Separate names with a comma.
Discussion in 'Mac Pro' started by jdm111, Nov 26, 2009.
I was browsing geekbench today and found this.
Please google "hyperthreading".
A true 16 core MP won't exist until 2011 at the absolute earliest.
According to the E5430 Harpertown CPU it is actually a 2008 MacPro3,1. The MacPro2,1 designation belongs to the 2007 model with x5365 Clovertown CPUs. Both machines had quad CPUs. Both are not Hyperthreaded. So one can safely assume that the report was not generated automatically. The core figure probably relates to the total number of cores of the machine.
Thats not physically possible since 8 core CPUs that fit a MP don't exist.
I said the machine and not the CPU.
Aren't we all disregarding that Gulftown is said to have six cores per chip, not eight?
I think that was in reference to the Sandy Bridge Mac Pro, which should at least have an option for 16 real cores. (although I suspect the lower end Sandy Bridge chips might have 6 cores, to improve yields. Intel mentioned something about doing that at one point, IIRC. Please don't quote me on that)
The 6 core variants = failed 8 core chips with nanosurgery applied to reduce the operational core count IIRC.
Hmm... Turn junk into income... Yes, I'd see that as a way to improve yields.
Thats been done for years. The PS3's CELL is actually an 8-core CPU with 1 intentionally killed to increase production yield (those with 1 dead core off the line aren't trash).
AMD does much the same thing; when you buy a Phenom X3 Triple Core CPU you're actually buying a failed quad core. To me, thats actually a bad thing. It shows that AMD's manufacturing quality is so low they have enough 1 dead core CPUs to launch an entire line of processors based on failure!
Hahahaha! Love this one! It would be like Harley Davidson actually making a line called the "Walk It" series. Bahaha...I hate Harley -.-'
Of course it has.
Actually, chips aren't totally consistent accross the entier wafer. Those on the outer rim are more apt to be less stable, and end up binned at lower clocks for example. Even the wafer itself can make a difference due to the location in the ingot it was struck from.
But as you increase the core count, the possibility of a core not working is higher. So they have the choice to toss them (recycle the material), and end up with reduced yeilds per wafer, or use micro/nanosurgery to mod them to a working part. As it happens, AMD's yields are quite good.
It's actually a very smart thing to do, as if it's not, the costs increase. This isn't new either.
Doesn't mean that their a cheap alternative to blades
I would like to say that this is not factual. The core is certainly operational - people have enabled it, but AMD was quick to clamp down on the problem (a motherboard manufacturer, IIRC)
It sounds a bit unreasonable when you consider that Intel has nor 8 core product with 32 nm technology. The six cores are IMO a result of the die area becoming available by the die shrink and staying with the 1366 socket. From 45 to 32 nm you simply do not get the place to fit another 4 cores and the memory bandwidth of the Nehalem design would not have been suitable for 8 cores.
Since Intel doesn't make any 8 core chips, your explanation seems unlikely.
Hasn't Intel done the two-die CPU before?
Indeed they did, but they still had to fit the resulting size into the space of the 1366 socket. And as I said Intel have no 32 nm octad in the plan, only the 45 nm Becton, which will also come in a much bigger LGA 1567 socket and have a completely different uncore design.
So from my point of view the theory isn't very plausible.
There have been speculations though that Gulftown will consist of three two core dies. I find this hard to believe but I guess it would be possible.
I was actually referring to the Beckton, not the Gulftown. Sorry about the confusion.
I recall an article some time back, but haven't relocated it with a search.