Slashdot Mirror
G4, PIII & E2K Compared
Lord Kano-The Gangster Of Love writes "A head to
head to head comparison of PowerPC G4, Intel Pentium III
and E2K ( CT : that russian chip) can be found at macintouch.
This has all of the makings for an interesting year,
indeed."
No. The Intel compiler is available on Windows NT 4, and I fail to see why Intel would deliberatly strip down its compiler to get worse performance on NT. The OS has almost no influence since 99.9% of the time (or something like that) should be spent executing user code. And finally go at www.spec.org and look at the details of the benchmarks: you'll see that most of the recents benchmarks are done on Windows NT, and that in the case of the Intel SE440BX motherboard with PII-450 Intel gets the same results with both Unixware and Windows. Windows gets even a slightly better score in SpecFP, but with a more slightly more recent version of the Intel compiler.
The article didn't quote PII numbers because it's about future chips. It didn't quote PIII numbers because "Intel will officially introduce its new Pentium III processor on February 28, and until then it isn't disclosing details about the chip's performance and pricing, nor allowing PC makers to announce specifications and prices for systems based on the new chip."
The article wasn't biased at all, no matter what you choose to read into it.
Not quite right, Compaq Precision WorkStation with
PII-450 gets 17.6 and the modell with XEON gets 19.0
what are you talking about? You never heard of plugins? Not all plugins are platform independent you know
try running lightwave 3D on NT with 2 xeons and you'll see why. Don't use them in win98 with MS Word-that's not gonna make a difference
if you take a look at real world performance, a single G3@400 beats a dual pentium II 450 nt sp4 at photoshop(PSBench5) This only shows ignorance of the Mac fanatics fooled by Apple's unreasonable claims about G3s. Intel Spec numbers look Ok to me. Exotic hardware? Strange compilers? Be more specific ? Performace computing magazine benchmarked a Dell with a PII333 once, SPEC95int was 14.1. And that was a Dell XPS with PII333 running UnixWare 2.0. hehre is URL, http://www.performancecomputing.com/reviews/hardwa re/9812.shtml Also, Type finger johnc@idsoftware.com | less and see what John Carmak has to say about Apple hardware and G3 in general.
Then why are tcl/tk programs so slow?
They'd be better off improving Javascript, and
allowing Java objects to be scripted from it.
This would allow Java programs to control
Dynamic HTML pages.
PA-8500 @ 440 MHz : 32 SpecInt95 and 54 SpecFp95
21264 @ 600 MHz : 32 SpecInt95 and 58 SpecFp95
too bad Apple "owns" the G4. that sucks
Are you contending for the world championship of bigotry and total stupidity ? Your logic is "with crap hardware and compilers the Pentium II goes slower, so G3 is a much better processor".
Sorry, but when you're benchmarking processors, you should benchmark processors, not how crappy motherboards and gcc could be in the worse case. This is just obvious scientific methodology. That's the rule for SPEC.
Anyway, decent motherboards do exist, just go to www.spec.org, you'll see that the Dell Precision Workstation 610 / 450 Mhz (Xeon 450: SPECint = 19.0), Siemens AG CELSIUS 2000 (Xeon 450: SPECint = 18.9).
We've never seen an x86 chip that's faster than intel before. I hear alot about Amd k7 but nobody thinks it will beat the merced-just the pentium 2.
hmm imagine if AMD used the E2k for the K8...
well, do some results surfin at www.spec.org.
I'll estimate my PII-505 to do around 20 SPECint95.
The powerpc processor is astounding, yes. Even apple's effort to increase the bus speed from it's previous 66 mhz bus speed to 100 mhz made SLIGHT difference instead of a staggering gain in performance, it peaked a little... which would indicate that the limiation is somewhere else, maybe the slow dma/33 speeds how about trying it with U2W scsi? Or it's the siomple fact that the bus is able to keep the processor fully saturated and the bottleneck is the processor itself... My guess is that a formal study will have to be done, one which i hope to do myself formally...
So, when a website reports the speed of PC hardware, you don't believe it, but when it reports that the Mac is faster than a PC you do believe it? Everyone except for your friend gets scores of 18, that means everyone is using magic compilers and the latest in alien technology?
It scares me to think that people will start developing web sites that are only accessible by Pentium III's. But, if you think about it, the real leading edge multimedia stuff on the web these days is all being driven by the porn industry.
Is porn the industry that Intel is trying to market the Pentium III's to?
Actually, the point to the Xeons is better SMP support, leading to superior SMP scaling. Buying a one-processor Xeon is pretty stupid, but if your company *must* execute x86 code and it has to be done as quickly as possible, then a four-way Xeon is de facto "as quickly as possible".
-- Guges
Yeah, that MacOS has had a little too much hot sauce. However, OS X will be very very very less bloated, thanks to a better extensability architecture than Extensions, Plug-Ins, DLLs or whatever you currently use. And, no, Mac is not more bloated than Windows. But the hardware? The new G3s run pretty good, hardware wise. 100Mhz main busses, (semi-good, but standard) PC 100 memory, 64bit PCI, etc, makes a pretty good architecture, i think, but maybe i'm wrong.
Hey, What do you mean by "regular" hardware. Like shit hardware? How about benchmarking a high quality brand PC such a Micron or Gateway? Come sometimes Compaq for exapmle can make a PII 400 system act slower than a PII300.
I am sorry for YOUR hardware.
Don't make people laugh. A G3/450 beating an SMP PII system ? Come on Apple did fool people, but I did not know that it was by this much.
Obviously there is a problem somewhere. Did he use the Intel compiler with the correct optimisations options ? Did he bought the real Spec benchmark suite, and where did he got instructions for compiling it ?
You see, Macintosh machines have crappy processors, but incredible bus architecture. That means that their 232Mhz (or whatever the fuck it is) can drive alongside a PII 300 (with the 66Mhz bus).
Gateway own Amiga, and created a subsidiary called Amiga Inc. for the development of a new "wonder machine" - We now know it is going to use QNX/Neutrino as the OS kernel, but the main task of Amiga Inc. is to provide standard APIs such as OpenGL for the new system.
.library and .device system, mainly reentrant shared code, 32-bit clean since 1986... The joy of nutcase command lines such as type "Hi from a new Window" >CON:0/0/640/256/New_Window_Title/AUTO/CLOSE/WAIT, Sempaphores, ARexx , the list just goes on and on...
Originally, they had earmarked one unnamed superchip for their system, but have since had second thoughts, and look like they're now heading for multiple targets (one of which could well be PPC, since QNX has already said neutrino will run on PPC)
PPC on the so-called "classic" Amigas is a weird hack where a 50MHz 68060 processor shares the system bus with a 233MHz 604e PPC (The architecture actually does RC5 decryption very well).
Amiga Inc. couldn't care less about this architecture - It was developed by a third party, Phase 5, who then got all sulky when Haage-Partner (with WarpOS - nothing to do with OS/2 Warp) developed a better system kernel than their own (PowerUP)
Even so, at least one company is trying to make a G3 "classic" Amiga. Incidentally, LinuxPPC does run on the 2-cpu boards, quite satisfactorily and stably, it's called Linux/APUS and is on sunsite.auc.dk
I often wish that people would look at some of the better aspects of Amiga system architecture when developing for other systems (I know BeOS actually did)
Datatyping, MUI , AHI/ARTAS, C ( mark you) OOP programming via DoMethod(), the
I find it interesting to compare:
Linux:
Enlightenment/GNOME/ImLib/ESD
Amiga:
Intuition/MUI/Datatypes/AHI
Is Raster an (ex-) Amiga user????
And which of the Alpha, PPC and SPARC can I stick into my socket 7 or Slot 1 mainboard?
If someone wants to compete with Intel they have to go the route of Cyrix and AMD, and be a plugin replacement. The average home user doesn't want to dump their Voodoos, AWE32s and such just for a faster CPU.
I am really getting excited about the G4. It will hopefully only use a small amount of power, and be SMP aware.
Wouldn't you rather buy G4s that each have 50% more FPU performance than a Xeon. Considering that a Xeon is about $4,000 and a G4 will be about $600.
BTW: Personally I love Alphas, but don't they use too much power to really be useful in an average PC?
It will also be nice to have Altivec, even in a worse case scenerio it will still be about twice as fast as Streaming SIMD, because it is twice as wide. It also appears that Spec marks always underestimate the power of the PPCs, and many other high end chips as well. I remember reading somewhere that they just use a type of test that isn't a true measure of performance, but it does get within about + or - 50-100%
In any case, the true G4s are going to be the ones that contain multiple cores on one chip. Two of these babys on a single chip will really scream.
It basically says that a dual-PII/450 has comparable performance to a PII-300 (see PS5bench result page), within 5-10%. Talk about bogus benchmarks. This also says much about the unbiaseness, and technical expertise of the authors of the article.
The difference being of course, you can get a reasonably fast PC with an Intel chip in it (or AMD or some other x86 compatible chip) for under $2k now. Any decent Sparc is at LEAST $10k and then I'm not even sure that buys you SCSI drives with it. SGI's and Alphas are no better and are just as much.
According to mot's website, the 400MHz G3 gets 18.8 int and 12.2 fp.
-a
How many applications are compiled with Visual C++, Borland C++, Watcom C++, Intel C++ ? I don't know. But in reality, I expect any CPU-intensive software done by people who aren't complete idiots, to be profiled and compiled by a decent compiler.
you just proved my point, most of the scores posted at www.spec.org are done with the Intel compiler..
Absolutly not, since the PowerPC ones are done with AIX and Motorola or IBM C/C++ compiler. How many applications are compiled with AIX and Motorola or IBM C/C++ compiler ? If you are rejecting the SPECint got from Intel, then you have even more reasons to reject results for G3/G4.
Don't forget : the x86 architecture instruction set is a hideous bodge job, and really, even the motorola 68k series was better in terms of design. (which is why a 50MHz 68060 can hold its own against far "faster" pentiums - just ask any amiga user(of course, they're going PPC now anyway...)
Even so, motorola realised they could do better...
The PPC chips at least have a decent instruction
set, despite being RISC .
I'd rather program (and have done) in PPC macro assembler than in x86 assembler any day...
The main reason people use x86 is simply because people use x86.
Let's not even mention 4 register crapness, segfaults, and brain-dead pipelining...
What are SpecInt95 and SpecFP95 of fastest Pentium II's (or PIII). The article curiously leaves that out. Could it be that the G3 won't look so overwhelming by comparison (as opposed to what the Mac ad campaigns would have you believe)?
PII-450 is 18 not 12 SPECint95??
I'm intrested to hear were did you get that number from?
Maybe that's why they picked their new slogan.. "This way in" huh huh huh.. :-)
When did Sun announce plans for x86 based systems? They have Solaris x86, true, but they don't sell systems for it, and they have announced that they will support Solaris on IA64, but Sun workstations with x86 processors?
Posted by Nr9:
They take the SUM of the times because it really isn't a benchmark, its designed to estimate the time a graphics pro will waste on filters....
anyway's in working with pentium II systems, i kind of believe this benchmark result but not the 18 specint95
ps... by compiler dependent i mean that the tester has to used a precompiled binary of photoshop and can't really optimize his system with another compiler
Posted by Nr9:
BLAH!!!
Intel claims high spec scores but using standard compilers and machines, a pentium II 450 does 12 specint95 and 12 specfp95
Posted by Nr9:
The Intel tests are done with intel using exotic machines and compilers
true personal testing shows about a 12 for both specint95 and specfp95 for the pentium II 450
Posted by Nr9:
because the shitty 18 specint95 numbers have been posted that much times....
a Pentim II at 450mhz has NOWHERE THE PERFORMANCE OF 18 SPECINT95
if you take a look at real world performance, a single G3@400 beats a dual pentium II 450 nt sp4 at photoshop(PSBench5)
Posted by Nr9:
a pentium II 450 actually(usually regular hardware and compilers) does 12 specint95 and specfp95
a G3 450 doing 20 shows that a G3 is much faster than the pentium
Posted by Nr9:
because it sux and hard to use
no applications use it
Posted by Nr9:
of course he didn't use the intel compiler because it just doesn't correspond to real world performance
most applications don't use the intel compiler so why should benchmarks use it
note that Apple was conservative during their bytemarks because they didn't use their MrC(which is used to compile the Mac OS compiler, which posted bytemark scores twice of those using the codewarrior compiler because it is more widely used...
since the majority(sadly) of intel users use windows, they should use ms vc++ or borland to test their systems
Posted by Nr9:
They take the SUM of the times because it really isn't a benchmark, its designed to estimate the time a graphics pro will waste on filters....
anyway's in working with pentium II systems, i kind of believe this benchmark result but not the 18 specint95
Posted by Nr9:
the site didn't mention the compiler they used....possibly intels'?
Posted by Nr9:
the G4 does 8 watts---still low power compared to other chips
Posted by Nr9:
because it sux and hard to use
little applications use it
Posted by retroman:
Whoever owns Amiga is NOT doing it. They won't even say what CPU the new AmigaOS will run on. There are several 3rd party PPC accelerators for the Amiga, but none to my knowledge are actually integrated into the system by the manufacturer.
For that matter, NO Amigas have been in production since '92/'93. Amiga International I think just now or shortly will resume manufacturing the Amiga 4000's out of NOS from the West Chester Commodore facility. Sheesh, 6 years without a new model... pathetic.
Posted by Nr9:
So, when a website reports the speed of PC hardware, you don't believe it, but when it reports that the Mac is faster than a PC you do believe it? Everyone except for your friend gets scores of 18, that means everyone is using magic compilers and the latest in alien technology?
i don't believe any Mac claim but I certainly don't believe the score of 18...DID YOU BENCHMARK THE THING YOURSELF? how many applications use the Intel compiler......
i believe the mac one because it used an application benchmark, PSBench5, which uses photoshop. It is not compiler dependent and it clearly shows the mac has the lead
Sun workstations from $2500, but w/ no SCSI
Gives a 17.2 for a 450Mhz Pentium II (no figures for the Xeon yet).
8 q3/cpu95-980831-03026.asc
http://www.specbench.org/osg/cpu95/results/res9
--
Matt. Want XML + Apache + Stylesheets? Get AxKit.
From CPU info center
450 Mhz Pentium II 17.2 - 12.9
450 Mhz Xeon 18.9 - 14.7
400 Mhz PowerPC G3 17.6 - 12.2
200 Mhz IBM POWER3 13.2 - 30.1
450 Mhz UltraSparcII 19.6 - 27.1
600 Mhz UltraSparcIII 35+ - 60+ (est)
575 Mhz Alpha 21264 30.3 - 47.7
667 Mhz Alpha 21264 44 - 66
1000 Mhz Alpha 21364 ~70 - ~120 (est)
250 Mhz MIPS R10000 14.7 - 24.5
What many people still dont seem to realize is that PPC 750 ("G3") is really descended from earlier low-power implementations, namely the 603 family.. its unfortunately lacks the high-end FPU of its bigger, older brother, the 604e, which packs one hell of a punch... as both Motorola and IBM are still making the things, I cannot understand the defocusing from that chip so early since the G3 really was no great replacement, and the G4 isnt here yet.
However, I know a few people that have touched the things and seen Altivec demos and its hard not to be impressed... too bad Exponential Technologies fell apart a few years ago, I heard of live demos back then at 500-600 mhz when PPC and Pentiums were just breaking 100.
>> Macintosh machines have crappy processors, but incredible bus architecture. That means that their 232Mhz (or whatever the fuck it is) can drive alongside a PII 300 (with the 66Mhz bus).
:) AFAIK, all current Pentiums also have a fast level 2 cache on a high-speed bus, which does wonders in relieving main buses of high loads of traffic... huge speed boost for any cpu that can pull it off.
>>
Hhahah, theres a fine example of an informed opinion... getting all too common here.
Macs have traditionally lagged behind high-end Intel main bus speeds, and like Intel systems, rely on a high bus-to-cpu speed ratio to deal with todays high cpu clockspeeds.
As far as the high-speed bus you refer to, that is a result of the new trend in three-tiered caches... level 1 is on chip, level 2 is interconnected to CPU on a very fast dedicated bus, and level 3 is what used to be called level 2
Let me just make one point simply:
PowerPC DOES NOT EQUAL PowerMac, Apple, or MacOS... so dont bring out the Apple laundry unless its fair then to diss x86 because of Windows... as we all know well here, there are more than one (or two) OSes, and hardware architectures are seemingly taking a backseat to functionality... how many platforms do we have Linux running on? Ive got a dual-processor PowerMac here that rarely sees Mac OS and absolutely rocks on other OSes (BeOS screams on dual 604e chips).
If anyone is interested, there is a big interview with Babayan (where he also speaks about Merced technology, future of the CPUs etc) about E2K att ml
http://www.ibusiness.ru/archive/story/babayan.h
and http://www.osp.ru/cw/1998/03/business/05.htm
Anyone cares to translate?
Great that you have done it.
:-)
;)
I understand Russian and was pretty excited after reading all these articles, but I'm not well at making translations
If I won't see your article posted on slashdot RSN, I'll write them as well reminding them about it.
no unfounded speculation on the k7? im hurt.
Log
Our shop is almost all Sun. 4 people got a 2,800 US dollar deal on USparc 5. IDE, 4Meg video. USparc 333 processor.
:)
Screw that. I got a dual PII 450 ATX Full Tower with Adaptec SCSI, G200 video, SCSI 9G and 32x CD for the SAME DAMN PRICE. Redhat preinstalled from atipa.
Sun guys are already whining about having to buy a SCSI card to add all their external drives. They are about out of slots already since the PC-Card (AMD 300 computer on a card)that Sun pawned off on them takes up two slots in their mid-tower which is short on expansion slots to begin with. I'll never fill all the slots on the Tyan mobo. I'll run out of IRQs first.
Some people need the Sun stamp of approval for that fuzzy feeling. Bah. (I keep telling them they should run Linux on the Sparc processor and Solaris on the AMD)
Suns makes really nice servers.
-k
Has anyone else noticed the extreme antisocial behavior inherent to many of the internet porn sites? And no, I'm not anti-porn, pro censorship, or anything else of that ilk -- my desktop backgrounds would shock many a straitlaced person.
However, I have encountered numerous porn sites which have a tendancy to be very antisocial when it comes to java scripting, not to mention eye-numbing graphics that are irritatingly distracting from the interest at hand (looking at people engaged in various natural/unnatural acts). Before surfing for porn I always have to disable javascript, lest my entire browser is hijacked! Unfortunately I need javascript for a couple of sites I use regularly (aviation weather sites mostly), else I would leave it disabled always. As an example of "hijacking", some sites refuse to ever let you leave, popping up another netscape window after you've closed or exited the porn site. The only way out is to quit Netscape -- you can't kill the windows fast enough to get out otherwise! Other sites pop up twenty or thirty (or infinite) browser windows to various advertisments -- extremely obnoxious (this slowed my xsession down once so bad I had to log in remotely and reboot -- and even that took awhile). Do these people think I'd ever give them one red cent for any product they're trying to push after treating my machine in such a way? Who the fuck are these idiots, to treat the internet, my machine, and my browser, as if it were a broadcast medium where they should control the content and determine where I surf to?
If all the porn sites become PIII only the only skin off my nose will be the time lost to dig out some old scripts and start ripping stuff down from the binaries news groups again. These idiots offend me, not by their content, but by the disdain, arrogance, and disrespect with which they treat me, the consumer, and my property.
The Future of Human Evolution: Autonomy
The width won't help much if you are performing operations that AltiVec isn't designed for.
Both AltiVec and KNI have considerable strengths. AltiVec, if I understand correctly, is optimized for DSP-style operations; multiply-accumulate and similar operations useful in filtering. KNI, if I understand correctly, is optimized for geometry operations, like componentwise multiplication of floating-point vectors. At something like signal processing or 2D image processing/filtering, AltiVec has a clear advantage. At something like 3D geometry processing or ray-tracing, KNI has the advantage. Which is more useful to you depends on what you are doing.
In practice, IMO AltiVec will speed Photoshop up some more and that's about it. IMO, KNI would have been useful a year ago for gaming but will now be useless because graphics cards will be able to do geometry transformations themselves. So, IMO neither is likely to be as big an advantage as their respective makers claim them to be.
Alpha is a canonical RISC machine, more so than
SPARC or PPC. The only simpler instruction set
out there is MIPS. This means Alpha will be hurt
even more by the problems that are plaguing all
high-end RISC designs now. It is becoming the
very antithesis of itself. Very very complex.
It's time to move beyond RISC.
Russians have always been great with theory, but poor with implimentation. I sincerely doubt that the E2K will ever exist. If it does, it won't be a Pentium or PPC killer. Anything can look good on paper. Putting it to silicon is a very different matter. The ALPHA is where it is at in terms of technology. Nothing else really compares.
..... okay,
Wait, I'm putting on my flame proof suit,
BEGIN FLAMING NOW:
-Master Switch, one more element in the machine
Personally, I could never figure out the point of the Xeon chips. They run maybe 15% faster than an equal Mhz P2, and they cost WAY too much.
I mean, I'm all for extra power, but come on!
Another non-functioning site was "uncertainty.microsoft.com."
The purpose of that site was not known.
Okay, I hate to break the news to you, but you've got it backwards. A PowerPC 750 (Apple calls it the G3) at equal Mhz to a Pentium-II is a faster CHIP, but the internals of Apple's computers (plus the only OS more bloated than Windows, MacOS) makes it slower.
Try running optimized Linux on A P2-450 and a G3/300, and (if you're not running on Apple hardware), the G3 will win.
Another non-functioning site was "uncertainty.microsoft.com."
The purpose of that site was not known.
PPC MacOS's equivalent to DLLs are shared libraries. PPC MacOS's shared library architecture is more sophisticated than Win32's DLLs. Instead of using virtual memory to map files onto memory, PPC MacOS's code fragment architecture does not require virtual memory and performs better without virtual memory. A shared library is a type of code fragment when it is loaded into memory, and a code fragment is simply a table of contents, storage space for globals and statics, as well as code. It is much harder to work with code fragments than DLLs if you are an assembly language programmer, but if you program in a normal programming language, the compiler handles the interaction with code fragments required to call functions in other code fragments. Simply, code fragments are a far more sophisticated, complex method to do the same thing that DLLs do without being slowed by virtual memory, memory page management, accessing files, etc. Code fragments are much more suited to a RISC environment than a CISC or Intel environment, which is why you probably will not be seeing a version of Windows which compares directly with a code fragment architecture any time soon.
DES Khaddafi KGB genetic jihad Uzi Rule Psix Qaddafi cryptographic Peking Mossad Legion of Doom Albanian Serbian Saddam
The G4's SIMD capabilities look like they maybe theoretically 4X as fast because the Pentium III has only 8 128Bit registers and requires a mode change compared to Altivec's 32 128bit registers. Plus you get the G3's integer performance, the 604E's FPU and up to 2MB of full speed backside cache. Drool.
If you are going to compare numbers on Unix, NT, Mac, and etc. You have to level the playing field. You can not compare Unix to any of the gui based systems unless you are running an x-session on the Unix machine. X-Windows, from personal experience, tends to need more horsepower than the other systems. Figiting with the mouse alone can chew up several percent in performance. There are several x-displays that will basically stop a system generating x-graphics. X-Eyes is bad enough but the rest are much worse. You can only compare a Unix benchmark to DOS benchmark unless you add the additional computation.
.85 out of the expected 1.8 increase for a dual cpu system badly handling state changes. When you have to save all of the registers and tranfer them to the other cpu so it can have its turn, you lose performance. Adding more cpus just makes it worse.
Does anyone know if PhotoShop is SMP wise. A lot of SMP systems lose 5% or so as the number of processors goes up. The rational is that it is spending