Advocacy
Dojo (HowTo)
Reference
Markets
Museum
News
Other
Myths
Press
General
Hack
Hardware
Interface
Software
Standards
People
Forensics
Web
CodeNames
Easter Eggs
History
Innovation
Sightings
Opinion
Martial Arts
ITIL
Thought
|
"Since the PowerPC alliance came together in 1991, it has largely kept its promise to offer microprocessors at roughly twice the price/performance ratio of Intel's x86" - Byte Magazine 11/96
If you have questions about what performance is, then read the understanding performance page. This is the more techie version of what's what.
There are two types of tests - processor tests, and application tests. Processor tests try to isolate the processor from the rest of the systems and exercise only the processors. Application tests tend to go more towards the real world - just see how fast the computer runs various applications or suites of Applications.
Byte - PPC has huge lead
|
| integer | floating point |
| 90mhz Pentium | 1 | 1 |
| 200mhz Pentium | 2.1 | 2.2 |
| 200mhz PowerPC 603e | 3.8 | 2.6 |
| 200mhz Pentium Pro | 2.8 | 3.5 |
| 200mhz PowerPC 604e | 4.3 | 4.4 |
240 MHz PowerPC 603e (the fastest today) would be roughly 4.5 times faster than a Pentium-90 - or equivalent to 428 MHz Pentium. Of course now you can get the Pentium at 233 MHz -- and the 603e at 300 MHz.
225 MHz PowerPC 604e would be roughly 5.0 times faster than a Pentium90- or equivalent to a PPro at 343 MHz. 604e's are available at 350MHz. The G3's are also available -- and they just blow away the integer performance of all other chips (a 266mhz G3 is nearly twice as fast at integer as the same speed 603e. Roughly the same in FP).
Roughly the PPC's scale linearly, and the MMX Pentium is about 10-15% faster than the regular P5 (with some leaps for very selective tasks). See MMX links for more information.
Byte uses small tests to exercise the CPU, and is good for measuring CPU performance -- but since many of the tests fit in an L1 Cache, it is not as good for measuring CPU-Memory or CPU-L2 cache performance.
HINT Marks is done by Ames Laboratory
|
| Double | Int |
| PowerMac 604e/200mhz | 6.4 | 9.5 |
| Pentium Pro 200mhz | 6.1 | 9.5 |
The PowerMac beat the PPro and that was running WinNT on the PPro. Remember the PPro takes a substantial performance hit running Win95 or 16bit apps like most users are doing. So the real world results of the Pentium are much lower than this. Furthermore, the compilers were not a very optimized compiler for the Mac (with some tests showing up as much as 100% faster on the Mac by using the Motorola or MrC compiler), and RISC computers are more sensitive to optimization than CISC.
Lastly the PPC 604 is available with at least 20% faster mhz than were tested.
SPEC - PPC has slight lead.
Spec is a large suite of performance measuring benchmarks supposably geared towards System and Processing testing (not just CPU, but also tests memory sub-systems and cache as much as processor).
This test is still the most popular suite, and one of the most respected. However, one must realize that it does not test pure system performance (I/O is mostly ignored), and it does not test CPU alone (Cache/Memory is a factor). PPC's tend to do a little better in either the more isolated tests, or the more general ones -- but in this middle ground they tend to only be marginally better than Pentiums. (The G3's faster cache and Memory systems should alleviate some of this).
These tests are unique in that they show the least amount of performance advantage for the PowerPC. The Spec suite is the most preferred by PC-types.
I have some issues with Spec for the following reasons --
- The test suite and source is not publicly available so individuals can not verify the results at home. (Most PC's would not get the performance of Intels reference boxes). The suites are licensed (not freeware) and are available to those that pay (a lot) for them.
- This suite has a large variety, but many of the benchmarks do not fit inside the processors L1 or L2 caches. This means that the tests are often not only measuring CPU performance but is more measuring CPU to Memory. The PPro system tested by Intel used a 4-way interleaved SDRAM sub-system to squeak every ounce of performance out, using a variant of the PPro (with a 1 Meg L2 cache) that is not readily available. Since users seldom have this configuration it is not exactly a practical test.
- Intel has done heavy optimization on their Reference Compiler for Spec's - this too gives them a performance boost on Spec's that may not be noticed in real world, since nobody builds shrink-wrap Software in the Wintel world with the Intel Reference Compiler. Using the other commercial compilers yields inferior SPEC results (and assumably processing results) by comparison. People are using the Motorola and IBM compilers. So "your mileage may vary".
- The hot boxes can only get their performance running 32 bit code (and often Unix). But the majority of Intels code base is still 16 bit code - where the Pentium Amateur actually out-performs the Pentium Pro (due to misalignment penalties and the like).
So Intel seems to do exceedingly well at Specs, especially compared to what the real world results (Application bench marking on regular boxes). It may be possible to achieve such results as Intel gets on Specs, but it seems completely impractical (rare) in the real world -- based on the application bench marking results. Ironically Bytemarks, while being a much more primitive test (CPU only) may be a better predictor of the real world results, because of other consideration and practical usages -- like the tests are done on the machines people have, with the compilers people use.
Digital Sapients, Inc.
An independent, third-party benchmark testing, conducted by San Francisco, CA-based Digital Sapients, Inc., has shown that Power Computing Corporation MacOS systems outperform similarly equipped and configured Dell Computer systems on average from 13-16% across a myriad of software applications.
http://www.powercc.com/Product/CPUs/Benchmarks/DS/index.html
Competitive Assessment Services
July 1995 - Executive Summary of an Independent Benchmark Study
The study showed that Apple Power Macintosh computers consistently performed faster than similarly configured Pentium processor-based PCs in a wide assortment of tasks, including graphics, publishing, and scientific applications.
These were Application tests to get the most "real-world" results.
http://product.info.apple.com/productinfo/factsheets/comparison.html
Linpack Benchmark -- Java Version (no JITc)
| PowerMac 8500/120 | 0.559 Mflop/s in 1.229 secs |
| Alpha 21064, 233MHz | 0.527 Mflop/s in 1.303 secs |
| P6-200 | 0.341 Mflop/s in 2.01 secs |
| 120 MHz Pentium | 0.315 Mflop/s in 2.181 secs |
This test was done before the JIT compilers came out (which made a major difference in performance on PC's, but has not been used to re-benchmark Macs).
The 604/120 is 63% faster than the 200mhz Pentium Pro
The 604/120 is 77% faster than the 120mhz Pentium
Also note that now days the 604e is much faster than the 604 and running at 250mhz, while the PPro is still only 200. (So the 604e/250 would be 400% faster than todays P6?!).
Mathmatica -
| PowerMac 8500/180 | 1.69 |
| HP Vectra XU6/200 | 1.071 |
| Dell Dimension XPS P166c, OS: Win95 | 0.496 |
The PowerMacs are available at much faster speeds than were tested.
The PC's did much better when running the OpenStep version of Mathmatica (but the Rhapsody versions have not been tested on a PowerMac, so that is not a fair comparison)
http://fampm201.tu-graz.ac.at/karl/timings30.html
Ingram Study on Performance -
An earlier Application performance test that shows that the PowerMacs with the older 601 processor were faster than pentiums at the same clock speed by about 50%. (PowerPC's have been almost continuously available at faster MHz than the Pentium counterparts).
http://www.halcyon.com/kegill/mac/win95/Ingram.html
SUPPORTING ARTICLES
- CPU Bench marking (very technical) - from C'T (German Computer Magazine - translated) - http://www.heise.de/ct/Service/English.htm/9611270/
- Berkley's Benchmarking Links
|
