Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Originally posted by Fender2112
Nice article.

I did not know x86 architecture was that old. You have to figure that Intel and AMD will eventually have to drop the x86. Or at least reavamp it. There definitely will be some growing pains. But that's part of growing up.

x86 architecture is very very old.

I remember speaking with Intel engineers during a recruitment drive at my university, and they did not have any plans on abandoning the x86 architecture then. This was back around when the 680x0 family was starting to go on the wayside. If they were going to do it, it should have been at that time. The PowerPC was not out yet, and they could have done a transitional chip to get out of the cruft that the x86 architecture is.

I believe the Pentium wasn't out yet. And this was supposed to be their really fast new chip because they started doing micro-ops. Should have spent the effort doing a new chip, and putting the marketing money behind it.

There was the i960 chip, which only managed to get printers to use it.

(Motorola had the 88000 series, IIRC.)
 
Mr. MacPhisto:

Now, answer THE question, oh ye who worships at the altar of Intel
You are in no position to accuse anyone else of worshiping a company.

If the Moto 74xx had branch prediction, cache, etc, FSB, EVERYTHING equivalent to the Intel chip - who would win in raw performance? The answer is obvious - it is the PPC. The fact that the G4 could compete with it quite well at 130nm with more cache and a faster FSB shows the potency of PPC because, admittedly, the G4 has not been pushed forward very much by Motorola. And as for the omission of the L3, not all G4s are equipped with the L3 cache.
Obvious? Fact? Yeah, perhaps its obvious when you can't think clearly about the object at hand for all your Apple worshiping. In reality the 745x is an out of date design that memory bandwidth and cache cannot compensate for. Certainly it could be reworked with better scalar floating point power and superior out-of-order execution, but as it stands now its just not that impressive. I've asked many times for someone to show me processor/memory benchmarks where the 17" AluBook 1ghz G4 with 166mhz FSB significantly beats the 15" TiBook 1ghz G4 with 133mhz FSB. If bandwidth were the problem I would expect 25% more of it to make some difference.

If IBM and Motorola decide to step it up on the consumer end to create a viable option to the x86 then they will surpass anything Intel can come up with.
Worship.

IBM has pledged to kill the Itanium and wants to have PPC architecture become more dominant - and I believe IBM will succeed, even on the consumer platform.
Worship.
 
Originally posted by MorganX
>>If SPEC marks are to a useful measure of CPU performance they should use the same compiler, an open source compiler is ideal for this as any optimisations added for one CPU will be in the source code and can thus be added to the other CPUs also keeping things rather more balanced. <<

This is just absolutely ridiculous and stupid. Benchmarking a CPU using code not optimized for it, will not measure anything except how slow a CPU can run software not written for it. This is utter stupidity.

Comparative benchmarks should be run using the best available compiler for each CPU. This will give the best indication of real world performance as each platform will run software optimized for it.

What good it is to use a Sh**y open source compiler, when you're never going to run Sh**y open source software on a given CPU?
Woah dude, calm down.
SSE2 shouldn't be used because its very easy to perform parallelism with instructions that are always the same i.e. SIMD, when SPEC benchmarking is used, I am pretty sure that they are just doing a lot of generic loops etc. that alot of "real world" appications wouldn't be doing all the time, thats also why Alitvec is not being used on the G5 SPEC benchmarks...doh!!

And as for the s**ty compiler stuff, oh come on man!! Thats GCC, maybe not the most fast thing ever, but thats the same one that compiles the same Linux that beats out Win2000 on many things such as webserver/SMB stuff, one the same H/W (of course its better coding helps, but MS has got Billions of dollars, surely they can do better no?)...
 
Fukui:

A lot of the sub-tests in SPEC are quite real world, the ones I use are all integer tests, and include GCC compiling, gzip and bzip2. I do not believe that using SSE2 is unfair; I think Apple just needs to work on GCC till it can make them some AltiVec code in SPEC.
 
Originally posted by ddtlm
Fukui:

A lot of the sub-tests in SPEC are quite real world, the ones I use are all integer tests, and include GCC compiling, gzip and bzip2. I do not believe that using SSE2 is unfair; I think Apple just needs to work on GCC till it can make them some AltiVec code in SPEC.
Yea, but with SSE2, it would be easy to get things running artificially high, kinda like apples "G4s velocity engine can provide up to XXX GFLOPS" statement, yea, right, maybe for something written in assembly!! I'm just saying that both had their respective SIMD units 'turned off", thats pretty fair because most people aren't gonna use it explicitly anyway.

And, as far as apple's "tweaks" of the benchmarks, most of you fail to remember Joswiaks (however you spell it) statement, that those "optimizations" would be in the final G5/panther release, special malloc and all...just that the prototypes on stage didn't have them set up by default. Regardless of everyone's "nyanyanya blablabla" mac/pc rantings the truth will speak for itself when tom's HW gets a hold of them and does some real third-party benchmarking, until then, its just throwing sand in the wind...
 
Originally posted by ddtlm
Mr. MacPhisto:


You are in no position to accuse anyone else of worshiping a company.


Obvious? Fact? Yeah, perhaps its obvious when you can't think clearly about the object at hand for all your Apple worshiping. In reality the 745x is an out of date design that memory bandwidth and cache cannot compensate for. Certainly it could be reworked with better scalar floating point power and superior out-of-order execution, but as it stands now its just not that impressive. I've asked many times for someone to show me processor/memory benchmarks where the 17" AluBook 1ghz G4 with 166mhz FSB significantly beats the 15" TiBook 1ghz G4 with 133mhz FSB. If bandwidth were the problem I would expect 25% more of it to make some difference.


Worship.


Worship.

I think the SDR bus helps to limit any improvements the PBs would have by going to 167MHz or using DDR RAM. 25% is not all that much of an increase when you're dealing with such slow speeds. Although I would imagine the 17" PB is a bit faster than the 15"


As for worship - I can accuse who I like when it is quite obvious that some people absolutely love what Intel is doing with the x86. I'll admit that I do have faith in IBM and their ability to produce a superior chip. Moto, not so much, but I think they can do better.

I'm also not debating the 74xx is out of date, my point was that it could still do quite well if those updates are made - and would likely be able to match many of Intel's current CPU's in raw performance. Seeing as the 74xx is outdated, it's impressive that with a redesign it could still be a pretty good chip.

But the 74xx needs to be improved upon and replaced if Moto/Apple/Intel intend on keeping a 32bit processor around for a while. Either that or we'll wait on the Power5 derivation next year, which should be a very impressive chip and should send the 970 into the low-end by 2005. That's still a long time to have to use the current architecture of the 74xx, but the 750VX may solve that problem.

I'd also hate to disappoint you, but I'm not an Apple worshipper so much. I love the machines regardless of processor because I it is reliable and efficient, but I'm also only a recent convert who has been in the PC world since the age of 5 over 20 years ago. I've experienced the x86 in many incarnations and am less impressed with every iteration. Admittedly, the Pentium M is probably the best thing Intel has gotten out of x86 in a decade, but I doubt it will last and I don't expect Intel to really come up with much for the future.

I've also experienced Centrino in real life and was not overly impressed. While they run cool, the laptop I used for a few days experienced quite a bit of lag and still suffered under a miserable OS (XP Pro). Everyday tasks fucntioned slower than they do on my 2GHz PC desktop (w/ 1GB). Menu's lagged somewhat on the IBM Thinkpad T40 w/ 1.2GB of RAM and a 1.3GHz Pentium M. There's no reason why there should be any lag in regular activity with that much speed and RAM. I ran Half-Life on the machine (while plugged in and at fullpower) and encountered ALOT of slowdown. My 2GHz P4 did a lot better with an older 32MB Radeon card (purchased over 2 years ago and running on older AGP tech). True, laptops aren't really meant for gaming, but this machine was supposed to run faster than my desktop.
It also managed to crash only a few hours out of the box (though that's not Intel's fault), thus why I keep my iBook and do more on it, despite the fact that it only has a 700MHz G3 - which is snappier than the 1.3GHz Centrino based machine. That has a lot to do with the OS, but still - color me unimpressed with Intel's latest and greatest.
 
Mr. MacPhisto:

I think the SDR bus helps to limit any improvements the PBs would have by going to 167MHz or using DDR RAM. 25% is not all that much of an increase when you're dealing with such slow speeds.
You need to re-examine you logic. At low speeds that 25% is most important; if the G4 really needed bandwidth then 25% more of it should have mattered. But noone can seem to demonstrate an advantage, which leads to the conclusion that the memory bandwidth is not the problem.

As for worship - I can accuse who I like when it is quite obvious that some people absolutely love what Intel is doing with the x86.
As soon as Cubeboy and myself labeled that article as the junk it is, you paint us as "loveing" x86, "salivating" over it, and "worshipping" Intel. Shame we had to burst your bubble there, but like I said that article is junk, you'll need to find a more effective way of dealing with it than cheap smears.
 
Originally posted by MorganX
Benchmark abuse has nothing to do with my comment, unless you are suggesting that compilers cannot be optimized without specific benchmark cheats.
...

Cheating is a separate issue not a part of my comment or much of the article. Maybe I should reread it.
You either missed my point or deliberately overlooked it. SPEC is a standard benchmark suite. Using a vendor-supplied compiler runs the risk of having the benchmark already coded in, irrespective of the optimizations.
 
Originally posted by MisterMe
Using a vendor-supplied compiler runs the risk of having the benchmark already coded in, irrespective of the optimizations.

I didn't miss the point. I missed the value of it. Running a third party supplied compiler runs the risk of having it optimized for one plat form and not the other, irrespective of any deliberate abuse.

As you have stated, these "cheats" on the x86 side are so common and widespread, Apple or Veritest should have been easily able to detect them, and then they would be having to explain themselves.

At some point you have to take the risk that adults and professionals will behave like adults and professionals.

Using your logic, your point, you must basically discount all benchmarking because there is a risk of cheating in any scenario. Including the one that was conducted.

Can I conclude that it is your opinion that the vendor-neutral compiler used was equally optimized for both platforms?
 
Does anyone actually know how much SSE2 improves performance?

Suggested Readings:
http://www.aceshardware.com/Spades/read.php?article_id=25000196

A few things to note from the article above:

“For SPECfp2000 the new SSE/SSE2 instructions offer about a 5% performance gain compared to an x87-only version (this is using ICC+IFC 5.0).”

“It is very hard for compilers to get SIMD code out of the source code, without the help of a human programmer. As the source code of SPECfp can not be changed, it is not possible to add hints, and it is not permitted to manually change the machine output.”

Pentium 4 2.0A
ICC+IFC 5.0
Intel D850MD
PC800 RDRAM
SPECfp Base: 732

Pentium 4 2.0A
ICC+IFC 6.0
Intel D850E (Newer Motherboard)
PC800 RDRAM
SPECfp Base: 764

Pentium 4 2.0A
ICC+IFC 7.0
Intel D850E (Newer Motherboard)
PC1066 RDRAM (Faster RAM)
SPECfp Base: 766

Even with faster ram and a newer motherboard, a Pentium 4 running ICC+IFC 7.0 compiled code only scored a little more than 4% better than the same Pentium 4 running ICC+IFC 5.0, which itself, offers only a 5% performance gain over x87 only code. I suspect it’s even less considering the better motherboard and faster ram. SSE2 can at most, offer only a 9% increase over x87 code with the newest compilers, that would effectively give a Pentium 4 3.0 GHz with SSE2 disabled a SPECfp score of at least 1065, nowhere near Apple's 600 some. Apparently, SSE2 is not the main reason the Pentium 4 scores so well in SPECfp.
 
Originally posted by MorganX
I see no reason not to take his G5 benchmark seriously, but I wouldn't give much credibility to his PC bencmark results. I don't have a problem with optimizing the G5 code, I have a problem with not optimizing the x86 code.

Using the potential abuse argument for an excuse not to optimize is meaningless. If you're going to cheat you can cheat regardless of what software you use. Given the description for the malloc library this is the only optimization I would say is a cheat just to inflate benchmark results.

Please note, that I said nothing about not taking the G5 Benchmark seriously, just Steve. I view it exactly the same as you. If you are going to "optimize" the G5, then you have to at least attempt to do similar optimizations to the "compitition."
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.