Slashdot Mirror
Intel Releasing PIII Xeon Today
BMIComp writes "Yahoo! news is reporting that Intel is going to introduce their Pentium III Xeon Chip, today. " .18 microns, 700 Mhz, and integrated cache. The article talks quite a bit about how the new Xeons are going straight for Sun's throat.
As far as I can tell, the most the Xeon can scale to is 8 processors.
And it doesn't do that well. Intel's 8 way chipset is crap. To grossly oversimplify it's two 4 way shared busses communicating over yet another shared bus. It performs about how you'd expect. Compaq has a better 8 way xeon chipset, but if they're not your hardware vendor, you're SOL.
Somewhere Scot McNeally is screaming "Help, I'm being menaced by a dwarf!"
Or not.
--Shoeboy
(former microserf)
Sun's architecture is based on high-speed switches connecting multiple fast/wide PCI or Sbus buses. The peecee architecture is based on having one to four PCI buses on another shared bus. Suns have memory buses no less than 1152 bits wide. Peecees generally have 256-bit memory buses. Summary: Suns can move far, far more bits around inside them than any peecee server can. Sun 4, Intel 0. It should be noted that this isn't entirely Intel's fault, but they aren't really helping anyone do something non-peeceeish. You _could_ build a Sun-like system with Xeons, but nobody does.
The peecee architecture is hopelessly obsolete. 16 bit bootstrap code. A BIOS. Forget boot monitors and intelligent peripherals, you know, things that make systems manageable and flexible (just try booting a peecee from tape - you can't - or over the network - that'll cost you extra). Both system architectures have their roots in the late 70s and early 80s. Sun started with a good design and have steadily improved and enhanced it. The peecee started with a poor design and have left it essentially unchanged for 20 years. Sun 5, Intel 0. Again, not all Intel's fault, but since there aren't really any other systems available using Intel CPUs, it counts against them anyway.
Now, even with these drawbacks, I could see the Xeons being a good choice for a midrange server if they cost much less than the Suns. Unfortunately, the Xeons are actually more expensive than the UltraSparcs, even at the same clock and with the same size L2 cache. If you have $4k to blow, you can get either a Xeon 500 w/2MB, or an UltraSparc 450 with 4MB. And the UltraSparc is faster, too. Oops, guess that's 3. Sun 6, Intel 0.
But it's not necessarily the truth.
The truth is that for a whole lot of the applications for which people are installing big UltraSPARC boxes, the applications are not CPU-bound, but rather I/O bound.
Having a better, faster Xeon Pentium III processor doesn't help with either of these things.
What helps with such applications are:
Which Seagate and Quantum and IBM are responsible for...
Think Adaptec, maybehaps?
Which means more than the 8GB that appears to be supported at this point.
If you're not part of the solution, you're part of the precipitate.
The thing about CISC is that they have a habit of using microcode to translate all of the complex instructions into smaller ones for the core of the processor (AFAIK). The time it takes to decode instuctions is considerable at this stage, sith several hundred instuctions.
Trust me on this one - it's at most one clock, and possibly less. This goes for the x86 instruction set especially; each byte of the variable-length instruction has a fixed purpose, instead of being completely random. To process this, you need to prefetch several bytes ahead and have three or four shifters and a MUX to select and align the next instruction while you're processing the current one, giving a throughput of one instruction fetched and aligned per clock (or more, if you add more silicon). Once the instruction is aligned, you read out each of the fields that might possibly be there, and use a MUX to select them. You have a combinational logic block that processes the opcode and tells you if this is an instruction that can be processed atomically or that needs to be broken down, and a lookup table giving a series of RISCian stages for multi-clock instructions. If the "multi-clock" flag is set, you stall the instruction fetch unit and route in instructions from the lookup table instead.
It should be noted that processing the argument-location byte may insert memory load/store instructions too. However, as these are very predictable, you don't need a table lookup for them (just stalls and instrucion register preloads at appropriate times).
Lots of extra silicon, but little extra time. You do much the same thing in a RISC processor, except without the alignment stage or the lookup table.
As the article says, Pentium III Xeons have been around for a while, but were topped out at 550 Mhz. This is just faster (and different layout, I think).
I cannot give you a whole lot of technical detail, but practical data I have. I have a SuperMicro S6DGU dual Xeon board on which I run two 500-MHz 512k units. I have compared these to a similar SuperMicro dual P3 board, and did some time comparisons in both SCO UNIX and in WinNT 4. In both cases, the Xeon was between 10 and 25% faster at the same clock/processor speed depending on the application.
At the time, I did not test multiprocessors, but I can only suppose that the margin would widen due to better SMP support within the Xeon/GX chipset. Also notable is the difference in chipsets between the BX and GX I tested with. The GX may make my setup a little faster with memory interleaving, more efficient bus arbitration, etc.
What remains to be seen is whether the cost difference justifies the performance difference for small servers, workstations, and hobbyist users. Can anyone kick in their deep technical knowledge of these chips?
-L
Intel might claim that they are going for the throat with the PIII Xeon, but they have a huge gap to close before marketing catches up with reality.
...same as a standard PIII.
As far as I can tell, the most the Xeon can scale to is 8 processors. At least that is the largest machine I can find for Xeon. Sun's midrange machines _start_ at 4 processors and, for now, go up to 64. The planned maximum for UltraSparc III is 1024. Sun 1, Intel 0.
UltraSparc is 64 bit, Xeon is 32.
Sun 2, Intel 0
UltraSparcs have had integrated caches for quite a while, as far back as SuperSparc I. On the current cpus, the desktop boxes can have up 2MB cache, the midrange servers 4MB, and the large Enterprise servers get 8MB cache. The PIII Xeon's claim to fame is the extra electronics they add to the processor card which allows some hardware admin of the processor. The cache size is 256KB
Sun 3, Intel 0.
Until Intel produces a cpu which can scale well and has an OS which supports that scaling, I don't think Sun has much to worry about.
And for those of you who think MHz is a consideration, bzzzt! In the world of 24x7 server farms, torque counts more than horsepower. The UltraSparc cpu can handle a much greater load without sweat than most other cpus. The UltraSparc was designed to handle massive loads. So, while the UltraSparc will likely lag behind an equivalently rated Intel cpu, which would you rather use to make a cross country houshold move: A Kenworth or a few dozen pick-up trucks?
In the immortal words of Socrates, who said; 'I drank what?'