problems.
Fly me to the moon Let me sing among those stars Let me see what spring is like On jupiter and mars
Of course some people are always going to want to have the fastest game machine on the block. But seriously, it's amazing the kind of performance you can get these days with cheap, low-end hardware. Yesterday I built a machine for $300 with a 3 GHz P4 and 1 Gb of ram. (I reused a hard disk, so that cut the price a little.) Sure I could have built a dual-core system, but I would have ended up with a machine that cost many times more, used tons of power, and had almost identical performance except when I had two cpu-intensive processes running at once (i.e., almost never).
Find free books.
Clovertown and Kentsfield are identical except for the socket and bus speeds they use. So, yes, Clovertown is the quad core version of Woodcrest (like Kentsfield, it's actually two dual-core processors on one package) and will work fine in the Mac Pro (unless Apple does something silly and uses firmware to block upgrades).
Give a man fire, and you warm him for the night. Set a man on fire, and you warm him for the rest of his life.
Agreed, but bear in mind that a single-bit error doesn't have to crash your system (or an app). In fact, it usually won't, because the amount of "critical" memory is very small relative to the total amount of RAM installed. Instead it will silently corrupt data. This could result in a momentary glitch in what's shown on the screen; it could result in an app delivering nonsensical results; or, far worse, it could result in bad data being written to disk or an app delivering subtly wrong results. Since all modern operating systems use all memory in your box for something (cache, usually) pretty much every single-bit error is going to screw something up.
I work with many ECC-using servers and there are typically one to five single-bit errors per month. Even though I understand the reasons for it, I am kind of bewildered that ECC isn't more common on high-end desktop systems. The RAM costs ~15% more, but gamers, for instance, are already willing to pay 50% markups (or more) for a 1% performance bump (if that). You could even market it as overclocker-friendly: the error checking will tell you when you're overclocking too high, and the error correcting will help you when you're right on the edge. It could also allow overclockers to identify DIMMs which can't keep up without the laborious process of "pull out a stick, run memtest overnight; put stick back in, pull out a different one, run memtest overnight; etc." (Or the worse one when DIMM has to be installed in pairs. Then you get the joy of testing every combination.)
I'm not ready to state that GA-965P-DQ6 is the "fist", but it's the first one I found out about (months ago). I've already got this MB running with a Core 2 Duo E6600, and it specifically states on the manufacturer "Ready for next generation Quad Core processor".
P roducts_Overview.aspx?ProductID=2295
http://www.gigabyte-usa.com/Products/Motherboard/
I vote to change the name of the article "Intel Quad Core Ready Desktop Mobo Spotted" to "Another Intel Quad Core Ready Desktop Mobo Spotted".
If you live in high altitude areas, ECC RAM is really necessary if you want reliable computing. Even with ECC, higher-end servers will "scrub" memory by periodically reading out every memory block to correct single-bit errors before they become double-bit errors.
The Virginia Apple Supercomputer initially used non-ECC nodes. Couldn't keep the machine up and they ended up selling off all of the original Xserve machines to get ECC Xserve's.
At Los Alamos altitude, the problem is even worse. It isn't uncommon for non-ECC workstations to crash every other day.
Cosmic rays can flip a bit, and have biological effects. Cosmic ray arrival rates are related to your altitude. Blacksburg VA may be only a thousand feet up, but Los Alamos is more like 8-9K--can't remember, exactly. But it's considerably higher than Santa Fe, which is about 7K.
Also, you may not be able to accept single-bit errors, even at the lower rate you'd experience at sea level. What's good enough for a Joe Gamer's PC in Denver (5K feet) may not be remotely good enough for mission-critical servers. Wall Street is at sea level, for the purposes of this discussion. But you can bet they don't always look at random bit-flips kindly.
As usual, how you spec a machine depends upon what you're planning to do with it. That even filters down into whatever spares you may stock for a server farm. Compare mfg. memory prices for large multi-CPU (think 8-64 CPUs, not some generic 4-way) servers running Solaris or HP-UX to memory available from the mass market vendors. Last time I did that, the mfg. memory was about 10X more expensive than something generic that would at least boot your system.
In that environment you need to be sure that the memory you're buying really is equivalent--not just that the machine will still boot. In that case, BTW, it turned out that there was equivalent memory available, without paying a huge vendor markup. It wasn't as cheap as the rock-bottom stuff (which would still have yielded a bootable machine) but it was cheap enough to justify buying a (much cheaper) machine just to stress test it before adding it to the ready spares bin for production systems. Sometimes a four hour support contract is still too slow, but you don't want to pay six figures for a hot backup system. Which also has to be under that expensive service contract.
If you have TB of memory to deal with, and mission-critical means minimize or eliminate flipped bits, the rules change considerably. It's a whole different world.
God only knows what gyrations the NSA must go through. Oh, wait, they just pay vendor rates, no matter the cost. My tax dollars at work...
What you do with a computer does not constitute the whole of computing.