Catch up on stories from the past week (and beyond) at the Slashdot story archive


Forgot your password?

Pentium III review 31

Dennis "Thresh" Fong wrote in about a review of the as-yet unreleased Pentium III. As we know, the Pentium III (Katmai) introduces 71 Katmai New Instructions (KNI): floating point SIMD instructions. While 3D-Now! and MMX cannot be used simultaneously to x87 FPU instructions, requiring the use of a state change instruction, apparently KNI will avoid this performance penalty by using a new processor mode. If you remember MMX was made multitasking independent (no specific MMX state to save/restore when switching between processes) by piggy-backing onto the FP-registers whose state is saved by all x86 multitasking OS's. Since the new mode is only activated if the OS knows about it, this is a safe way for Intel to improve performance.
Hemos:This is strange-I posted this before, but it seems to have vanished-can't figure out why. We're looking into it.
This discussion has been archived. No new comments can be posted.

Pentium III review

Comments Filter:
  • by Anonymous Coward
    Doesn't the x86 have enough friggin instruction already? I hope AMD doesn't play Intel's game, and ignores these extra warts.
  • Thresh of Doom/Quake asskicking fame reads Slashdot? Cool! :)
  • This story was up, got nearly 100 comments, and then vanished from Slashdot. Now it's back. Why?
  • Was talking to a games developer on Saturday, who had access to both a P3 and a K7. He wouldn't say how much faster the K7 was, but left the impression that there is no contest, even comparing with games optimised for P3.

    It seems likely to me that Intel will only catch up with the K7 when they bring out Merced, but I wouldn't even bet on that! (yup - it would not surprise me at all if the top end K7s that are being produced when the Merced comes out will be faster than the Merced)

  • You ever used GIS? Especially ArcView, takes a hell of a lot of processor and is very vital to many businesses.

    But I do think you're right that for most uses other than games you don't need anything this powerful.
  • I am ignorant in the Intel CPU world. Although I have experience w/ MIPS and other simple RISC procs, I must be incapable of figuring out how MMX and what other Katmai instructions could aid in ``Central Processing.''

    Should not these instructions and capabilities sit outside the CPU in ancillary chips (video/what else)?

    How does this help in OS speed and capability? Is it that we can perform FP ops better amongst process context switches? I'm assuming graphical subsystems outside the OS.

  • MMX didn't do shit. Katmai new instructions won't either... I wish Intel would stop trying to build EVERYTHING into the cpu and let the companies that are experts at it, like TDFX or NVIDIA handle it via add on boards...
  • I must have never heard about this. . . from the sound of the article, it seems that a Celeron 366 is extremely overclockable. Was this a slip by Intel? It has to be really well known, because it got tremendous scores on all those tests. Anybody got more info. on this processor? Just curious.

  • Too many people seem to be looking at this from precisely the wrong angle. They're saying 'normal software doesn't use that much CPU, games can get the performance more easily from 3d cards, ergo we don't need P3/500 systems'.

    They're right.

    On the other hand, Intel has produced these chips. There will be people around who've *GOT* P3/500 systems. These are chips which can run KDE or similar with 10% of their power; you could run software-rendered Quake 2 as the WALLPAPER, and the user won't notice except that his wallpaper freezes up when he recalculates something big. What do *you* plan to do with them?

    At the moment, my machine is 80% idle at most times; if you looked only at instruction traces from the CPU, you'd guess I was a special-purpose device for checking Mersenne numbers for primality. Some of you would see the same result; some would be told they were running HLT instructions 90% of the time, some would be told they're using 20% of the cycles on a box which is basically doing RC5 trial encryptions.

    But Mersenne numbers, RC5 trial encryptions, and much more so HLT loops, are not directly useful to the man whose box is providing the cycles. Quake 2 wallpaper would be nice to watch, but not what I'd call useful. Speech recognition might need fair numbers of cycles, but again it'll usually need them while the user is speaking rather than all the time.

    It's embarrassing to have to ask the question 'what useful things can we do non-interactively with a dedicated P2/400', but we've got to that point now.

  • You want things on the CPU die.

    Really, you do.

    Why? Because, on the CPU die, a 128-bit 400MHz bus is ten lines of VHDL. It's 0.06 millimetres wide. You can have dozens of them lying around. Off the CPU die, it's 128 IO pins - and the CPU's only got space for 600 or so such - and 128 PCB traces which have to be at least 0.5mm apart; it's a *big* orange ribbon on your motherboard.

    What you have at the moment between your CPU and your graphics card is a 66MHz 64-bit bus. If the graphics card was on the special-purpose PCB inside the P3/500, you'd have a 250MHz 64-bit bus. If it were on the CPU die, you could have as many 500MHz 128-bit busses as you pleased.

    I *want* integration. OK, I can't upgrade the graphics and the CPU separately - but you don't do that often anyway. You have big upgrades where you yank the motherboard and most of the cards, and replace the whole pile. Why not just replace the CPU?

    Yes, Intel might not have the in-house expertise in gfx chip design - but nvidia would cost them less than $10^9, and that's a sum any bank would be pleased to lend them tomorrow.

    *Everything* belongs in the CPU MCM. Seymour Cray knew that in 1986.

  • I'll be amazed if the successor to the K7 (the K7's already taped out) doesn't deal with KNI; AMD isn't quite yet in a position to become incompatible with the way Intel's steering IA32. It's OK if some AMD stuff doesn't run on Intel; if any Intel stuff doesn't run on AMD then AMD gets horrendously bad publicity.

    Remember, AMD get to implement it after seeing how Intel did it; they can stick in four KNI pipelines where Intel had only one. Just for that reason I'd expect them always to be ahead of Intel's last major revision.
  • Does anyone else see Merced being renamed Pentium IV? Looks like the Pentium Line is turning into a Street Fighter.

    Pentium Pro
    Pentium MMX
    Pentium II
    Pentium III
  • by Snoox ( 11939 )
    What happened to this news, you put it here a couple of days ago and then got rid of it? I thought I had gone nuts, was seeing things when suddenly this item was gone, not even in the history. Oh well.. :)

  • Uhm... Thresh didn't write the review. I dont know what you are smoking...
  • I must be getting old-fashioned or something - but I still can't see the great benefits of overclocking. Especially bus overclocking, which stresses your complete system beyond its specifications. So you get a 10 percent CPU performance increase (but was CPU your bottleneck?). On the other hand, your system becomes more unstable. Personally, I'd rather have a slightly slower, but completely reliable system! On the other hand, I guess Windows typically crashes before your CPU/mainboards gets the chance...
    Note that I do see the reason those brain-dead Celerons are overclocked, as you seem to be able to get it to run 50 percent faster. But do people really notice less than 10 or 15 percent?
  • shtml

    That's the other one that got yanked after attracting around 100 comments.

  • granted that's without a fan but damn! something just isn't right when i can cook my jiffy pop over my processor.
  • The article last week was removed because I exposed the Intel conspiracy to destroy the Alpha.

    Just pretend this is your first time reading this, and nobody gets hurt ;-)
  • Intel has been using CISC technology for ages. This is the processor that can only do 1 instruction per cycle. This makes for a processor that can't handle big loads, and why you don't see many Intel systems as huge servers (and if they are they have multiple processors for that reason), and most people go with Sun Sparcs.

    The alternative to this was RISC a processor design that lets you process several instructions per cycle. This is what Mac's use, and of course, Macs do suck, but thats because of the OS. You stick Linux on any mac, and watch it fly. Not to metention that Mac's make good crackers, for things such as DES and RC5.

    At some point, Intel will figure out that CISC is old, and its out of its league, switch to RISC, and act like they created the damn thing.

    Another bad thing about the Pentium III is that is Un-Overclockable. One of the nice things about the older processors is with a few jumper settings, you could get PII's to run as high as 504mhz! And this is from a PII 333/300!

Happiness is twin floppies.