MIPS vs Pentium vs Alpha Windows NT Benchmarks
Bill Glazier
NEW 32-BIT APPLICATION SOFTWARE BENCHMARKS SHOW MIPS-BASED
PCs AS FASTEST FOR MICROSOFT WINDOWS NT
MIPS-Based PCs Run Windows NT Applications Over 200 Percent Faster
than Intel Pentium and 34 Percent Faster than Alpha AXP
MOUNTAIN VIEW, Calif.-September 20, 1993 -MIPS Technologies, Inc.
today released benchmarking studies which show MIPS-based PCs
operating native 32-bit Microsoft Windows NT software applications
significantly faster than PCs based on other microprocessors. The
tests, which involve running four different applications with fixed
sets of data, show that a MIPS-based PC (with a MIPS. R4400 micropr-
ocessor at 150 MHz) runs these applications on the average of 202
percent faster than PCs with an Intel Pentium microprocessor (60 MHz),
297 percent faster than PCs with an Intel 80486 DX-2 microprocessor
(66 MHz) and 34 percent faster than PCs with an Alpha AXP micropro-
cessor (150 MHz) from Digital Equipment Corporation. On MIPS-based PCs,
certain application tasks ran 459 percent faster than 80486-based PCs,
333 percent faster than Pentium-based PCs and 60 percent faster than
Alpha-based PCs. [See attached summary]. The benchmarks are designed
primarily to test CPU performance rather than other aspects of a
computer system, such as video and disk performance.
"MIPS-based personal computers are the fastest systems for running
Windows NT applications," said Tom Whiteside, president of MIPS
Technologies. "Because the benchmarks are based on real applications,
they can be the basis for practical comparisons of Windows NT platforms.
These application benchmarks prove that MIPS-based PCs deliver the
highest real performance and the best price/performance. We will see
similar performance leads as other application benchmarks become
available.
MIPS-based systems have the power to take full advantage of Windows NT
and new applications, such as business critical, client/server and
power applications."
The application benchmarks all run under the final version of Windows NT
(Version 3.1), with the exception of the Alpha AXP PC for which a final
version is not yet available (Release Candidate 1.514 was used). The
benchmarks are based on the beta versions of four Windows NT applications:
Texim Project from Welcom Software Technology Corporation; MaxEDS and
MaxROUTE from Massteck Ltd.; Image-In from CPI, Inc. and PV-WAVE Advantage
from Visual Numerics Inc.. The applications span a variety of markets,
including management or administration, scientific and engineering.
The benchmark for Texim, a sophisticated project management application,
involves two steps. The first loads a project file and measures the
time to simulate time analysis on it 100 times. The second loads a
large project and measures the time it takes to simulate resource
allocation 50 times. The MaxEDS benchmark, a computer-aided design
package for electronic circuit boards, involves loading files for a
small and medium circuit board, routing one board and loading a
components file and placing the parts on the other. The benchmark for
Image-In, an image editing application, performs three discrete filtering
operations on a 100 kilobyte image file. The benchmark for PV-WAVE, an
interactive data analysis and visualization application, provides three
volumetric renderings of a cube.
In addition to the application benchmarks, several other benchmarks also
shows the performance and price/performance advantages of MIPS-based PCs.
One benchmark, based on the Hamilton C-Shell, generates calendars for a 20
year period. A MIPS-based PC runs the benchmark 72 percent faster than a
Pentium-based PC and 65 percent faster than an Alpha-based PC. In
addition,
benchmarks from BYTE magazine compare the performance of various PCs to a
486-based PC. A MIPS-based PC runs 320 percent faster than a 486-based PC
and 72 percent faster than a Pentium-based system.
MIPS-based PCs include systems from Acer America Corporation; Deskstation
Technology, Inc.; NEC Technologies, Inc. and NeTpower, Inc. MIPS-based
PCs are designed to optimally run Windows NT and offer the best
price/performance. MIPS-based PCs smoothly run existing MS-DOS and
Microsoft Windows application software, but their power becomes
especially apparent with 32-bit Windows NT applications. New 32-bit
applications are now becoming available.
MIPS Technologies, Inc. designs and supplies the world's most advanced
RISC processor technology. The company tests, certifies and licenses
its processor technology to its semiconductor partners which provide
processors for the computer system and embedded control markets. MIPS
microprocessors are used by a number of computer industry leaders,
including Acer Technologies, Pyramid Technology, Siemens Nixdorf AG,
Silicon Graphics, Inc., Sony, Tandem Computers Incorporated, NEC
Technologies, Inc., NKK and Tektronix, Inc. MIPS RISC architecture
components are available from Integrated Device Technology, Inc.,
LSI Logic Corporation, NEC Corporation, Performance Semiconductor
Corporation, Siemens AG and Toshiba Corporation. MIPS Technologies, Inc.
is a wholly owned subsidiary of Silicon Graphics, Inc. and is
headquartered in Mountain View, Calif.
- end -
-----------------------------------------
Spreadsheet with Benchmark timings
-----------------------------------------
Machine RISCStation 4400 466-DE DeskPro M 5/60 DECpc AXP/150
Company NEC Dell Compaq DEC
Processor MIPS R4400SC 80486DX2 Intel Pentium/60 21064/150
Speed(MHz) 150 66 60 150
Memory (MB) 32 32 40 32
Primary Cache 16K/16K 8K 8K/8K 8K/8K
Secondary Cache 1 MB 128K 256K 512K
Display 1280/G384 1024/ATIUltra 1024/QV 1024/QV
Hard Disk SCSI SCSI IDE SCSI
Bus Type EISA EISA EISA EISA
Windows NT 3.1 (Final) 3.1 (Final) 3.1 (Final) RC 1.514
Benchmarks total elapsed time (sec)
Texim Project 101.76 257.00 172.00 133.10
Time Analysis (100)+ Resource (50)
(MIPS is 252% faster than 486DX2-66)
(MIPS is 69% faster than Pentium/60)
(MIPS is 31% faster than Alpha AXP/150)
Image-In 49.05 113.37 73.62 55.01
Waves + Ripples + Emboss
(MIPS is 231% faster than 486DX2-66)
(MIPS is 50% faster than Pentium/60)
(MIPS is 12% faster than Alpha AXP/150)
MaxEDS 148.96 684.00 496.00 239.00
(MIPS is 459% faster than 486DX2-66)
(MIPS is 333% faster than Pentium/60)
(MIPS is 60% faster than Alpha AXP/150)
PV-WAVE 59.73 147.00 94.00 n/a
volume rendering of cube
(MIPS is 246% faster than 486DX2-66)
(MIPS is 57% faster than Pentium/60)
App Performance Advantage
(MIPS is 297% faster than Intel 486DX2-66)
(MIPS is 202% faster than Intel Pentium/60)
(MIPS is 34% faster than Digital Alpha AXP/150)
Other Benchmarks:
Software Development elapsed time (sec)
1764.93 2850.00 2163.00 n/a
(MIPS is 61% faster than 486DX2-66)
(MIPS is 23% faster than Pentium/60)
Hamilton C-Shell elapsed time (sec)
20.63 54.60 35.50 34.00
(MIPS is 265% faster than 486DX2-66)
(MIPS is 72% faster than Pentium/60)
(MIPS is 65% faster than Alpha AXP/150)
BYTE Benchmarks (compared to 486DX2-66)
3.02 1.00 1.76 n/a
(302% faster than 486DX2-66)
(76% faster than Pentium)
MIPS is a registered trademark of MIPS Technologies, Inc.
Pentium is a trademark and Intel is a registered trademark of Intel
Corporation.
Alpha AXP is a trademark of Digital Equipment Corporation.
Windows NT is a trademark of Microsoft Corporation.
Texim Project is a registered trademark of Welcom Software Technology
Corporation
MaxEDS, MaxROUTE and Massteck are trademarks of Massteck Ltd.
Image-In is a registered trademark of CPI, Inc.
PV-WAVE Advantage is a trademark of Visual Numerics, Inc.
Hamilton C-Shell is a trademark of Hamilton Laboratories
Len Schultz
I am posting this followup to make some clarifications to Bill Glazier's
(SGI) posting on the relative performance of the NEC RISCStation 4400 and
the Compaq DeskPro M 5/60. There was an unfortunate error in the
calculation of the relative percentages in execution speed. I shall
summarize:
>>>>> On Tue, 28 Sep 1993 20:46:23 GMT, bgla...@buffalo.mti.sgi.com (Bill Glazier) said:
Bill> NEW 32-BIT APPLICATION SOFTWARE BENCHMARKS SHOW MIPS-BASED
Bill> PCs AS FASTEST FOR MICROSOFT WINDOWS NT
Bill> MIPS-Based PCs Run Windows NT Applications Over 200 Percent Faster
Bill> than Intel Pentium and 34 Percent Faster than Alpha AXP
--text deleted--
Bill> Machine RISCStation 4400 466-DE DeskPro M 5/60 DECpc AXP/150
Bill> Company NEC Dell Compaq DEC
Bill> Processor MIPS R4400SC 80486DX2 Intel Pentium/60 21064/150
Bill> Speed(MHz) 150 66 60 150
Bill> Memory (MB) 32 32 40 32
Bill> Primary Cache 16K/16K 8K 8K/8K 8K/8K
Bill> Secondary Cache 1 MB 128K 256K 512K
Bill> Display 1280/G384 1024/ATIUltra 1024/QV 1024/QV
Bill> Hard Disk SCSI SCSI IDE SCSI
Bill> Bus Type EISA EISA EISA EISA
Bill> Windows NT 3.1 (Final) 3.1 (Final) 3.1 (Final) RC 1.514
Bill> Benchmarks total elapsed time (sec)
Bill>
Bill> Texim Project 101.76 257.00 172.00 133.10
Bill> Time Analysis (100)+ Resource (50)
Bill> (MIPS is 252% faster than 486DX2-66)
257/101.76 - 1 = 152%
Bill> (MIPS is 69% faster than Pentium/60)
Bill> (MIPS is 31% faster than Alpha AXP/150)
Bill> Image-In 49.05 113.37 73.62 55.01
Bill> Waves + Ripples + Emboss
Bill> (MIPS is 231% faster than 486DX2-66)
113.37/49.05 - 1 = 131%
Bill> (MIPS is 50% faster than Pentium/60)
Bill> (MIPS is 12% faster than Alpha AXP/150)
Bill> MaxEDS 148.96 684.00 496.00 239.00
Bill> (MIPS is 459% faster than 486DX2-66)
684/148.96 - 1 = 359%
Bill> (MIPS is 333% faster than Pentium/60)
496/148.96 - 1 = 233%
Bill> (MIPS is 60% faster than Alpha AXP/150)
Bill> PV-WAVE 59.73 147.00 94.00 n/a
Bill> volume rendering of cube
Bill> (MIPS is 246% faster than 486DX2-66)
147/59.73 - 1 = 146%
Bill> (MIPS is 57% faster than Pentium/60)
Bill> App Performance Advantage
Bill> (MIPS is 297% faster than Intel 486DX2-66)
Bill> (MIPS is 202% faster than Intel Pentium/60)
Bill> (MIPS is 34% faster than Digital Alpha AXP/150)
Note that 7 of the 8 tests were between 23% and 76% faster on the MIPS.
This average heavily weights the one test that performed poorly on the
Compaq. The component files for MaxEDS were less than 100K in size, and
may have fit well into a 1 MB L2, but not into a 256K L2. Real test
results may vary from that predicted above.
Please note the following about the entry level Compaq DeskPro M 5/60:
The IDE hard disk is considerably slower than Compaq's SCSI offerings.
This may not have affected these benchmarks.
Compaq utilizes a 256K L2 write-through cache of their own design in the
entry model. It is estimated to be 10% slower than a 82430 PCIset or a
82496 Cache Controller of the same size. Larger WB caches would be a
better match agains the 1 MB NEC cache.
The 60 MHz Pentium is estimated to be 10% slower than the 66 MHz Pentium.
The standard Microsoft Compliler does not use all the Pentium
optimizations. Other compiler companies, and Mircosoft in the future,
offer better performance on the Pentium.
While I am not claiming that tweaking a Pentium system will make it faster
than a R4400 system on these benchmarks, I do claim that the gap is not as
big as stated by SGI. I would certainly like to see the results on faster
Pentium systems.
-----------
Len Schultz, lsch...@ichips.intel.com
Microprocessor Division 6, Technical Marketing
Intel Corp., M/S JF1-19, 5200 NE Elam Young Pkwy,
Hillsboro, Oregon 97124-6497
This is a private posting. I do not speak for Intel.
--
-----------
Len Schultz, lsch...@ichips.intel.com
Microprocessor Division 6, Technical Marketing
Intel Corp., M/S JF1-19, 5200 NE Elam Young Pkwy,
Hillsboro, Oregon 97124-6497
Ronald F. Guilmette
>
>NEW 32-BIT APPLICATION SOFTWARE BENCHMARKS SHOW MIPS-BASED
>PCs AS FASTEST FOR MICROSOFT WINDOWS NT
>
>MIPS-Based PCs Run Windows NT Applications Over 200 Percent Faster
>than Intel Pentium and 34 Percent Faster than Alpha AXP
>
>
>MOUNTAIN VIEW, Calif.-September 20, 1993 -MIPS Technologies, Inc.
>today released benchmarking studies which show MIPS-based PCs
>operating native 32-bit Microsoft Windows NT software applications
>The benchmarks are based on the beta versions of four Windows NT applications:
>Texim Project from Welcom Software Technology Corporation; MaxEDS and
>MaxROUTE from Massteck Ltd.; Image-In from CPI, Inc. and PV-WAVE Advantage
>from Visual Numerics Inc.. The applications span a variety of markets,
>including management or administration, scientific and engineering.
The numbers posted are certainly impressive, but for the sake of the work-
station-oriented among us, you should post the SPECmark numbers also.
(Getting the gcc component of the SPECmark suite running on Windows-NT
takes a small amount of fiddling due to the absence of a fork() system
call, but the patches are not that hard to do, and I think I even know
where you might be able to obtain a copy of the necessary patches.)
--
-- Ronald F. Guilmette ------------------------------------------------------
------ domain address: r...@netcom.com ---------------------------------------
------ uucp address: ...!uunet!netcom.com!rfg -------------------------------
Bill Glazier
lsch...@ichips.intel.com (Len Schultz) writes:
|>
|> I am posting this followup to make some clarifications to Bill
|> Glazier's
|> (SGI) posting on the relative performance of the NEC RISCStation 4400
|> and
|> the Compaq DeskPro M 5/60. There was an unfortunate error in the
|> calculation of the relative percentages in execution speed. I shall
|> summarize:
|>
|> Note that 7 of the 8 tests were between 23% and 76% faster on the MIPS.
|> This average heavily weights the one test that performed poorly on the
|> Compaq. The component files for MaxEDS were less than 100K in size,
|> and
|> may have fit well into a 1 MB L2, but not into a 256K L2. Real test
|> results may vary from that predicted above.
|>
|> Please note the following about the entry level Compaq DeskPro M 5/60:
|>
|> The IDE hard disk is considerably slower than Compaq's SCSI
|> offerings.
|> This may not have affected these benchmarks.
Remember, the Compaq Pentium was the SAME PRICE as the MIPS machine. A
SCSI
drive would be more expensive. In any event, it WAS NOT AVAILABLE at time
of purchase.
|>
|> Compaq utilizes a 256K L2 write-through cache of their own design in
|> the
|> entry model. It is estimated to be 10% slower than a 82430 PCIset or
|> a
|> 82496 Cache Controller of the same size. Larger WB caches would be a
|> better match agains the 1 MB NEC cache.
Again, a larger cache would be more expensive. A Larger cache or systems
based on unavailable Intel chipsets was not available at time of purchase.
|>
|> The 60 MHz Pentium is estimated to be 10% slower than the 66 MHz
|> Pentium.
We were unable to purchase a 66mhz Pentium from Compaq. Technology
demonstrations are great (I can demo today a 185 mhz MIPS R4400), but the
reality is no one is shipping Pentium/66 in any volume today.
|>
|> The standard Microsoft Compliler does not use all the Pentium
|> optimizations. Other compiler companies, and Mircosoft in the
|> future,
|> offer better performance on the Pentium.
The MIPS numbers were generated using the standard Microsoft compilers.
We, too, could use proprietary, not commercially available compiler
technology
to boost our numbers. But, that's not what customers see!!!!!!!
Microsoft
chose NOT to use much of the Pentium optimizations for their compiler.
That's
unfortunate, but reality for someone buying an Intel Pentium system today,
and for a while to come.
|>
|> While I am not claiming that tweaking a Pentium system will make it
|> faster
|> than a R4400 system on these benchmarks, I do claim that the gap is not
|> as
|> big as stated by SGI. I would certainly like to see the results on
|> faster
|> Pentium systems.
Not to belittle Len, whose points are well taken, but our benchmarks were
designed to make an apple to apples comparison, as best we could, for
systems
at a given price point which are *currently available*. A comparison
of those systems, today, says that MIPS Technology is roughly 2x the speed
of comparable Intel Pentium technology in a Windows NT environment.
I am sure, over time, Pentium systems will improve, as will MIPS systems.
More compiler optimizations for MIPS and Pentium will be
adopted by Microsoft. And higher frequency, lower cost chips will
come to market. But the fact that the x86 architecture is 'running out
of gas' will remain. Those in the know suggest Intel will dump the x86
instruction set and adopt a true RISC architecture with P6/P7.
We hope to continue to expand our base of application benchmarks so the
real performance differences will be known. Any ISV with NT code that they
wish
us to benchmark should contact me at bgla...@mti.sgi.com.
Bill Glazier
MIPS Technologies
'Becasue Windows NT Is Too IMprtant To Entrust to Intel'
Bill Schneller
>>MIPS Technologies
>>'Becasue Windows NT Is Too IMprtant To Entrust to Intel'
So you are going to trust MicroSoft instead?
Mike O'Connor
Guilmette) writes:
:>NEW 32-BIT APPLICATION SOFTWARE BENCHMARKS SHOW MIPS-BASED
:>PCs AS FASTEST FOR MICROSOFT WINDOWS NT
:>
:>MIPS-Based PCs Run Windows NT Applications Over 200 Percent Faster
:>than Intel Pentium and 34 Percent Faster than Alpha AXP
:
:The numbers posted are certainly impressive, but for the sake of the work-
:station-oriented among us, you should post the SPECmark numbers also.
Windows performance is not just about SPECmarks -- from what I was
reading, it's in the disk and graphics areas that Alpha NT needs work.
...Mike
--
Michael J. O'Connor | Internet: m...@fmsrL7.srL.ford.com
Ford Motor Company, OPEO | UUCP: ...!fmsrl7!opeo!mjo
20000 Rotunda, Bldg. 1-3001 | Phone: +1 (313) 248-1260
Dearborn, MI 48121 | Fax: +1 (313) 323-6277