Slashdot Mirror

If I go to the SPEC web page, I can look up a "All SPEC CINT2000 Results Published by SPEC". This table tells us the following:

Compaq Computer Corp AlphaServer DS20E Model 6/667
1 cpu: base 424 peak 444

Intel Corporation Intel VC820 (1.0 GHz Pentium III)
1 cpu: base 407 peak 410

There's also "All SPEC CFP2000 Results Published by SPEC". It says:

Compaq Computer Corp AlphaServer DS20E Model 6/667
1 cpu: base 514 peak 577

Intel Corporation Intel VC820 (1.0 GHz MHz Pentium III
1 cpu: base 273 peak 284

So yes, you're right. It's somewhat unfortunate that AMD hasn't published specs for their Athlon CPU using the industry-standard SPEC benchmarks (Instead using Winbench, which I spit upon) because they're reputed to have 16% faster SPECfp scores than Intel. On top of that, it looks like the new athlons coming out with full speed cache are a little over ten percent faster than the other models, at least as far as integer math goes. This seems to translate into something like a 11% increase in game performance, which is all *I* really care about.

So it looks like the new Alpha chip is 4-8% faster at integer math than the P3; The division is wider for floating point, at 47-51%, which is pretty dramatic. It looks like the numbers back you up.

However, read the following clip from this page at Compaq Canada:

Here is an offer that AlphaServer 1000/ 1000A or 1200 systems owners may find hard to refuse. Upgrade to a powerful AlphaServer DS20E system and get a (CDN) $7,500 instant rebate when you spend at least (CDN) $60,000.

This indicates to me that it's probably not that hard to spend CDN$60,000 on one of those systems. That translates (At today's rates) to US$40,655.92.

By comparison, if I price out "Compaq ProLiant CL380 Intel® Pentium® III 800MHz/256-Tower Model 128MB Qty=1 Proc, NC3123 per Node" with two 800mhz P3s per node, over 800mb of ram per node, a nice raid5 made up of 9gb disks, and so on, it comes to US$40,878.00. The "per node" I mentioned is because this is actually two Dual P3 systems in one rackmount unit, intended to work as a failover box (I assume.) That system is pretty well loaded at this price and configuration.

Unfortunately, there's no way to get a price on an ALPHAserver without actually talking to a salesmonkey at Compaq, so I can't tell you what a system with 1.6gb of ram, a raid controller (with only 4 9.1gb disks, mind you, should be cheap) and so on will cost. Hmm, there's a 404 clicking on a link on Compaq's page... oh, never mind.

One Alpha chip is faster than one P3 chip, or even one athlon (though the athlon might have better integer math... AMD has always been pretty good at that.) I'm guessing that one Alpha chip will cost you more than two P3 chips, based on what I've seen of the pricing on Alpha equipment in the past. With PC systems, you also have a lot more flexibility in who you can purchase from, interoperability, and so on. With the Alpha systems, you can't just go down to Fry's and buy a new mainboard.

The Alpha chip certainly is faster. Is it cheaper per SPECint2000 or SPECfp2000 point? Probably not.

If I go to the SPEC web page, I can look up a "All SPEC CINT2000 Results Published by SPEC". This table tells us the following:

Compaq Computer Corp AlphaServer DS20E Model 6/667
1 cpu: base 424 peak 444

Intel Corporation Intel VC820 (1.0 GHz Pentium III)
1 cpu: base 407 peak 410

There's also "All SPEC CFP2000 Results Published by SPEC". It says:

Compaq Computer Corp AlphaServer DS20E Model 6/667
1 cpu: base 514 peak 577

Intel Corporation Intel VC820 (1.0 GHz MHz Pentium III
1 cpu: base 273 peak 284

So yes, you're right. It's somewhat unfortunate that AMD hasn't published specs for their Athlon CPU using the industry-standard SPEC benchmarks (Instead using Winbench, which I spit upon) because they're reputed to have 16% faster SPECfp scores than Intel. On top of that, it looks like the new athlons coming out with full speed cache are a little over ten percent faster than the other models, at least as far as integer math goes. This seems to translate into something like a 11% increase in game performance, which is all *I* really care about.

So it looks like the new Alpha chip is 4-8% faster at integer math than the P3; The division is wider for floating point, at 47-51%, which is pretty dramatic. It looks like the numbers back you up.

However, read the following clip from this page at Compaq Canada:

Here is an offer that AlphaServer 1000/ 1000A or 1200 systems owners may find hard to refuse. Upgrade to a powerful AlphaServer DS20E system and get a (CDN) $7,500 instant rebate when you spend at least (CDN) $60,000.

This indicates to me that it's probably not that hard to spend CDN$60,000 on one of those systems. That translates (At today's rates) to US$40,655.92.

By comparison, if I price out "Compaq ProLiant CL380 Intel® Pentium® III 800MHz/256-Tower Model 128MB Qty=1 Proc, NC3123 per Node" with two 800mhz P3s per node, over 800mb of ram per node, a nice raid5 made up of 9gb disks, and so on, it comes to US$40,878.00. The "per node" I mentioned is because this is actually two Dual P3 systems in one rackmount unit, intended to work as a failover box (I assume.) That system is pretty well loaded at this price and configuration.

Unfortunately, there's no way to get a price on an ALPHAserver without actually talking to a salesmonkey at Compaq, so I can't tell you what a system with 1.6gb of ram, a raid controller (with only 4 9.1gb disks, mind you, should be cheap) and so on will cost. Hmm, there's a 404 clicking on a link on Compaq's page... oh, never mind.

One Alpha chip is faster than one P3 chip, or even one athlon (though the athlon might have better integer math... AMD has always been pretty good at that.) I'm guessing that one Alpha chip will cost you more than two P3 chips, based on what I've seen of the pricing on Alpha equipment in the past. With PC systems, you also have a lot more flexibility in who you can purchase from, interoperability, and so on. With the Alpha systems, you can't just go down to Fry's and buy a new mainboard.

The Alpha chip certainly is faster. Is it cheaper per SPECint2000 or SPECfp2000 point? Probably not.

If I go to the SPEC web page, I can look up a "All SPEC CINT2000 Results Published by SPEC". This table tells us the following:

Compaq Computer Corp AlphaServer DS20E Model 6/667
1 cpu: base 424 peak 444

Intel Corporation Intel VC820 (1.0 GHz Pentium III)
1 cpu: base 407 peak 410

There's also "All SPEC CFP2000 Results Published by SPEC". It says:

Compaq Computer Corp AlphaServer DS20E Model 6/667
1 cpu: base 514 peak 577

Intel Corporation Intel VC820 (1.0 GHz MHz Pentium III
1 cpu: base 273 peak 284

So yes, you're right. It's somewhat unfortunate that AMD hasn't published specs for their Athlon CPU using the industry-standard SPEC benchmarks (Instead using Winbench, which I spit upon) because they're reputed to have 16% faster SPECfp scores than Intel. On top of that, it looks like the new athlons coming out with full speed cache are a little over ten percent faster than the other models, at least as far as integer math goes. This seems to translate into something like a 11% increase in game performance, which is all *I* really care about.

So it looks like the new Alpha chip is 4-8% faster at integer math than the P3; The division is wider for floating point, at 47-51%, which is pretty dramatic. It looks like the numbers back you up.

However, read the following clip from this page at Compaq Canada:

Here is an offer that AlphaServer 1000/ 1000A or 1200 systems owners may find hard to refuse. Upgrade to a powerful AlphaServer DS20E system and get a (CDN) $7,500 instant rebate when you spend at least (CDN) $60,000.

This indicates to me that it's probably not that hard to spend CDN$60,000 on one of those systems. That translates (At today's rates) to US$40,655.92.

By comparison, if I price out "Compaq ProLiant CL380 Intel® Pentium® III 800MHz/256-Tower Model 128MB Qty=1 Proc, NC3123 per Node" with two 800mhz P3s per node, over 800mb of ram per node, a nice raid5 made up of 9gb disks, and so on, it comes to US$40,878.00. The "per node" I mentioned is because this is actually two Dual P3 systems in one rackmount unit, intended to work as a failover box (I assume.) That system is pretty well loaded at this price and configuration.

Unfortunately, there's no way to get a price on an ALPHAserver without actually talking to a salesmonkey at Compaq, so I can't tell you what a system with 1.6gb of ram, a raid controller (with only 4 9.1gb disks, mind you, should be cheap) and so on will cost. Hmm, there's a 404 clicking on a link on Compaq's page... oh, never mind.

One Alpha chip is faster than one P3 chip, or even one athlon (though the athlon might have better integer math... AMD has always been pretty good at that.) I'm guessing that one Alpha chip will cost you more than two P3 chips, based on what I've seen of the pricing on Alpha equipment in the past. With PC systems, you also have a lot more flexibility in who you can purchase from, interoperability, and so on. With the Alpha systems, you can't just go down to Fry's and buy a new mainboard.

The Alpha chip certainly is faster. Is it cheaper per SPECint2000 or SPECfp2000 point? Probably not.

Yes, MHz does make a difference, but people need to please stop touting SPEC results without taking into account other factors when comparing CPU performance...

SPEC benchmarks are designed with minuscule datasets to reduce RAM and Cache bottlenecks... some interesting articles found at the STREAM homepage discuss how some CPU manufacturers boosted L2 cache on their chips, while ignoring RAM bandwidth considerations, simply to get higher SPEC results...
Memory bandwidth results (MByte/s) for recent HP, RS/6000, and esp. Alphas single-CPU workstations show they can play around with much more data located in RAM alot quicker than a 733 PIII, even at low clock speeds (400MHz for POWER3 and PA-RISC 8500... the Alphas were 21264s clocked at equal to or slightly less than 733MHz, with results being about double those of the PIII)... note that the PIII wasn't a xeon... shouldn't make a big diff though, because the architecture is similar (nearly identical) for PIII and PIII-xeon... look at the results yourselves, it's innarestin...

that being said, all Sun Ultra workstations performed a little worse than 'equivalent' HP, IBM, and DEC(Compaq) workstations regarding RAM bandwidth... Ultra60-360s perform so poorly that the PIII 733 gets twice as many MFLOPS on a particular test (but the lead is much less on two others... a 450MHz-Ultra might tie or surpass them)...

And ALL that to say that SPEC is only useful if you're comparing systems which will be doing computations on teeny tiny datasets. At least that's the case for SPECcpu95, I don't know about SPECcpu2000. Furthermore, the SPECfp benchmarks focus mainly on double-precision floats, to the expense of single-precision floats... this might indicate why results for SGI machines make them look pokey, considering some of their CPUs (r5000) are optimized for single-precision MADD instructions, because of their ubiquity in doing 3D work...

Here are some sites that contain benchmark results and/or link to sites with benchmark results:
SPEC website
the CPU Info Center
FutureTech SGI info

I can't understand why a lot of people here seem to think that Sun is the all-knowing, untouchable God of Microprocessor/System design. Sun's systems consistently underperform when matched up against similar systems from almost all of their competitors.

Don't take my word for it, check out some of the benchmark results from the SPEC website. Sun's products are always near the bottom of the pile.

Not only are their products inferior, but they can't seem to deliver a new processor and they keep pushing back the schedule for the UltraSPARC III. I'm no Intel fan but I don't think that even they could learn anything from Sun's engineering.

There already is a great benchmark for processors, and it's called SPEC. Yes, it's not open source, but it's really quite reliable for comparing CPUs of any architecture. As slashdot user "cweber" pointed out in his post, they have been doing this for 11 years, and they periodically revise their benchmark suite to stress CPUs more uniformly.

The open-source method. This is really good to ensure that there are no cheaters at the benchmark level.

Tom's interesting ideas on Crusoe. This stems from the fact that SPECmarks don't quite approximate real usage that Crusoe depends on to use it's hotspot optimizations. However, we are interested in the raw sustained speed of the processor (in this case), not the speed of the OS or it's task swap latency. Tough problems to solve.

Open-source means that the benchmark code will be able to take advantage of the best compiler available for the target CPU (see comment at end).

The Bad:

Anyone who has done benchmarks knows that even small variations in system config can have strage or harmful effects on the benchmark results. This open-source effort is going to have to have a database of hardware configs in order for this to be useful.

The Ugly:

Vendors are going to oppose this (at least not support it). Why? Because plain and simple they have an interest in promoting the most favorable statistics possible about their products. They want to keep feeding you "polygon fill rates" and "texels per second" because their card may not stand up in a direct test program comparison. Plus, they are just dying to convince you that they have new BogusMarketingAcronym (tm) technology and their competitor does not. Nevermind that SSE and 3Dnow do pretty much the same thing -- companies have an interest in differentiating themselves as much as possible.

If this benchmark actually takes off (and gets widely accepted), we might get cheaters at the firmware or hardware level. This has happened before -- although which company it was and which benchmark they cheated I can't remember. I can't find it on the net or remember to save my life (sigh)...

I also need to say something to the people who think a processor should be judged independently of a compiler. This is just plain dumb. Why? Because a processor and it's compiler are a team. You can't use one without the other. When a chip is designed, there is a direct information dependence between the chip architects and the compiler writers. They are designed as a pair (ideally), and they should be tested as such. If a given compiler has great optimizations, then great! That means the compiler understands its target real well. It is a win for both the CPU and the compiler for pulling it off. This compiler is going to do the same kinds of optimizations when vendors use it to write programs, so that helps the comparison between benchmark code and apps.

However, I can see the need to compare not only the best compiler, but GCC as well, because of its broad acceptance. But if you are serious about performance, and want to get every once of juice out of your chip, you use the vendor provided compilers, not GCC. Don't get me wrong, GCC is great for compliance and portability, but it usually doesn't compare well with vendor compilers for generated code speed (with the possible exception of IA-32).

Ars Technica also published, a while back, some good information regarding CPU benchmarks. Check it out if you are interested in SPEC or CPU benchmarks in general.

uh no, try again:
http://www.spec.org/cgi-bin/order/
those dont look like open source prices to me...

--Siva

Keyboard not found.

Why would anyone use an alpha processor?

Well performance for one. Have you seen the specs on these things?
And like the other Anonymous Coward pointed out: quality.
You have to think of this not as a Compaq product but as one from Digital Semiconductor. The guys who not only created Alpha, but StrongARM, PCI Bridge, and other networking components (e.g. Tulip). To me, Digital Semiconductor is the most respected name in the industry.

Now the real question: Why would anyone use an Intel processor?

I'll be the first to point out that I have limited experiance with the x86 architecture. But I'll also be the first to point out that this architecture is terrible! I mean when I think about the performance (and quality) that goes into the majority of the worlds desktops -- as an engineer and geek I'm disgusted!

Note that there is a new 1U rack version of the DS10, called the DS10L (code-named "Slate"), that is very attractive for highly compute-intensive tasks. There's a picture of a rack full of these in the Linux section of Compaq's web site.

Can you point to any independent results to back up this claim? I can point you to the Standard Performance Evaluation Corporation (http://www.spec.org), which has published benchmark results indicating a 500MHz G4 is about as fast as a 500MHz Pentium III.

Mind you, I'm not saying all CPUs running at the same frequency are equal, only that the latest Pentium and PowerPC CPUs happen to be. A 500MHz Alpha 21264 is quite a bit faster than either, and a 500MHz PA-RISC 8600 is considerably faster still.

Sheesh - go look at this and see that Mhz nowdays (has it ever?) is irrelevant. A 733 Intel CPU gets a rating of 336 while an Alpha 667 get 413!!! And I forgot where I saw it, but Intel says the 1Ghz rates a 410 on this. STILL lower than the 667 Alpha. (on par, but lower)Total marketing BS. I'll admit the speed wars are great, but let's remember what Intel ISN't telling you - that they HAVE to run their chips faster because they're inherently slower designs (sorry for the rant)

The closest thing we have to a universal measurement of CPU performance is the SPEC 95 benchmark. If you want to see how the various processors measure up, look them up there. SPEC has also come out with a newer benchmark, SPEC CPU 2000, but there aren't many submissions for the newer benchmark, and imo its SPECfp part is broken (too integer-intensive). I personally really really doubt the value of dividing benchmark results by MHz and calling it MHz "efficiency", but you wouldn't be the first to do it.

-- Guges --

The closest thing we have to a universal measurement of CPU performance is the SPEC 95 benchmark. If you want to see how the various processors measure up, look them up there. SPEC has also come out with a newer benchmark, SPEC CPU 2000, but there aren't many submissions for the newer benchmark, and imo its SPECfp part is broken (too integer-intensive). I personally really really doubt the value of dividing benchmark results by MHz and calling it MHz "efficiency", but you wouldn't be the first to do it.

-- Guges --

Check out the SPEC marks:
http://www.spec.org
An Alpha 21264-based machine at 667 MHz is about twice as fast as a 733 MHz Pentium III with 800 MHz RDRAM. You do the math.

Taking a look at Intel's own posted benchmarks of the 800mhz Coppermine running on a 133mhz bus gives:

SPECINT - 38.4 SPECfp - 28.9

The fastest one is one in a Dell system, which has the numbers I originally quoted (SPECint: 38.9, SPECfp: 32.4). Proof is here for SPECint and here for SPECfp .

Now, I'd compare these to AMD's benchmarks, but AMD hasn't published SPECINT results, and only publishes the base SPECfp results. (Which, by the way, show the Athlon soundly thrashing the Coppermine

You are a liar. AMD Athlon 750 MHz, SPECint: 33.0, SPECfp: 26.5. (800 MHz Coppermine: 38.9 and 32.4). Proof: SPECint and SPECfp . AMD hasn't published results for the 800 or the 850, because those processors have such poor performance.

In any event, comparing SPEC scores is a rather _bad_ way to judge system performance. If you know enough to extrapolate new benchmark scores from current ones, you should also know that there are much better real world tests available.

SPEC is not the be-all and end-all of benchmarks but it is *the* standard benchmark for scientific commputing, and by far the most respected CPU benchmark in the world. I would *really* love to see TPC-C results for Athlon, but they haven't been published yet (gee, I wonder why?)

If you like, you can watch a 700mhz Athlon kick the snot out of a 733mhz PIII Coppermine running on a 133mhz bus over at Ace's Hardware

And I should trust some ma and pa benchmarks over the most professional and industry standard benchmarks for what reason? For starters, spec is a dot-org and Ace's is a dot-com, so SPEC is inherently less biased. I don't trust benchmarks from dot-com sites.

So next time, please take your results somwhere else or provide a real source for them.

Well, it is next time, and the source is all above.

Taking a look at Intel's own posted benchmarks of the 800mhz Coppermine running on a 133mhz bus gives:

SPECINT - 38.4 SPECfp - 28.9

The fastest one is one in a Dell system, which has the numbers I originally quoted (SPECint: 38.9, SPECfp: 32.4). Proof is here for SPECint and here for SPECfp .

Now, I'd compare these to AMD's benchmarks, but AMD hasn't published SPECINT results, and only publishes the base SPECfp results. (Which, by the way, show the Athlon soundly thrashing the Coppermine

You are a liar. AMD Athlon 750 MHz, SPECint: 33.0, SPECfp: 26.5. (800 MHz Coppermine: 38.9 and 32.4). Proof: SPECint and SPECfp . AMD hasn't published results for the 800 or the 850, because those processors have such poor performance.

In any event, comparing SPEC scores is a rather _bad_ way to judge system performance. If you know enough to extrapolate new benchmark scores from current ones, you should also know that there are much better real world tests available.

SPEC is not the be-all and end-all of benchmarks but it is *the* standard benchmark for scientific commputing, and by far the most respected CPU benchmark in the world. I would *really* love to see TPC-C results for Athlon, but they haven't been published yet (gee, I wonder why?)

If you like, you can watch a 700mhz Athlon kick the snot out of a 733mhz PIII Coppermine running on a 133mhz bus over at Ace's Hardware

And I should trust some ma and pa benchmarks over the most professional and industry standard benchmarks for what reason? For starters, spec is a dot-org and Ace's is a dot-com, so SPEC is inherently less biased. I don't trust benchmarks from dot-com sites.

So next time, please take your results somwhere else or provide a real source for them.

Well, it is next time, and the source is all above.

Taking a look at Intel's own posted benchmarks of the 800mhz Coppermine running on a 133mhz bus gives:

SPECINT - 38.4 SPECfp - 28.9

The fastest one is one in a Dell system, which has the numbers I originally quoted (SPECint: 38.9, SPECfp: 32.4). Proof is here for SPECint and here for SPECfp .

Now, I'd compare these to AMD's benchmarks, but AMD hasn't published SPECINT results, and only publishes the base SPECfp results. (Which, by the way, show the Athlon soundly thrashing the Coppermine

You are a liar. AMD Athlon 750 MHz, SPECint: 33.0, SPECfp: 26.5. (800 MHz Coppermine: 38.9 and 32.4). Proof: SPECint and SPECfp . AMD hasn't published results for the 800 or the 850, because those processors have such poor performance.

In any event, comparing SPEC scores is a rather _bad_ way to judge system performance. If you know enough to extrapolate new benchmark scores from current ones, you should also know that there are much better real world tests available.

SPEC is not the be-all and end-all of benchmarks but it is *the* standard benchmark for scientific commputing, and by far the most respected CPU benchmark in the world. I would *really* love to see TPC-C results for Athlon, but they haven't been published yet (gee, I wonder why?)

If you like, you can watch a 700mhz Athlon kick the snot out of a 733mhz PIII Coppermine running on a 133mhz bus over at Ace's Hardware

And I should trust some ma and pa benchmarks over the most professional and industry standard benchmarks for what reason? For starters, spec is a dot-org and Ace's is a dot-com, so SPEC is inherently less biased. I don't trust benchmarks from dot-com sites.

So next time, please take your results somwhere else or provide a real source for them.

Well, it is next time, and the source is all above.

Taking a look at Intel's own posted benchmarks of the 800mhz Coppermine running on a 133mhz bus gives:

SPECINT - 38.4 SPECfp - 28.9

The fastest one is one in a Dell system, which has the numbers I originally quoted (SPECint: 38.9, SPECfp: 32.4). Proof is here for SPECint and here for SPECfp .

Now, I'd compare these to AMD's benchmarks, but AMD hasn't published SPECINT results, and only publishes the base SPECfp results. (Which, by the way, show the Athlon soundly thrashing the Coppermine

You are a liar. AMD Athlon 750 MHz, SPECint: 33.0, SPECfp: 26.5. (800 MHz Coppermine: 38.9 and 32.4). Proof: SPECint and SPECfp . AMD hasn't published results for the 800 or the 850, because those processors have such poor performance.

In any event, comparing SPEC scores is a rather _bad_ way to judge system performance. If you know enough to extrapolate new benchmark scores from current ones, you should also know that there are much better real world tests available.

SPEC is not the be-all and end-all of benchmarks but it is *the* standard benchmark for scientific commputing, and by far the most respected CPU benchmark in the world. I would *really* love to see TPC-C results for Athlon, but they haven't been published yet (gee, I wonder why?)

If you like, you can watch a 700mhz Athlon kick the snot out of a 733mhz PIII Coppermine running on a 133mhz bus over at Ace's Hardware

And I should trust some ma and pa benchmarks over the most professional and industry standard benchmarks for what reason? For starters, spec is a dot-org and Ace's is a dot-com, so SPEC is inherently less biased. I don't trust benchmarks from dot-com sites.

So next time, please take your results somwhere else or provide a real source for them.

Well, it is next time, and the source is all above.

From this and this you can see that the GeForce is on par with a $38k SGI.

rbf who is typing this on an Alpha running Debian GNU/Linux 2.1 with Linux 2.2.13.

This was true, but is now only true for floating point. Just look at the SpecInt results.

• AlphaServer GS60E Model 6/700, 700 MHz 21264A , L1: 64KB(I)+64KB(D) on chip , L2: 8MB (I+D) off chip, RAM: 8GB (4x2GB), OS: Tru64 UNIX V4.0F , result: SPECint = 39.1, SPECint_base : 34.7
• Microstar MS-6167 mainboard, AMD Athlon processor 650MHz, L1: 64KBI + 64KBD , L2: 512KB(I+D) @ 1/2 CPU clock speed, RAM: 128 MB PC-100 SDRAM, OS: Windows NT 4.0 Workstation (Service Pack 5), SPECint: 29.4, SPECbase: 29.4

The gap not very big: 30% at best, in reality maybe 10%, when using gcc in both cases. The difference is that with the Athlon you can run StarOffice, or W95+ software DVD player, and millions of other software with Linux, Windows NT or W98.

Which failure of the Athlon ??? Just have a look at SpecINT results. Athlon is competitive with the Alpha Mhz for Mhz, and is near the top. In fact the Athlon could be currently the second fastest microprocessor that ever existed on planet Earth (the first being the Alpha). Of course for floating point, the Athlon is well behind the Alpha, but the main point is that the gap is closing between RISC systems and IA32: comparing the 49 SPECfp for the latest Alpha to the 13 SPECfp for the PIII-550, one might choose to go for the Alpha ; but compared to the 22 SPECfp of the Athlon-650, one could seriously hesitate.

I don't know how IBM are comparing their performance to Sun's Starfire. Pretty meaningless without giving any details. btw the Starfire is over 2 years old now. I don't think Sun are yet officially supporting their 450Mhz UltraSparc-II in volume on Starfire yet, even though it's been out for a while. (btw, you can get US-II 450's with 8MByte of 2nd level cache - clocked at 450Mhz!) Also, Sun's UltraSparc-III will be shipping in volume this December, starting at 600Mhz, and from early SPEC 95 benchmarks I've heard of it's about 10% faster (in fp) than an 600Mhz EV67 (Alpha 21264A) and they haven't even finished optimising it yet. SPEC int should be very good too.

Latest SPEC results here - 600Mhz Athlon has SPEC int/fp of 27.2/21.6. 667Mhz EV67 (Alpha 21264A) has SPEC int/fp of 37.5/65.5. The more competition, the better! (that includes competition between OS's) .

Take a look at Intel's Compiler.. If "Intel is throwing support towards projects that are things I care about" where does this fit in?

Also, take a look at some SPEC results for Intel Hardware and notice that most of the results (even Intel's newest entry are with Intel's compiler. If your really interested, look around on the SPEC site. Even AMD reports results using Intel's compiler.

Maybe you should look at Apple, at least they use GCC as thier compiler.

Take a look at Intel's Compiler.. If "Intel is throwing support towards projects that are things I care about" where does this fit in?

Also, take a look at some SPEC results for Intel Hardware and notice that most of the results (even Intel's newest entry are with Intel's compiler. If your really interested, look around on the SPEC site. Even AMD reports results using Intel's compiler.

Maybe you should look at Apple, at least they use GCC as thier compiler.

Take a look at Intel's Compiler.. If "Intel is throwing support towards projects that are things I care about" where does this fit in?

Also, take a look at some SPEC results for Intel Hardware and notice that most of the results (even Intel's newest entry are with Intel's compiler. If your really interested, look around on the SPEC site. Even AMD reports results using Intel's compiler.

Maybe you should look at Apple, at least they use GCC as thier compiler.

Take a look at Intel's Compiler.. If "Intel is throwing support towards projects that are things I care about" where does this fit in?

Also, take a look at some SPEC results for Intel Hardware and notice that most of the results (even Intel's newest entry are with Intel's compiler. If your really interested, look around on the SPEC site. Even AMD reports results using Intel's compiler.

Maybe you should look at Apple, at least they use GCC as thier compiler.

Here's a short summary for those not inclined to look it up for themselves:

450 Mhz 7400(G4) 21.4 SPECint95 20.4 SPECfp95
Intel SE440BX2 Motherboard (550Mhz PIII) 22.3 SPECint95 15.1 SPECfp95
Compaq XP1000 (667Mhz Alpha) 37.5 SPECint95 65.5 SPECfp95

Check www.spec.org. It provides standard benchmarking code, and collects benchmarks for everything from PCs up to Big Iron. It will certainly have comparisons between PIIIs and the PPC-750, and should have Athlon data as soon as AMD gets around to compiling the benchmark software.

However, the "new and improved (tm)" SPECweb99 is almost ready: See the SPECwww99 home page

There's some people complaining that they can't see the actual #'s, so here they are

• Where did you get the figures
• Why aren't the figures on the official SPEC site like they should be with all relevant data, like compiler version, hardware availability dates etc. etc.
• Everyone else in the industry is obliged to write 'estimated' disclaimers all over unsubmitted results. Why isn't AMD? Are they afraid to admit they used the Intel compilers (what else could they have used?).

Keep in mind what these are ...just tiny benchmarks... not REAL world applications..

http://www.spec.org/osg/cpu95/CINT95/

http://www.spec.org/osg/cpu95/CFP95/

You can also argue about whether your normal applications are compiled with all the nifty compiler switches the benchmarked programs are. Again this doesn't negate the fact they are real programs.

SPEC does a reasonable just in trying to make an "apples to apples" comparison under adverse conditions. Running your own specific applications as a benchmark is best. But as a reasonable substitute SPEC isn't nowhere near in the same class as FLOPS, MIPS, or "Seive".

The marketing folks bend the rules as to what are Specmark numbers ( running on a simulator doesn't officially count). There suppose to be real machines.

Keep in mind what these are ...just tiny benchmarks... not REAL world applications..

http://www.spec.org/osg/cpu95/CINT95/

http://www.spec.org/osg/cpu95/CFP95/

You can also argue about whether your normal applications are compiled with all the nifty compiler switches the benchmarked programs are. Again this doesn't negate the fact they are real programs.

SPEC does a reasonable just in trying to make an "apples to apples" comparison under adverse conditions. Running your own specific applications as a benchmark is best. But as a reasonable substitute SPEC isn't nowhere near in the same class as FLOPS, MIPS, or "Seive".

The marketing folks bend the rules as to what are Specmark numbers ( running on a simulator doesn't officially count). There suppose to be real machines.

Anyone know how much the 8500 costs? It has 1.5Mb of level 1 cache - it has 150M transistors, to the Alpha 21264's 'mere' 15M. It must cost loads... This cache probably skews the SPEC results quite considerably when comparing to 'real world' cases - the SPEC marks scale pretty well with cache size I've heard... I also see that the 8500 doesn't seem to scale at all well at SPECfp as you add extra processors, compared to the other chips.

Real world usage can vary immensly from the SPEC values, depending on what you're doing. I have friends who've compared various machines for high end computations (fluid dynamics) and they found the SGIs ran/scaled the best, even though they didn't have the best SPECfp results for a single chip - it's their massive data buses that do the trick. Actually, the PA-RISC 8500 doesn't have a complete Fortran compiler yet... Most people I know consider the Alpha to be let down by it's IO/bus data-rate,etc. Yes, it's better than PC, but it's not much compared to the other high-end RISC guys, especially SGI, though I expect this difference to change...

A final note, a problem you get with high-speed processors is that they become nice little microwave transmittors (at 700-800+ Mhz, I think it was) and so you really need to reduce the power (the PA-RISC 8500 consumes 85W @ 440Mhz) and up the shielding when clocking at this rate, because otherwise you'd get a REAL pizza cooker/toaster in your computer...

Unfortunately the SPEC marks are never compiled with gcc because it isn't as fast as Intel's compilers, which I presume AMD will use :-). I hope this list will be updated when the K7 is out, since it is probably a good indicator of Linux integer performance.

Of course, you also need realistic benchmarks.

To get its hand on a good x86 design

Last year AMD had sale of something like 2.5 billion dollars. Last _quarter_, Motorola had sales of 7.2 billion. And Intel? 7.1 billion last quarter.

Precisely. AMD is too small to be taking on Intel. Motorola could take the K7 design and run with it.

The rumour mill has Compaq buying AMD. I think they need to be bought by someone who knows how to run a fab and who won't be in competition with AMD's customers.

x86 is a dying technology

Merced is rumoured to be a flop, McKinley is rumoured to be late, Alpha is still not taking off. Sure x86 is dying, but it has been for years and that hasn't stopped Intel making a fortune on it. Remember, Windows 2000 is either going to be W98/DOS-based or late or both, so where does that leave non-x86 designs for the mainstream? Some years off.

The PowerPC 750 is a damn fine chip.

I'm sure it is, and Motorola should keep building it, but its not where the volume is, and that's not going to change. And while they are doing OK, they don't have any sort of performance lead over the `dying' x86 chips: According to The CPU Info Centre they get maximally 17.6 SPECint95s at 400MHz in Motorola's 0.22um process, while Intel is well into the twenties at 0.25um (I think).

Apple MrC with PowerMathLib:
NUMERIC SORT: Iterations/sec.: 107.164610 Index: 2.769469
STRING SORT: Iterations/sec.: 9.606252 Index: 4.222528
BITFIELD: Iterations/sec.: 3571869546.392807 Index: 612.688667
FP EMULATION: Iterations/sec.: 20.041185 Index: 9.635185
FOURIER: Iterations/sec.: 1602.031493 Index: 1.813977
ASSIGNMENT: Iterations/sec.: 1.510908 Index: 5.756497
IDEA: Iterations/sec.: 462.599684 Index: 7.077719
HUFFMAN: Iterations/sec.: 168.458324 Index: 4.681348
NEURAL NET: Iterations/sec.: 3.053514 Index: 5.166690
LU DECOMPOSITION: Iterations/sec.: 52.623917 Index: 3.106855

This from the bytemark of a PowerMac 604e/180Mhz, compared to a P90 (which would get index=1.0). They both came from a page at Apple, that I linked to slashdot and that seems to have been removed since (no longer available from Altavista). Do some calculations: take the BITFIELDS, iterations per second column ; compute the number of clock cycles taken by iteration considering the 180Mhz frequency of the processor ; conclude.

The "index" column shows that the PPC604e was 600.0 times faster than a P90 (code that you don't run is very fast indeed). This pretty much killed the benchmark. This is why bytemark is plain bogus.

A SIMILAR BYTEMARK BENCH WAS USED BY APPLE TO CLAIM THAT G3 ARE TWICE FASTER THAN PII. Actually they might have taken another compiler to make this claim, but the outrageously better results for the G3 are for the exact same reason, and Apple perfectly knew that its benchmarks were bogus.

Even PC Magazine, a company that almost no one would argue would be biased in favor of Apple vs. Wintel, admits that a G3/400 beats a Dell Dimension XPS T500 Pentium III on tests using an application *widely reported* to be heavily optimized for MMX and SSE! All that with a 20% slower clock speed!

If it is Photoshop benchmark, it is mostly a loop, so not very characteristic of typical applications. Also last time I saw the people added the time of the different filters, which is stupid because it resulted on the slower filters (i.e. unoptimised) being the very preponderant, independantly of how often they are used in real life. Other debatable benchmarks are http://macspeedzone.com/4.0/Wi nvsMacmathematica.html (we don't know which compiler is used, which optimizations, etc.).

Real benchmarks are SPEC CPU benchmarks (see www.spec.org). G3 are about 15% faster than PII at the same clock rate. And of course, a very important point is that all processors aren't available at the same clock speed at the same time. PIII/500 are commercially available that do 22 SPECint. I don't know if I can buy a G3/500 *now* ; and Alpha have had consistently a higher clock than other CPUs.

Another source of SPEC numbers is http://infopad.eecs.berkeley.ed u/CIC/summary/local/ ; if you want to measure the "advocacy effect", have also a look to http://macspeedzone.com/4.0/WinvsM acSPECint.html, which amusingly have lower SPECint numbers for PII than the best published and verified at spec.org (dig spec.org, maybe they took the very first results for PII, or the worse ; there are several different entries with PIIs).

Perhaps you've heard of Project Appleseed? No?
http://exodus.physics.ucla.edu/appleseed/appleseed .html
Imagine that--an independent study of G3 vs. x86 and other processors...hmm...a PII/300 is over 33% slower than a G3/266 (13% slower clock), and over 17% slower than a Rev. A iMac (29% slower clock).

The funny thing is that the cluster of 8 G3/266 would have the same performance of a single 21264/667Mhz for floating point (or 2 21264 for integer code). The 21264 may be more expensive than 8 G3, but it is sure much more easy to debug a mono-processor code :-).
You should have noted the fine prints:
This investigation was initially motivated by the impressive single node performance we achieved on our well-benchmarked suite of plasma particle-in-cell (PIC) simulation codes [4-5] on the Macintosh G3/266, as shown in Table I. This was due in part to the availability of an excellent optimizing Fortran compiler for the Macintosh produced by the Absoft Corporation [6]
and of course:
Computer Push Time Loop Time
Macintosh G3/300: 1750 nsec. 191.1 sec.

One iteration runs in 1.75 microsecond (i.e. 500 cycles), which means that it is really short code and loop intensive (about 100 millions loops), very likely to be optimiseable. This is a specific application, that couldn't be generalized. The annoying thing is that, contrary to SPEC which allows any submitter to choose its own compiler, we don't know how they decided to choose the others' compilers. They are also daily Mac users, they probably know very well how to optimize, and choose the compiler for their Mac.
But all in all, it is a good point for G3. Now if only Apple/Motorola released official SPEC numbers for G3, instead of plain bogus bytemarks...

You're absolutlely correct, but the scope of this report wasn't to test themselves.
This shows one weakness of linux in this field. For instance go to spec.org and do a search for operating systems.
Results:
linux: 0
windows: 257
out of 2314 records.

I get this from CINT95 Results Looks like the Pentium can at least keep up in (I'm guessing) integer arithmetic. This isn't addressing "memory bandwidth"; but doesn't this mean the instruction scheduling is as good?

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.

Wrong. For instance, Dell Precision Workstation did 17.6 SpecINT with a PII-450, and the Precision Workstation 610 does 19.0 SpecINT. SpecFP is 15.2.

The important point is that all this is commercially available:

• The Dell Precision Workstation 610 is available at about $5000 without monitor and stripped downat Dell Online Store
• The SPEC benchs were done with the commercially available Windows NT 4.0 and Visual C++ (for the libraries).
• The Intel C/C++ compiler used, designed to be integrated with Microsoft Visual C++ IDE (as a plugin-compiler), is can be ordered by these means for about $429.

And I used the Dell example only because there were uncontestable proofs of the performance of the PII/Xeon. This doesn't imply that dirt cheap PII systems correctly designed couldn't reach the 17 SPECint.

And for compilers, there are no proofs that they are all that bad ; the point is moot anyway since the Intel Compiler is cheap ($400). Any software company having performance intensive applications, and compiler performance problems, would be able to afford it.

a G3 450 doing 20 shows that a G3 is much faster than the pentium

Now I'm waiting for you to provide informations of how to get a G3 doing measured 20 SPECint (where to get compiler, OS and hardware).

Wrong. For instance, Dell Precision Workstation did 17.6 SpecINT with a PII-450, and the Precision Workstation 610 does 19.0 SpecINT. SpecFP is 15.2.

The important point is that all this is commercially available:

• The Dell Precision Workstation 610 is available at about $5000 without monitor and stripped downat Dell Online Store
• The SPEC benchs were done with the commercially available Windows NT 4.0 and Visual C++ (for the libraries).
• The Intel C/C++ compiler used, designed to be integrated with Microsoft Visual C++ IDE (as a plugin-compiler), is can be ordered by these means for about $429.

And I used the Dell example only because there were uncontestable proofs of the performance of the PII/Xeon. This doesn't imply that dirt cheap PII systems correctly designed couldn't reach the 17 SPECint.

And for compilers, there are no proofs that they are all that bad ; the point is moot anyway since the Intel Compiler is cheap ($400). Any software company having performance intensive applications, and compiler performance problems, would be able to afford it.

a G3 450 doing 20 shows that a G3 is much faster than the pentium

Now I'm waiting for you to provide informations of how to get a G3 doing measured 20 SPECint (where to get compiler, OS and hardware).

Wrong. For instance, Dell Precision Workstation did 17.6 SpecINT with a PII-450, and the Precision Workstation 610 does 19.0 SpecINT. SpecFP is 15.2.

The important point is that all this is commercially available:

• The Dell Precision Workstation 610 is available at about $5000 without monitor and stripped downat Dell Online Store
• The SPEC benchs were done with the commercially available Windows NT 4.0 and Visual C++ (for the libraries).
• The Intel C/C++ compiler used, designed to be integrated with Microsoft Visual C++ IDE (as a plugin-compiler), is can be ordered by these means for about $429.

And I used the Dell example only because there were uncontestable proofs of the performance of the PII/Xeon. This doesn't imply that dirt cheap PII systems correctly designed couldn't reach the 17 SPECint.

And for compilers, there are no proofs that they are all that bad ; the point is moot anyway since the Intel Compiler is cheap ($400). Any software company having performance intensive applications, and compiler performance problems, would be able to afford it.

a G3 450 doing 20 shows that a G3 is much faster than the pentium

Now I'm waiting for you to provide informations of how to get a G3 doing measured 20 SPECint (where to get compiler, OS and hardware).

• SpecInt results 09/1995 (before PPro)
• SpecInt results 12/1995 (PPro)
• SpecInt results 1Q 1996
• SpecInt results 2Q 1996

I'd really really like to see your benchmarks. (Would be Bogomips, methinks).

• SpecInt results 09/1995 (before PPro)
• SpecInt results 12/1995 (PPro)
• SpecInt results 1Q 1996
• SpecInt results 2Q 1996

I'd really really like to see your benchmarks. (Would be Bogomips, methinks).

• SpecInt results 09/1995 (before PPro)
• SpecInt results 12/1995 (PPro)
• SpecInt results 1Q 1996
• SpecInt results 2Q 1996

I'd really really like to see your benchmarks. (Would be Bogomips, methinks).

• SpecInt results 09/1995 (before PPro)
• SpecInt results 12/1995 (PPro)
• SpecInt results 1Q 1996
• SpecInt results 2Q 1996

I'd really really like to see your benchmarks. (Would be Bogomips, methinks).

www.spec.org