PDA

View Full Version : SPEC results for top processors.


ddtlm
Oct 13, 2002, 03:07 PM
Here, this should take you to the story in question:

http://www.aceshardware.com/read_news.jsp?id=60000436

You should find a HTML table showing SPEC results for top-performing processors at this time. To summarize, the P4 (2.8ghz) takes #1 in the integer tests, the Athlon (2.25ghz) takes #2 in the integer tests, and the Alpha 21264C (1.25ghz) takes #3. In the floating tests, the Itanium II (1.0ghz) takes #1, the Power4 (1.3ghz) takes #2, and the Alpha 21264C (1.25ghz) again takes #3.

Other contestants include the Ultrasparcs from Sun and the PA-RISC chips from HP. Unfortunately Moto/Apple are not into testing with SPEC, probably because it doesn't make the G4 look good.

For those that don't know, SPEC is quite possibly the best cross-platform becnhmark out there. This does not mean it is perfect, but you know a lot of systems don't have Photoshop available so we've can't include it. :) Notice since lots of competing companies use SPEC, we can be confident that SPEC is not a tool of Intel and AMD designed to make Apple look bad (which is one opinion I have heard voiced somewhere).

dricci
Oct 13, 2002, 03:24 PM
From what I've read, SPEC isn't designed for the Power PC or AltiVec, making its results inacurate, hence the reason Apple/Moto choose not to participate.

ddtlm
Oct 13, 2002, 03:31 PM
dricci:

If SPEC is somehow unfair to PPC, then how do you explain the high ranking of the Power4, which is also essentially a PPC chip, and doesn't even have a vector unit to back it up!? Also, "vector units" such as SSE and SSE2 are used by Intel and AMD, so why cannot AltiVec be used?

I think the answer is what I suggested before: the G4 does poorly in SPEC.

dricci
Oct 13, 2002, 03:40 PM
The G4 does poorly in SPEC because SPEC is not designed to use AltiVec, at least last time I checked. The Power4 can produce more raw Power than the G4 without help from something like AltiVec, where the G4 relies on AltiVec enabled applications and tests to get it's performance. That's just how it was designed, and when used properly, it does pretty well.

It's just further proof that benchmarking is stupid and inaccurrate. If you want to get real "benchmarks," use each processor in a regualar task under normal conditions to see which reigns supreme.

ddtlm
Oct 13, 2002, 04:13 PM
dricci:

Gimmie a break. When the G4 looses its because the benchmark is "unfair" and when it wins it is because the G4 "is powerful."

It's just further proof that benchmarking is stupid and inaccurrate.
Oh, even better. When the G4 cannot produce sufficiently high number the entire idea of benchmarking becomes useless?

The G4 does poorly in SPEC because SPEC is not designed to use AltiVec
Thats ignorant rubbish. I've already told you that other processors make use of their SIMD isntructions fine, and all you can do is claim that the test was not designed for AltiVec? There are at least 5 or 6 different, totaly incompatible instruction sets represented in the results I linked... SPEC isn't "designed" for any of them, nor should it have been. Thats the whole idea behind a fair benchmark.

If you want to get real "benchmarks," use each processor in a regualar task under normal conditions to see which reigns supreme.
What a wuss-out way to avoid the issue.

beatle888
Oct 13, 2002, 04:19 PM
how come the Intel P4 comes so close to
the POWER4? I thought the POWER4 with
its 8 instruction sets per cycle could toast
the Intel chips even if the POWER4 has
a slower ghz rating?

nixd2001
Oct 13, 2002, 04:23 PM
Originally posted by dricci
If you want to get real "benchmarks," use each processor in a regualar task under normal conditions to see which reigns supreme.

Can you suggest something that can be quantified? Without such an ability, there's no meaningful comparison. Until alternative (or "better") quantified comparisons are available, the best anyone can do is compare such numbers.

dricci
Oct 13, 2002, 04:24 PM
Troll: Think whatever you want, but I don't believe in benchmarking, unless it involves actually using the device in normal, every day operations, not running caluclations through it.

All I am saying, is SPEC is not the best benchmark for Moto PPC chips, due to differences in how caluclations are preformed. I don't know exactly how it all works, but that's just the way it is. How else could some benchmarks say current G4s are up to par with the latest Pentium 4s or AMD Athlons? See, it's all on what you test, and how it's tested. Some processors are not good with certain benchmarks, and the G4 and SPEC are a good example of one of those.

I'm unsub'ing from this thread, I see it going nowhere fast. Everywhere you go seems to turn into a troll fest. You enjoy trying to start trouble here. Why? I don't know.. I'm sure you could find something more productive to do with your SPEC-passing P4s... so go do it.

ddtlm
Oct 13, 2002, 04:24 PM
beatle888:

The Power4 can issue 8 but only "retire" 5 per cycle, so it's max sustained rate is only 5 instructions per clock. Even then, like all other processors, the Power4 will usually be doing fewer than 5.

ddtlm
Oct 13, 2002, 04:29 PM
dricci:

Not only did I not really attack the G4 in my original post, but I had intended it to inform people as to the performance of the Power4 relative to other top chips. Note that the Power4 is doing well. It's a shame that you decided you needed to fight a loosing battle in the name of the G4. It's also a shame that you have hijacked this thread and turned it into something much less civilized than it was supposed to be.

beatle888
Oct 13, 2002, 04:33 PM
whats strange is that the INTEL P4 is doing
just fine against the POWER4 which is suppose
to knock the socks off of any current desktop
chip. this has me a little worried. why does
Intel always beat us?

ddtlm
Oct 13, 2002, 04:35 PM
beatle888:

Well the Power4 is at "only" 1.3ghz and has been there for something like a year, so faster versions are imminent. The current Power4 is a server chip so ultimate performance is not a big deal, and reliability is job #1. IBM is very very conservative with it.

If a "Power4 Lite" comes out a something like 2.0ghz and has an AltiVec unit, it should do very well. There is no gaurentee that it will be the best, but it will be a lot better than where we are now.

nixd2001
Oct 13, 2002, 04:57 PM
Originally posted by beatle888
whats strange is that the INTEL P4 is doing
just fine against the POWER4 which is suppose
to knock the socks off of any current desktop
chip. this has me a little worried. why does
Intel always beat us?

IBM put a lot of effort into not only having large memory bandwidth but also actually using it. This sort of characteristic is probably not going to show up in SPEC ratings. TPC etc might be more capable of benefiting from this.

Oh, and remember that this is for a CPU core. Each die contains two cores and each multi chip module (MCM) contains 4 dies, so you've got 8 cores in a nice "convenient" unit if you've got the money to spend :p

nixd2001
Oct 13, 2002, 05:05 PM
Originally posted by dricci
Troll: Think whatever you want, but I don't believe in benchmarking, unless it involves actually using the device in normal, every day operations, not running caluclations through it.

All I am saying, is SPEC is not the best benchmark for Moto PPC chips, due to differences in how caluclations are preformed. I don't know exactly how it all works, but that's just the way it is. How else could some benchmarks say current G4s are up to par with the latest Pentium 4s or AMD Athlons? See, it's all on what you test, and how it's tested. Some processors are not good with certain benchmarks, and the G4 and SPEC are a good example of one of those.

I'm unsub'ing from this thread, I see it going nowhere fast. Everywhere you go seems to turn into a troll fest. You enjoy trying to start trouble here. Why? I don't know.. I'm sure you could find something more productive to do with your SPEC-passing P4s... so go do it.

Chill out dude!

You're not the only one to raise the idea of "troll fests". What's actually happening is that people differ and some people keep asking questions and sometimes threads develop into discussions not every wants.

I wonder whether the concept of trolls exists in academic threads? Maybe someone from the 'brane theory of everything pops over to the superstring theory of everything rumors board and suggests 'branes might be a useful concept - and then gets trashed for being a troll? Probably not....

ddtlm
Oct 13, 2002, 06:24 PM
nixd2001:

Each die contains two cores and each multi chip module (MCM) contains 4 dies, so you've got 8 cores
However the "Power4 Lite" is almost certainly going to only have a single core per die in order to bring the current extremly high prices down to within the reach of "desktop" that IBM says they are targeting.

Of course the "Power4 Lite" will also sport something a lot like AltiVec if not AltiVec itself, and that will boost performance by a lot in certain areas.

alex_ant
Oct 13, 2002, 07:56 PM
Originally posted by dricci
I don't know exactly how it all works, but that's just the way it is.
Best quote ever!

bousozoku
Oct 13, 2002, 08:53 PM
If the truth be known, some compilers automatically generate vector instructions and others don't. GCC does not automatically generate any vector instructions as of 3.0, according to what I've read. Perhaps, if it did, the G4 would perform far enough up the list that it would only be remotely far away from the top ten. This would also help create games and 3D modeling with much better performance.

Perhaps, if Apple (or Metrowerks) would go further...

ddtlm
Oct 13, 2002, 09:55 PM
bousozoku:

I think that the GCC 3.x branch may generate some AltiVec, and it will certainly get better at it as time passes. I'm sure it's not as good as Intel's compiler, but I wasn't about to tell my little G4 defending buddy that. :)

As it is, I would expect the 1.25ghz G4 to land integer scores near the PA-RISC and UltraScarc, based on crude extrapolations based on now out of date tests done by c't (I think). Not real reliable. The biggest weakness of the G4 is really in floating point, and although AltiVec could boost some of that (perhaps a lot), I am pretty sure that some of the SPEC float tests are double precision which would still be bad for the G4, regardless of the compiler.

But fear not, we are almost assurred that there are chips better than the G4 on the way.