Slashdot Mirror
AMD's Triple-Core Phenom X3 Processor Launched
MojoKid writes "AMD officially launched their triple-core processor offering today with the
introduction of the Phenom X3 8750. When AMD first announced plans to introduce tri-core processors
late last year, reaction to the news was mixed. Some felt that AMD was simply planning to pass off partially functional Phenom
X4 quad-core processors as triple-core products, making lemonade from lemons if you will. Others thought it was a good way for AMD to increase bottom line profits, getting more usable die from a wafer and mitigating yield loss. This is an age-old strategy in the semiconductor space and after all, the graphics guys have been selling GPUs with non-functional units for years. This full
performance review and
evaluation of the new AMD Phenom X3 8750 Tri-Core processor shows the CPU
scales well in a number of standard application benchmarks, in addition to
dropping in at a relatively competitive price point."
3 cores sounds "wrong" (it should be apower of 2, right?), but with 3 cores, you can connect each core to every other one on an internal bus much more easily than with 4 cores, since you need fewer busses, and they do not need to cross.
SJW n. One who posts facts.
Looks like AMD's marketing and sales dept isn't being very smart here, pricing them the way they are. X3 chips are $20 cheaper than X4, and $5 cheaper than 2.2 GHz X4s. And with those benchmarks they are definitely not competitive against intel's 2-core and 4-core offerings. Come on guys! If you don't let go of some of the margins and price them aggressively against Intel you're going to die.
A couple more reviews that aren't as, um, positive:
http://www.pcper.com/article.php?type=expert&aid=550&pid=2
http://www.techreport.com/articles.x/14606
http://www.bit-tech.net/hardware/2008/04/23/amd_phenom_x3_8750/1
Everyone already does that. That's one of the reasons that Celerons used to be so popular with the overclocker crowd. When Intel didn't have enough of one kind of Celeron but had too many of another, they would mark down the faster chips or disable some cache on a P3.
Due to yields, if you buy a slow processor there is a good chance that it is capable of running quite a bit faster. When you buy a top of the line processor, that's much less likely.
GPU makers have been known to do the same thing. I remember when you could flash a low end card (one of the GeForce 4s?) to be a more expensive one (more shaders) and you might end up with a working card (wasn't disabled due to errors, just to 'meet quota').
This is normal. If they didn't do this, people would have to buy the faster chips which would cause their price to drop.
Comment forecast: Bits of genius surrounded by a sea of mediocrity.
That may have happened, but usually when chips are marked down it's because they didn't perform within specs in the higher slot. The fact that they don't show obvious problems in the hands of an overclocker doesn't mean they didn't meet the maker's QC cutoffs.
What I'm listening to now on Pandora...
PS3 uses the CELL processor built with 8 cores and one is disabled, leaving you with 7 cores-one for the OS and 6 for games/apps. And it will boot and run a linux image, yellowdog, which is a ported centos. So there ya go, you can buy one if you want one. There's more exact specs at the links, that is a basic and probably sort of flawed summary.
AMD systems are already radically different from how PCs used to be constructed ten years ago. Memory controller integration (NUMA in a multi-socket configuration) and a non-shared front-side bus come to mind, as does the point-to-point bus used between the processor and the south bridge (HyperTransport).
Contrast with Intel's "solution" which involves two sets or north and south bridges. Hardly elegant, and fails to expose the NUMA properties that the north bridges mitigate between one another.
Once AMD gets the clockspeed bit tuned in, I expect Phenoms to hit the high-performance market like a bar of soap in a sock. HPC likes memory bandwidth, but they like low memory latency even more and that's where AMD has Intel by the goolies. (ever wonder why even Athlon X2s hold their own in game benchmarks? doesn't matter how many gigahertz there are in the chip, games have datasets far larger than that 6-meg L2 cache.)
I don't know, when they double the speed of the 333Mhz processors, all of the CPU manufacturers labeled their chips as 667 Mhz. So, there must be some thought given to the the Christian tech sector. Of course, it could just be that they thought there were more Christian CPU buyers than Satanic ones.
;-)
The speeds were in reality 333.33... and 666.66..., so simple rounding produces 333 and 667. Perhaps they were merely using better mathematics than when they named the 133 and 266.
The beauty of it (from an engineering point of view) is that every core has been designed with 3 HT links. One goes to the memory, and two connect to other cores. So really, in a four-core system, there is an additional latency because information needs two hops to reach all of the cores. Three cores is the max AMD can do while still keeping latency at its lowest.
AMD's cores (the compute engines inside a single chip package) are NOT connected by HT links. HT links are used for communication with devices OUTSIDE of the chip package, and run at a clockspeed much less than that of the core clock.
AMD's cores are connected by a full speed crossbar switch, much, MUCH faster than HT. Most people really don't get that HT is chip-to-chip or chip-to-chipset, and that AMD has a fullspeed crossbar in the die. To say it one more time: AMD's cores within the same chip are connected at full CPU speed, and every core is exactly two hops to another: core-to-switch-to-core.
I am a viral sig. Please help me spread.