[KT]

Intel is trying to discredit Apple's new G5, using the same old lame "evidence" used elsewhere on the web.

From Intel's own "Write to Know" column, an employee asked the following:
Q: When I heard Steve Jobs on the news recently claim that the PowerMac G5 is the world’s fastest personal computer, I was skeptical. But now I have seen the specifications, and it appears that he is amply justified in his claim. With music and video applications driving the need for performance these days, Apple seems to have scored a home run. Do we have anything on our roadmap that will catch up to them anytime soon? Are we now one generation behind?

A: The Apple/Steve Jobs announcement received a lot of media coverage. Initial reports touted Apple’s claims as you noted. But more detailed analysis reveals significant embellishment. For example, the frequency is a claimed 4 GHz, but that results from counting two processors, each running at half that speed. Additionally, the benchmarks used are not current, and the compliers used were not appropriate. The Dell system used for comparison had certain performance features, such as hyperthreading, which were disabled.

Most technical Web sites have now analyzed and discounted Apple’s claims.
Bill Siu, vice president and general manager
Desktop Platforms Group

[redcalx]

Correct me if I am wrong but when the Intel dude claims that the compilers were not appropriate, isnt that in the G5's favor?

I was under the impression that the tests run were with the G5 running 32 bit applications rather than 64.

I could be wrong though.

[KT]

Originally posted by redcalx:
Correct me if I am wrong but when the Intel dude claims that the compilers were not appropriate, isnt that in the G5's favor?

I was under the impression that the tests run were with the G5 running 32 bit applications rather than 64.

I could be wrong though.

This article should set things to rest:
http://www.applelust.com/oped/amc/ar.../a030718.shtml

[CubeBoy]

Hannibal recently interviewed the PPC970s chief architect, compiler expert, and a PR guy, it seems that the version of GCC used by Apple during the benchmark was already more optimized for the G5. I don't see why everybody is pointing that x86 in general is more optimized for GCC when we're talking specifically about the Pentium 4 which has very few optimizations and some of them actually hurt the Pentium 4 in performance. (eg insufficient registers for march=Pentium 4).

From the interview:

"As it turns out, they were in fact wrong before WWDC, but the version of gcc that Apple was using for WWDC had the correct values in it. (I'd imagine that this pre-WWDC obfuscation of the 970's vector latencies was done deliberately, but I forgot to ask him about that."

"The gcc scheduler is not really designed ideally for a processor like the 970 and the Power4 and others, and that's a lot of what the IBM and Apple teams have worked on"

It should be noted that GCC is not able to schedule for the Pentium 4 at all (but it can for nearly every other x86 cpu and the PPC970), having well scheduled code is really quite important for most floating point code.

Also, regarding hyperthreading, here are two official SPEC submissions, one by Dell, the other by Intel, both machines used the same compiler(s), processor, and are equipped very similarly. The Intel submission had hyperthreading turned on, the Dell had hyperthreading off.

Dell SPECint Rate Base: 12.6
Dell SPECfp Rate Base: 12.7

Intel SPECint Rate Base: 13.8
Intel SPECfp Rate Base: 13.6

All in all, a nice 10% boost in SPECint Rate and 7% boost in SPECfp Rate with hyperthreading turned on.

[CubeBoy]

Oh yeah, here's the links for the SPEC submissions:

Dell P4 SPECint rate
http://www.spec.org/cpu2000/results/...1113-01829.asc

Dell P4 SPECfp rate
http://www.spec.org/cpu2000/results/...1113-01828.asc

Intel P4 SPECint rate
http://www.spec.org/cpu2000/results/...1202-01888.asc

Intel P4 SPECfp rate
http://www.spec.org/cpu2000/results/...1202-01889.asc

Both systems had a 3.06 GHz Pentium 4 with code compiled by ICC 6.0 for the C/C++ source code and IFC 6.0 for the F77/90 source code. Both had PC1066 RDRAM and sufficient memory for all the tests (since the memory footprint for any benchmark in the suite is between 100 and 200 megs). The Dell used a Microquill Smartheap Library which might explain why it did particularly well in SPECfp rate.

[dli537]

CubeBoy, you should read Hannibal's interview
again.

You showed the following paragraph from the interview.

"The gcc scheduler is not really designed ideally for a processor like the 970 and the Power4 and others, and that's a lot of what the IBM and Apple teams have worked on"

It says gcc scheduler is not good for 970 and Power4. It did not say anything about Pentium 4.
Where did you get the impression that GCC can not schedule for Pentium 4? They mentioned nothing about Pentium 4 in the interview.

[CubeBoy]

Originally posted by dli537:
CubeBoy, you should read Hannibal's interview
again.

You showed the following paragraph from the interview.

"The gcc scheduler is not really designed ideally for a processor like the 970 and the Power4 and others, and that's a lot of what the IBM and Apple teams have worked on"

It says gcc scheduler is not good for 970 and Power4. It did not say anything about Pentium 4.
Where did you get the impression that GCC can not schedule for Pentium 4? They mentioned nothing about Pentium 4 in the interview.

I didn't get that statement from the interview, thats just the fact when it comes to GCC. If you look at GCC 3.3's source code, you'll find that it's able to schedule for nearly every x86 cpu (Pentium, Pentium Pro, AMD K6, Pentium 2, Pentium 3, AMD Athlon, AMD Opteron) but not the Pentium 4. The testers used -mcpu=pentium4 which only tweaks some instruction costs for some very specific problems (in my opinion, a rather pathetic attempt to make up for the absence of a scheduler). Not having a scheduler is going to seriously hurt Pentium 4 performance in floating point code, which really needs to be well scheduled.

Regarding the GCC scheduler for the PPC970, he said it wasn't designed ideally for the PPC970 (and generally speaking, this is true for nearly all cpus) and that Apple and IBM has spent alot of their time and resources working on it or in other words, optimizing it.

I should've been more clear in my previous post.

[CubeBoy]

Whoops, accidently pressed reply instead of edit. Delete this.

[Superchicken]

hehe I wonder how Intel's guy would claim the Real world tests were botched eh?

[Spliffdaddy]

because everybody else gets better results with the P4 than Apple did.

[nvaughan3]

Originally posted by KT:
[B]Intel is trying to discredit Apple's new G5, using the same old lame "evidence" used elsewhere on the web.

Link please?