Yeah, SPEC scores contain real world benchmarks, and assuming that the individual scores on the tests are pretty evenly distributed, the SPEc score would be a pretty good method of determining performance. But sometimes, a processor/chipset combo simply excels at one or two particular components of the benchmark. In that case, real world performance can be vastly different.
The Fugitsu excels at art (image recognition using neural networks, C), the IBM at galgel(computaional fluid dynamics, F90), the SGi at swim (shallow water modeling, F77) and applu (partial differential equations, F77) All three have roughly similar aggregate scores. Yet the strengths of each patform lie in different areas.
(fwiw, the Powermac g5 showed strengths in mesa (3d image library, C), galgel and equake(seismic wave propagation simulation, C). The Intel machines, at least in Veritest's protocols, also performed strongly on those three tests.)
Obligatory Blackadder reference
on
Isn't It Ironic?
·
· Score: 5, Funny
Edmund:Baldrick, have you no idea what irony is?
Baldrick:Yeah, it's like goldy and bronzy, only it's made of iron.
Why don't you read the benchmarks before speculating? Some of the benchmarks are written in Fortran 90. Neither f2c nor g77 (available from fink) support this code, so Apple used NAG Fortran f95 v4.2
All of these are documented in the Veritest report, which includes a complete breakdown of results.
As for the intel compiler, the fact that icc produces good code for AMD processors has been known for some time. For those of us who prefer to use free compilers, the gcc results are still of some interest.
Of course, to really compare "gcc performance" one might choose to subtract out the Fortran programs-- those were compiled by the non-free NagWare Fortran. Or you could choose to compare those propriatary results with scores published on SpecBench. The F90 programs are galgel, facerec, lucas, and fma3d. The F77 programs are wupwise, swim, applu, mgrid, sixtrack, and apsi.
Let's massage the data into submission...
Re:benchmarks; can't live with or without them...
on
G5 Benchmark Roundup
·
· Score: 1
In my experience, PC doofuses have always been big with the benchmarks. It's like a bragging right to them. "I tweaked my dual Smockron 4500 and got it up to 313.3 on SPECdickweed_base!"
Meanwhile, us Mac doofuses (and I use the term with the greatest affection) spend that same time actually working. Because we need the extra cash to feed our $4000-a-year Mac habit.
Sure, certain models of x86 boxen costs less than macs, but the benchmarks cost $500.. I mean validation might well be worth something to the insecure, but is it really worth spending 20% of your budget on a benchmark suite? Remember, spec results are only valid for a machine of identical specification.
It is likely that Microsoft would attempt to verify that this exploit exists before paying. After all, it's traditional to enclose a sample of your art before beginning any decent blackmail campaign. But this sample "exploit" could provpide enough information to both expose the hackers to criminal prosecution, and allow Microsoft to distribute patching programs in future games.
Here's a discussion of Whitney's patent problems. In then end, though, he won his court cases, although, by that time, his patents were within a few years of expiring.
McBride claimed that SCO has the right to audit IBM's customers. "We have other rights under the contract we are looking at. For example, we can audit IBM customers. SCO has audit rights on its customers," he said.
McBride plans to use those audit rights to ensure that
" you simply take the copies of AIX that are out there and send them back to us, or you destroy them and give us notice of the date of destruction. It calls for that in the contract to certify that destruction has taken place."
The migration from AIX to other sources could well be a shot in the arm for US IT firms. It'll be just like the Y2K crisis, except with shorter deadlines and nastier side effects. I think that Wall Street could well afford to be offline for a few months, couldn't it?
make: There *is* an alternative make. I don't rember the name.
BSD make? It's largely incompatible with gnumake. I suppose we can replace those nasty autotools with good old fashioned imake, as well.
emacs: Ok, this is GNU. there is Lucid Emacs/XEmacs.
Is the goal of this excercise to get rid of GPLed stuff, or simply to rid ourselves of GNU utilities?
If it's the former, I'm sure you could, with a lot of effort, graft the linux kernel onto a BSD (thus earning the undying emnity of BSD pedantics who preach about the innate superiority of an integrated operating system over a disoganized group. that simply releases a kernel every few months and calls it an "operating system".
If it's the latter, simply download the sources from GNU and fork them. Announce your intentions as loudly as possible. Mimic the flashiest new feature sets that you can find. Be sure to accept patches from anyone, anywhere, and for heaven's sake, don't ask you developers to assign copyrights to you.
Within a couple of years, people will abandon the GNU toolset in droves, and start to contribute only to your forks. RMS will be left with a library of obsolete, incompatible, and neglected software. The BTG (Better than GNU) toolchain conquers on.
Soon, a SCO will discover that some idiot has copied code from their sources, sue you for copyright infringement, and run your BTG operations into the ground. And then. RMS will gaze fondly at his beloved GNU, and laugh at your misfortune.
Apple are using their compilation and runs of the SPEC benchmark s/w in a way that (they claim) only tests the processor(s), but use the results to justify their claim of 'fastest personal computer' (ie 'system'), NOT, 'fastest processor', so I don't see it as a valid test.
Q4: What components do CINT2000 and CFP2000 measure?
A4: Being compute-intensive benchmarks, they measure performance of the computer's processor, memory architecture and compiler. It is important to remember the contribution of the latter two components -- performance is more than just the processor.
If Apple had turned out record setting Spec scores, it would be perfectly justified in assuming the mantle of "fastest personal computer." Right now, I think they have to settle for the title of "fastest personal computer using GCC", which is less than catchy.
The SPEC benchmarks are of real world applications. They're just not as relevant outside the scientific community. They were originally developed to give researchers a quick way to compare disparate architectures-- so 20k wasn't frittered away on some underperforming workstation. Every few years, some hardware vender figures out a way to fit the benchmarks in cache-- and the benchmarks are subsequently revised to include tougher workloads.
NAG doesn't specify the gcc they used, but they do indicate that the Redhat 9.0 version of NAGware is compiled against glibc-2.3.2. From what I've seen, glibc is usually used in conjunction with gcc-3.2.
The "trial version" is a bit of a red herring. NAG states
These are full and complete NAGWare f95 Compiler implementations. If after the trial period you obtain a full licence from NAG there is no need to reinstall the compiler.
Yep, that's right - the Dell/Intel benchmarks were provided by Apple, and had been taken with Hyperthreading turned off (sugar put in the petrol tank, to use your analogy).
Xeon, we must add, has SMT (HyperThreading) which shouldn't help at all with SPEC data sets, but should help with real performance.
Indeed, these benchmarks show that hypethreading doesn't make much of a difference at all.
With hyperthreading, a pentium IV 3.066 Ghz (850EMVR chipset) achieves 1073 in floating point, 1085 in integer. Without, it achieves 1069 floating point and 1085 integer.
The astute observer will notice that these scores outdistance apple's gcc 3.3 tests by a considerable margin, but the difference is not due to hyperthreading.
Now scroll down. Notice the ASUS P4PE system. This system also uses a 3.066 Ghz Pentium IV, but boasts only a 872 fp, and 1034 integer score. And the tester used Intel compilers!
It's still a far cry from 889 int/693 fp (Dell Dimension 8300) but perhaps it demonstrates that Hyperthreading doesn't mean shit when computing spec2000 and that spec2000 scores vary a great deal from platform to platform.
The 640 k limit derives from the fact that the 8086 and 8088 used a 20 bit addressing nus, allowing the computer to address 1,048,576 bytes. As the original IBM PC used a the eight bit 8088 chip, this address space was divided in 16 pages of 65536 bytes each. Of that original meg, 384k (the upper memory) was set aside for video RAM (128k), bios ROM, and other miscellaneous features. This left 640k for real programs.
The PC was soon succeded by the PC/AT, which used the 16bit 80286. The 80286 could be used in one of two modes-- real (which was architecturally identical to the 8086, and thus limited to a 20 bit address bus), or protected virtual mode, which could use a 24 bit address bus to address 16MB of memory.
Unfortunately, DOS doesn't run in protected mode, and thus the programmer is foced to use all sorts of hacks to access extended memory (memory beyond 1024 k). On the 8086, one could use expanded memory, which switched up to four 16k pages of memory into the upper memory area.
For some odd reason, the 80286 while running in real mode, is also able to address another bank between 1024k and 1078k This is called high memory. Portions of DOS can remain resident here, freeing up 45k of lower memory. Woohoo!
The release of Panther doesn't make your copy of Jaguar any less useful - it doesn't detract from Jaguar's functionality, ease of use or anything else.
That's assuming that: 1) the developer tools will continue to be compatible with Jaguar, and 2) that any security flaws found in Jaguar will be unexploited.
mpeg4 is out; pixlet is in
on
Jaguar is Over
·
· Score: 2, Funny
I wasn't at MacWorld, but I did watch the speech over my 56K line. I was most impressed with the pixlet component of the presentation-- absolutely beautiful.
Slashdot Mirror
sigh, I recant my statement.
Half a dozen variations of the quote have appeared on the Internet, all credited to Baldrick. But the BBC America siteagrees with me.
The Fugitsu excels at art (image recognition using neural networks, C), the IBM at galgel(computaional fluid dynamics, F90), the SGi at swim (shallow water modeling, F77) and applu (partial differential equations, F77) All three have roughly similar aggregate scores. Yet the strengths of each patform lie in different areas.
(fwiw, the Powermac g5 showed strengths in mesa (3d image library, C), galgel and equake(seismic wave propagation simulation, C). The Intel machines, at least in Veritest's protocols, also performed strongly on those three tests.)
Edmund:Baldrick, have you no idea what irony is?
Baldrick:Yeah, it's like goldy and bronzy, only it's made of iron.
from Amy and Amiability
Why don't you read the benchmarks before speculating? Some of the benchmarks are written in Fortran 90. Neither f2c nor g77 (available from fink) support this code, so Apple used NAG Fortran f95 v4.2
Yes, there is a GNU Fortran95 compiler, but it's "in a pupal state."
is here
The G5 results are
SPECfp_base2000: 840
SPECint_base2000: 800
SPECfp_rate_base2000: 15.7
SPECint_rate_base2000:17.2
All of these are documented in the Veritest report, which includes a complete breakdown of results.
As for the intel compiler, the fact that icc produces good code for AMD processors has been known for some time. For those of us who prefer to use free compilers, the gcc results are still of some interest.
Of course, to really compare "gcc performance" one might choose to subtract out the Fortran programs-- those were compiled by the non-free NagWare Fortran. Or you could choose to compare those propriatary results with scores published on SpecBench.
The F90 programs are galgel, facerec, lucas, and fma3d. The F77 programs are wupwise, swim, applu, mgrid, sixtrack, and apsi.
Let's massage the data into submission...
In my experience, PC doofuses have always been big with the benchmarks. It's like a bragging right to them. "I tweaked my dual Smockron 4500 and got it up to 313.3 on SPECdickweed_base!"
Meanwhile, us Mac doofuses (and I use the term with the greatest affection) spend that same time actually working. Because we need the extra cash to feed our $4000-a-year Mac habit.
Sure, certain models of x86 boxen costs less than macs, but the benchmarks cost $500.. I mean validation might well be worth something to the insecure, but is it really worth spending 20% of your budget on a benchmark suite? Remember, spec results are only valid for a machine of identical specification.
Monty Python did it best
It is likely that Microsoft would attempt to verify that this exploit exists before paying. After all, it's traditional to enclose a sample of your art before beginning any decent blackmail campaign. But this sample "exploit" could provpide enough information to both expose the hackers to criminal prosecution, and allow Microsoft to distribute patching programs in future games.
Surely, the more money the US spends on space, the less it can spend on waging war? Or am I missing something?
Apparently, you are unfamiliar with the concept of "running a deficit".
Here's a discussion of Whitney's patent problems. In then end, though, he won his court cases, although, by that time, his patents were within a few years of expiring.
McBride plans to use those audit rights to ensure that
The migration from AIX to other sources could well be a shot in the arm for US IT firms. It'll be just like the Y2K crisis, except with shorter deadlines and nastier side effects. I think that Wall Street could well afford to be offline for a few months, couldn't it?
The hour long discussion has been archived on mp3
Wouldn't a grammar Nazi capitalize every noun, as in German?
make: There *is* an alternative make. I don't rember the name.
BSD make? It's largely incompatible with gnumake. I suppose we can replace those nasty autotools with good old fashioned imake, as well.
emacs: Ok, this is GNU.
there is Lucid Emacs/XEmacs.
Is the goal of this excercise to get rid of GPLed stuff, or simply to rid ourselves of GNU utilities?
If it's the former, I'm sure you could, with a lot of effort, graft the linux kernel onto a BSD (thus earning the undying emnity of BSD pedantics who preach about the innate superiority of an integrated operating system over a disoganized group. that simply releases a kernel every few months and calls it an "operating system".
If it's the latter, simply download the sources from GNU and fork them. Announce your intentions as loudly as possible. Mimic the flashiest new feature sets that you can find. Be sure to accept patches from anyone, anywhere, and for heaven's sake, don't ask you developers to assign copyrights to you.
Within a couple of years, people will abandon the GNU toolset in droves, and start to contribute only to your forks. RMS will be left with a library of obsolete, incompatible, and neglected software. The BTG (Better than GNU) toolchain conquers on.
Soon, a SCO will discover that some idiot has copied code from their sources, sue you for copyright infringement, and run your BTG operations into the ground. And then. RMS will gaze fondly at his beloved GNU, and laugh at your misfortune.
From the SpecCPU 2000 Faq
If Apple had turned out record setting Spec scores, it would be perfectly justified in assuming the mantle of "fastest personal computer." Right now, I think they have to settle for the title of "fastest personal computer using GCC", which is less than catchy.
The SPEC benchmarks are of real world applications. They're just not as relevant outside the scientific community. They were originally developed to give researchers a quick way to compare disparate architectures-- so 20k wasn't frittered away on some underperforming workstation. Every few years, some hardware vender figures out a way to fit the benchmarks in cache-- and the benchmarks are subsequently revised to include tougher workloads.
Dixie_Flatline wrote:
If you don't like it, ignore it. All negative mods are meta-moderated 'unfair'. ALL OF THEM
Doing your best to feed the trolls, I see...
The "trial version" is a bit of a red herring. NAG states
Eh? The Register speculates
Indeed, these benchmarks show that hypethreading doesn't make much of a difference at all.
With hyperthreading, a pentium IV 3.066 Ghz (850EMVR chipset) achieves 1073 in floating point, 1085 in integer. Without, it achieves 1069 floating point and 1085 integer.
The astute observer will notice that these scores outdistance apple's gcc 3.3 tests by a considerable margin, but the difference is not due to hyperthreading.
Now scroll down. Notice the ASUS P4PE system. This system also uses a 3.066 Ghz Pentium IV, but boasts only a 872 fp, and 1034 integer score. And the tester used Intel compilers!
It's still a far cry from 889 int/693 fp (Dell Dimension 8300) but perhaps it demonstrates that Hyperthreading doesn't mean shit when computing spec2000 and that spec2000 scores vary a great deal from platform to platform.
The 640 k limit derives from the fact that the 8086 and 8088 used a 20 bit addressing nus, allowing the computer to address 1,048,576 bytes. As the original IBM PC used a the eight bit 8088 chip, this address space was divided in 16 pages of 65536 bytes each. Of that original meg, 384k (the upper memory) was set aside for video RAM (128k), bios ROM, and other miscellaneous features. This left 640k for real programs.
The PC was soon succeded by the PC/AT, which used the 16bit 80286. The 80286 could be used in one of two modes-- real (which was architecturally identical to the 8086, and thus limited to a 20 bit address bus), or protected virtual mode, which could use a 24 bit address bus to address 16MB of memory.
Unfortunately, DOS doesn't run in protected mode, and thus the programmer is foced to use all sorts of hacks to access extended memory (memory beyond 1024 k). On the 8086, one could use expanded memory, which switched up to four 16k pages of memory into the upper memory area.
For some odd reason, the 80286 while running in real mode, is also able to address another bank between 1024k and 1078k This is called high memory. Portions of DOS can remain resident here, freeing up 45k of lower memory. Woohoo!
It's actuall $799, for the eMac.
The release of Panther doesn't make your copy of Jaguar any less useful - it doesn't detract from Jaguar's functionality, ease of use or anything else.
That's assuming that:
1) the developer tools will continue to be compatible with Jaguar, and
2) that any security flaws found in Jaguar will be unexploited.
I wasn't at MacWorld, but I did watch the speech over my 56K line. I was most impressed with the pixlet component of the presentation-- absolutely beautiful.