Follow Slashdot stories on Twitter


Forgot your password?

K7 Benchmarking 57

Quite a number of people have written in with the word about more specs on the K7, and its performance versus the PIII. Here's a little teaser: the spec K7 FPU performance is 40% faster then the PII. Check out Ace's Hardware for more information.
This discussion has been archived. No new comments can be posted.

K7 Benchmarking

Comments Filter:
  • Do Intel (or even Alpha) CPUs encode mpeg video in realtime? On my K6-2/400, Windows takes like 7 minutes to cram a 2 gig AVI file into a 50mb MPEG1. It's not an incredibly long time or anything.
  • Posted by DanTucny:

    A recording made of the dinner presentation on the K7 where these latest performance figures were announced is available from JC's ( []), it is in Real Audio format and has a few mirrors available...
    Of course, you've probably already been there and heard it :)
  • Can the K7 compress 30 jpeg images a second at 640x480? Can the K7 encode video to MPEG in realtime? How long does the K7 take per setiathome block? How long does bzip2 take to compress a 200 meg file on the K7?
  • The P3-450 can handle real-time MPEG-2 DVD quality compression if the application supports SSE, and the higher clock-speed versions can handle it without SSE.
  • Looks slashdotted already. If you like me can't get through, try JC's page [] for more K7 info and rumours.
  • Digital looked at MMX and decided they didn't really need that stuff since they were already fast enough at that sort of thing. So for the 21264 they did the MVI Motion Video Instructions instead, which are supposed to speed up MPEG compression. Don't know of any benchmarks though,
  • Jpeg is mostly floating point,

    Rubbish! The IJG library which is part of SPECint95 is integer-only and that is the software that everyone uses, since it has a very liberal license.

  • Actually, JPEG compression is already part of the SPEC suite [] so when the real SPEC results are released you just have to look at the breakdown. Another of the SPEC marks is a run of gcc (v 1.38 I think).

    Unfortunately the SPEC marks are never compiled with gcc because it isn't as fast as Intel's compilers, which I presume AMD will use :-). I hope this list [] will be updated when the K7 is out, since it is probably a good indicator of Linux integer performance.

  • Now, I've gotta go dry the drool out of my keyboard...
  • The K7 will be released in 500, 550, and 600 MHz speed grades initially /w 1/2 speed L2 cache (off-die of course). There is support for slower L2 (1/3) and faster (full), but reportedly these will be the "low end" and "high end" versions of the chips designed to compete with the Celeron and Xeon respectively.
  • by cdipierr ( 4045 ) on Friday June 11, 1999 @04:35AM (#1855722) Homepage
    Since it's slashdotted, I'll post some more correct info here.

    According to Ace's page a 550 MHz K7 /w 512kb L2 cache running at 1/2 speed is compared to a 550 MHz PIII Xeon (not sure the cache size) /w SSE enhancements. Using that, the results are as follows:

    SpecFP - 36% faster
    SpecInt - 6% faster

    The 600 MHz K7 is of course faster, turning in something like 43% and 15% respectively, but a fair comparison is of like speeds.
  • For the PIII Xeon cache was reported as 512K.
  • AMD claims that the K7 will be released initially at 500, 550 and 600 MHz. There have been reports that they will be able to reach 700 MHz on the current 0.25 process, and at least 1 GHz on the newer 0.18 process which they are adopting later this year. They plan to keep about 50 MHz faster than the top-of-the-line Intel offerings.

    If these spec results are to be believed, this could be a definite Xeon killer once SMP systems are available. Gee...a processor that outperforms a Xeon, for oh about a tenth of the price. Hard decision there...
  • The 200 MHz bus speed is the bus going to/from the processor(s) to the MMU, PCI bus, etc. The memory will still operate at 100/133 MHz for the first systems.

    This 200 MHz bus will be really shine in SMP. It utilizes a point-to-point design that is more scalable than Intel's SMP design which uses one bus (at 100 MHz on current systems) with dedicated lines for each processor.
  • IIRC, the Xeon versions of Pentium II/III have full-speed L2's.
    Christopher A. Bohn
  • Moreover, five new DSP extensions were added, allowing the K7 to more efficiently decode multimedia files like MP3 audio files, or do software processing of communications algorithms like ADSL, Meyer said."


    Um yeah, what special thing do you need to do for ADSL, unless you have some really bogus implementation? Most of them go with just a straight Ethernet connection, so the processor has nothing to do with it, save for interpreting the data that comes in over the NIC.
  • Consider the CPUs, though, too. I mean, a K6-3/450 is roughly equivalent in Mhz and L2 to a P2 Xeon 450. That doesn't make it suitable for the same tasks ... same goes for K7 & P3. If the K7 at equivalent clock speeds is really 6% faster integer and 35% faster fp than a *Xeon*, then I'd be willing to pay about the same for a K7 500 as I would for a P3 (non-Xeon, now) 550 or 600 -- although I think AMD will price them somewhat lower than that, given their history. But that assumes that spec95 benchmarks really represent real life, and we all know that there's no perfect benchmark. (Spec95 ain't bad, but ...)
  • You know, I don't think we're going to get those kind of benchmarks until the K7 actually ships. (Hey AMD! Free publicity if you send me a K7! :) ANYway, barring that event, here's a way to estimate those performance metrics: base them on the P3 500, which someone around here just might have floating around. Conservatively, here's what I would guess, based on K7 being 6% faster integer and 35% faster than a *Xeon* at equivalent clocks:
    • Jpeg is mostly floating point, with some integer thrown in for good measure. I would say that a K7 500 can do ... what, about 20% faster than a P3 500 on this? (Considering the K7 500 is almost 50% faster than a P3 500 in specfp95.)
    • MPEG encoding and seti@home are similar, except that SETI@home is a double-precision floating-point beast. As I understand it, the K7 is pipelined on double-precision, whereas the P3 is only partially pipelined. So, I would guess the K7 will wax the P3 at SETI@home, but I can't even guess by how much. MPEG encoding should (somewhat like JPEG) be about 20% faster, as a conservative estimate.
    • bzip2 is primarily an integer program, so it might be 10% faster in the processor ... but as pointed out by someone else, compressing a 200M file is more of an I/O test than a processor test. :) [NB: the 128K L1 cache might make a big difference here. bzip2 definitely won't just fit into L1 on a K7, but if it has good cache locality then that could really help it here.]

    You know the drill ... your mileage may vary. I'm basically talking off the top of my head, but these should be educated guesses. One thing's for certain, we won't know any real numbers until someone gets their paws on a K7 system loaded with Linux and actually times these tests.

  • In a word: it won't suffer much. The optimizations for the two processors, in the sense of FPU pipelining issues, are almost identical.

    The main thing "Intel optimizations" do is put FADDs (add) between FMULs (multiply), because you have to have an FADD after an FMUL (maybe two FADDs? I forget) to get reasonable FPU throughput on the P6-series of CPUs (Ppro, P2, P3, Celeron). This is because the two pipelines of the P6s can do an FADD and and FMUL in parallel ... but can't do two FADDs simultaneously (for instance). The K7 structure is very similar, except for a few things: 1) fewer restrictions on what can execute simultaneously. 2) lower latency on complex (FDIV and FSQRT) instructions, and 3) FMUL (and FADD of course, in the other pipeline) can execute partially in parallel with an FDIV. All these mean that anything "Intel optimized" (for the P6) will run great on a K7. In fact, anything optimized for the P6 is already 75% optimized for a K7 ... the difference is that the ordering of FDIV relative to FMULs and FADDs is important in the K7, allowing for some further tuning if that's important to you. :)
  • I stand corrected. I was going on "Jpeg in general involves lots of DCTs", and I think of those as floating-point. :)
  • great news, I just read over the site and apperantly (if this information is legit) the K7 is a kick-butt chip and well worth the wait and quite possible all the hype its been recieving. but one question still lingers....
    there are rumors going around from as low as 200 all the way up to 900 dollars, I want to know exactly the price range for this chip. AMD says that it is aiming this chip towards High end server buisness, what like a XEON I say? some xeon's sell for more than 3000 dollars, how much is this chip going to cost, can I possible afford one for my system?
  • The Setiathome is mostly single precession, there for you can probably use the 3D Now unit to actually double it's performance.

  • I think a lot of your "real-world" questions
    pertain more to the disk I/O performance
    than chip performance... I don't think those
    are necessarily good tests for a CPU...
    You need some hard-core computationally
    intensive tests... The SETI@home one was
    a great suggestion. I remember a few years
    ago using those encryption-breaking
    blocks as a great benchmark (RC5, etc).

  • by Ellis-D ( 19919 ) on Friday June 11, 1999 @04:45AM (#1855738) Homepage
    Well it appears that the info we had was correct as AMD did show a presentation including bencmarks at the dinner it hosted tonight. I got a little snippet of info from a usenet posting that JC posted and I thought should be posted here as well. Check it out:
    I've just returned the dinner meeting at which Dirk Meyer (VP of Eng.
    AMD) had a presentation. My first impression is that K7 looks very promising.
    Mr. Meyer told us that AMD was indeed announcing K7 this month (June
    '99) at 500, 550 and 600 Mhz. It has 22 Million transistors on a 184
    mm square die at .25 micron process.

    The first release of K7 will have 512K of L2 cache at half-speed.
    At 600 MHz, K7 is %115 faster in SpecInt95 than a PIII Xeon 550Mhz
    with 512K full speed cache. At 550 K7 is %106 faster in SpecInt.
    At 600MHz, K7 is %143 faster than the same PIII Xeon at SpecFPBase.
    At 550Mhz, K7 is %136 faster (these numbers are interpolated visually
    from a slide which means 143 was closer to 140 than 150).
    At 3D Winbench 99 V1.2 (null driver) on win98, K7 600 is at %146
    faster than the said PIII Xeon using SSE optimizations.

    There are new Integer SIMD instructions, DSP type instructions for
    MP3, AC (audio) etc. and cache prefetch instructions. Microsoft will
    support 3DNow in an upcoming Visual Studio release.

    At initial launch, there won't be any MP systems. All motherboards
    (from Asus, Biostar, Gigabyte, FIC and one other I couldn't catch)
    will use AMD chipset. Via, ALI and SIS are designing their own
    chipsets to be released before the end of this year.

    These are most of my notes during the one hour presentation and Q&A
    afterwards. All errors are my own. I speak only for myself etc., etc.
    What a great day for x86! hehe at least if you arn't Intel. This post was posted at 12:50 AM on June 11th.

    I ate my tag line.
  • It's supposed to be able to handle up to 8 processors. Unlike Intels quad arrangements :)
  • >Um yeah, what special thing do you need to do for ADSL, unless you have some really bogus implementation? Most of them go with just a straight Ethernet connection, so the processor has nothing to do with it, save for interpreting the data that comes in over the NIC.

    There are three major ways to deploy DSL or broadband or any other new communications medium:
    1. Use an Ethernet half-bridge. This is a common solution, and the one you obviously have in mind, but some would say it's pretty bogus.
    2. Use a dedicated DSL/whatever card. This is The Right Thing in an abstract sense. In the real world, Ethernet cards (I hate the term "NIC" which was popularized mostly by MS) are produced in such great quantity and are therefore so cheap that it's hard for a dedicated card to be price-competitive.
    3. Use a WinModem-like card that relies on the CPU to do half of the work. This is obviously what the original author had in mind, and is truly bogus.

    So, in other words, your original question is a very good one, but your "background information" doesn't really account for all of the possibilities very well.
  • The K7 will be released in 500, 550, and 600MHz variants. This has been heavily hinted since November, and was confirmed by the CEO of AMD (Jerry Sanders) himself at an annual shareholders meeting (I think that's when it was).

    The L2 cache of the K7 will be a half the clock of the processor. The 1/3x MHz idea was put together because AMD wasn't certain that the SRAM market would be able to supply 300MHz SRAMS for the K7-600's L2. Thankfully, this is not a problem.

    Incidentally, Kryotech's Super-G will be out this year, likely at 1GHz in Q4, with a hypercooled K7. It *will* be expensive, but it will be *worth it*. AMD will have two 180nm processes ready by Q4, which will make the K7 a lot cheaper to make and a lot more voluminous (eg: there will be more of them). Figure that you might see an 800MHz K7 by end of year if AMD deems it necessary, that's one great core for MHz!

    PC News'n'Links []

    PS: K7 and mP6 look to be the fastest current cores for rc5, per MHz. They may both be faster than the mighty K5, once optimized for.
  • "PS: K7 and mP6 look to be the fastest current cores for rc5, per MHz."

    Heh...I mean x86 cores, of course. ;)

  • Is the 450MHz PA-8500 out? I know one of the cardinal rules of spec95 is never compare MHz-to-MHz, since one of the architectural tradeoffs you make for higher performance is necessarily lower clock sometimes.

    Apologies...I have not kept up well with HP's offerings. :(

  • by JCholewa ( 34629 ) on Friday June 11, 1999 @06:30AM (#1855744) Homepage
    The K7 is going to be priced comparatively to the Pentium III, not the Pentium III Xeon, from what I've been told. The estimates among my local group are:

    $400 or slightly above for 500MHz

    $550-ish for 550MHz

    $700 or so for the 600MHz version, though they may want a more respectable (eg: high) premium for the fastest x86 process of all time

    These prices are slightly higher, mostly, than our extrapolations of PIII pricing around late July, where K7 will start to pick up volume. Despite the performance delta, AMD will likely make the part available to high end consumers in pricing, plus they want to pummel down Intel's high end ASP so they choke on their own Celerons.

    AMD's DDR L2 "Viper" version of the K7, in Slot-B, will compete against Xeon. It will also happen to destroy Xeon in spec -- even more utterly than regular K7 does. Cascades looks like it'll be toasted a bit, too, unless Intel puts up a surprise and gives it 1MB L1 on-die.

    BTW: K7's integer score beats out HP's mighty PA-8500 (which has 1.5MB L1 on die), I'm told. It may be the 2nd or third highest specint95 core out there.

    Also, K7 kicks ass at rc5 -- pass it along!

    PC News'n'Links []
  • And more facts as they come to hand as all other sites seem to be suffering /. effect...

    K7 apparently will be released at a clock speed of at least 500 MHz. However, there is speculation, for a number of reasons, that it might be released as high as 600 MHz. The first justification I've seen cited for this is that the K7's L2 cache is planned to run at an initial speed of 1/3 of the processor's speed, and when the bus will be at 200 MHz, it is illogical for the L2 cache to have a slower clock speed than the FSB. At this stage however there is no definite evidence to say that the K7 will be released above 500 MHz.

    An alt. site to check out for up to speed info is CMP net []. They carry an article [albeit a little dateded] article on the specs from a competitive Intel perspective. Makes interesting reading.
  • Ok, so AMD claims the AMD 550 will be faster than the Xeon 550, but how about Floating Point ? I mean, I am a 3D Artist, and all my CPU time is spent rendering, how does the new K7 compares aghinst the Xeon (or PIII) on FPU ? AMD has been known to suck in such tests, I was planning in getting a Dual Xeon, shuld I whait for the dual K7 (anybody knows if it's quad possible ?)


    P.S. Sorry for the bad speeling, english is not my native language

  • Since its a whole new architecture - for i86 people at least - involved it would be very interesting to see computation on large data, so RAM I/O speed could be tested by this alongside CPU performance. PCI-performance would also be interesting, supposedly theyd have to work around some of Intel patents.

    I cant wait for this real competition to Intel.

  • I concure. Price is the main factor to myself and a few million others. How much punch can I get for the buck. IF I have to spend 50% more for the same Mhz and L2, that contest is already over.

  • To all of those: Probably. And Quickly.
  • According to the blurb on the front page, the FPU is abou 40% faster than the Xeon. I don't know if that's true, but I sure hope it is! (I don't have a K7 on my desk so I can't say for sure...but if someone generous wanted to send me one.....I'd let you know ASAP) ;)
  • On paper K7's FPU looks VERY promising, and
    the first benchmarks seem to indicate the same.
    Let's hope it will not suffer too much when
    running code optimized for Intel's range of
  • Okay, I'm only guessing, but this could be caused
    by the difference between the full speed level-2
    cache of K6-III vs the 1/2 speed one of K7.
  • Man, I hope that NEVER happens... I hate WinModems, and WinxDSL cards would just suck.

    Of course, most of us (I hope!) are running some form of *nix operating system, so we won't want them anyways! More processing time for RC5! ;-)

  • The K7 used in the benchmark against K6-3 wasn't a K7 Revision C.

    You will expect a very much better benchmark for K7 with Revision C.

"To take a significant step forward, you must make a series of finite improvements." -- Donald J. Atwood, General Motors