Slashdot Mirror
Intel Cancels 800 MHz Xeon
goingware writes: "This article at C|Net tells how Intel canceled plans to produce an 800 MHz Xeon. They had feedback from major OEMs telling them they wanted fewer speedbumps with larger incremental improvements. I think that's a positive step, actually. I know from doing performance analysis of software that simply making a speedbump to a processor doesn't win the end-user that much, it's mainly for marketing reasons. This is because the performance of real systems these days is limited so much by memory access times and other factors. It would be better if manufacturers concentrated on engineering improvements that would result in real performance gains rather than notching up the clock speed."
This seems to be a good thing for computer sellers, rather than computer buyers.
If the cpu speeds increase in less frequent steps, then the cpu prices will also decrease in less frequent steps (IMHO).
This seems more like an attempt by Intel to "throttle down" the market. Also it makes their marketing campaigns look better if they can demonstrate these artificially-contrived "big improvement steps". Which in turn leads to higher prices asked by Intel, for those big steps.
This seems to be a subtle money grab from the consumer...
That's because the 700 Mhz is a 100 Mhz bus CPU, while the 733 Mhz is on a 133 Mhz bus... so those two CPUs are for different platforms and are really two different product lines. The next big thing after the 700 is not 750, and after 733 it is 800.
Indy (as does Indigo2, both out in 1993) had a 400MB/s memory bandwidth and a 266MB/s gio-64 bandwidth. The R4k Indigo (circa 91 I think) had these same limits. So if an earlier post is right that the p3 has a 400MB/s bandwidth, sgi's had that about a decade ago.
The O2 (which you can get for under $1500 now) has a 2GB/s memory bandwidth, and the Octane has it's switch at around 1.6GB/s (memory bandwidth is dependent on cpu clock). Origin's memory bandwidth scales linarly with the number of nodes you have (each node having up to two processors and its own ram). So when the pentiumpro had just started coming out, with what intel pushed as its really fast l2 cache -> cpu bandwidth (1.2GB/s), there were computers from sgi with faster system memory (though more latency of course). Interesting.
In short instead of all them fast CHIPS, we need to work on fast SYSTEMS.
That means:
All these things means that with our current chipsets we have the potential for much faster computers...where are they ?
PS. That beer/mdma guy is like Barney, everybody wants him dead.
"Semper in excretum set alta variant"
Nowadays, we have ultra-fast x86 CPUs, but chipsets that hold them back. I used to have a 486 motherboard that did memory interleaving to speed up memory accesses. I suppose when we had 70ns SIMMS it was more important, as well as cheaper to implement extra memory busses due to the lower pin count on a SIMM compared to a DIMM.
Anyway, it would make sense for the current x86 chipsets to have interleaving, although with SDRAM burst reads, it might be difficult to get the timing right. Maybe that's what has prevented it in commodity chipsets? Otherwise, I suppose you could have 4 byte interleaving thus:
DIMM0:hell
DIMM1:o wo
DIMM2:rld!
DIMM3:!!!!
You would still need some really low-latency memory in-between the main memory and the processor, and I guess the cost is another barrier to use in commodity chipsets.
It's interesting to note that the Alpha architecture has up to 8 times the memory bandwidth at 5.2GB/s than Athlon at 600MB/s or 12 times the Pentium at 400MB/s (although, an Alpha machine of such can cost $13,000) - check out http://www.microway.com/products /ws/alpha_21264.html for more in-depth information.
Enjoy,
J
Mainframes don't just have faster memory - in fact, the DRAM used may even be slower. It's the overall I/O, the speed at which the mainframe talks to peripherals and storage, the speed and caching in the storage systems, and the ability of the mainframe to do all these things at full blast simultaneously. The speed advantage to a mainframe, as you indicate, isn't one of CPU power per se - it's the ability the mainframe has to walk and chew gum whilst simultaneously rubbing its tummy, so to speak.
The memory difference isn't just one of memory interleaving (many boards do that now), or the memory-side bus. All PC processors get their speed from outrageous multipliers - which accounts for a couple of things to today's systems:
1: The tight code loops of a lot of benchmarks operate mainly from cache - creating way-high scores.
2: There isn't that much difference between a 1 GHz processor and a 600 MHz processor in real-world usage. Some things will be faster, but many more virtually unaffected.
Mainframe buses don't have the bandwidth restrictions that PC buses have. And when you think about it, we have 10x multipliers on PC processors, but the bus has only improved 4x since the 486 and the glory days of the 33 MHz bus. Most servers need faster I/O buses, not faster processors. When 64-bit 2x PCI is commonplace (or something better), and the FSB hits 250 MHz, and the operating systems finally become worthy of all that horsepower is when the PC will really start to make a dent in the mainframe's world. Until then, there's a reason why a mainframe will cost you hundreds of thousands of dollars, and a PC server will cost (at most) tens of thousands. PC servers are neat, and they do a pretty good job at what they are designed for - but it ain't no mainframe.
- -Josh Turiel
-- Josh Turiel
"2. Do not eat iPod Shuffle."
All in favor, say "HI!"
The practice of releasing new chips with negligible differences in clock speeds every few months has really seemed stupid to me. The sort of people who have to have the latest and fastest processor aren't going to be impressed when their brand new expensive X Mhz processor is superseded by the even newer X+25 MHz chip in two months time, and many of them will probably wait.
And it must also be costing Intel money to keep releasing incremental upgrades, simply because they then have to lower the prices of the now slower chips to get them to sell in the face of their latest speed king.
This pissing contest between AMD and Intel hasn't done Intel any favours at all, and it's probably time for them to take stock of where they went wrong.
---
Jon E. Erikson
Jon Erikson, IT guru