I was never a fan of the K6 line. I recently pulled a K6-2 300MHz (512K L2) system out of the trash and found that the performance was well below that of my PII-300.

One of the first things I noticed about the K6-2 system was that the Windows GUI got a little choppy when I tried playing MP3 and AAC music in the background (streamed over LAN from my file server). The PII-300 had absolutely no problem multitasking with music playing. The PII could also handle DVD playback just fine with GPU assistance; the K6 had trouble even with a Radeon VE.

Years ago I had a 500MHz K6-2 w/ 2MB L2 that I tried to max out. No matter what I did, I couldn't get the performance anywhere near my Katmai-550.

Yeah, the K6-3+ is faster thanks to the on chip cache, but I highly doubt it's enough to get it close to Athlon or PIII performance. Plus, Super 7 boards in general are just flaky as hell.

The Athlon 500Mhz beats the PIII 500Mhz in FPU Math.
The PIII only is faster if the application supports SSE. In 1999/2000, SSE was not widley used.

A 550Mhz K6-3+ can be slightly faster in Integer Math in some Situations.
In Floating Point, no Chance at all.

The Athlon got faster Bus (200Mhz effectiv (DDR)), faster Memory Speed..
So it depends heavly on the Platform used for the K6-3+.

The PII also benefits from better Bandwiths. Where the K6 needs really a decent Mainboard to unleash some power. On the PII Platform you mostly have LX440/BX440...

Only if you are plagued by early Apollo Pro Chipsets you will have a slow experience..

QBiN wrote:

F2bnp wrote:

No, no way in hell. A K6-III+ 550 is roughly as fast as a Pentium II 350 (sometimes faster, sometimes slower). A Pentium III Katmai 550 will annihilate the poor K6, so imagine what an Athlon/Duron can do which are even faster than the Katmais.

According to a compilation review by Tom's Hardware, the standard K6-III/500 "walked all over the Pentium II/450". With the die shrink allowing the K6-III+ to reach 550 and 600+ with overclocking, I would think the K6-III+ proves a decent counter to the Katmai core P-III's @ 500/600MHz... that is until you get to floating point performance. It'll lag a bit there.

I honestly don't see how it "walks all over" the Pentium II 450. First off, such a CPU is absent from the test so we've only got their word to lean on. If you look at most tests, the K6-III 450 is behind the Celeron 400 or equaling it. The Pentium II 450 is slightly faster than the Celeron 400, since it's using a 100MHz FSB instead of 66MHz and also enjoys a 50MHz higher frequency.

But then again, most of these tests are very unfair to both Pentium II/Celeron Mendocino's and K6 CPUs, since a lot of them make heavy use of SSE (that's why you see later Celerons really pull ahead of the Mendocinos) and others were released much later than these processors' release. I find 3DMark01 a terrible benchmark to compare different CPU architectures and would only use it to compare to CPUs of the same architecture. For example, in 3DMark 2000, a K6-III+ can score really well thanks to 3DNow! optimizations. It would be really misleading to compare it to a Pentium II. So, I can't really call this a definitive test/benchmark suite.

swaaye wrote:

What would be really interesting is data for minimum frame rates. 😁

This is very important too. Average framerates are great, but they don't show the whole picture. K6 systems are indeed Pentium MMX systems on steroids as Imperious said. That's why I keep mine and use it regularly. I selected a Baby AT board and the whole system is in a small Baby AT case, very reminiscent of my first computer which used a Pentium 133. I loved that machine, it served as a family computer for many years, until both my father and brother got their own systems, at which point they gave the Pentium 133 all to myself. To this day I remember the feeling of receiving my very very own computer. There's certainly a lot of nostalgia involved (I remember my brother teaching me how to use DOS with some simple commands, navigating folders etc).
My K6-III+ reminds me of my very first computer. It is really slow if you compare it to the Pentium II and Pentium III, but it is fast enough to make me happy. It is a shitty system, but hell, it is my very own piece of crap that I've put together and tweak regularly 🤣 .

For unoptimized code the K6 shines due to it's short pipeline. There it can be on par with an early Duron at the same clock speed.
Everything else, like compiled with P2 or Athlon as target platform, FPU intensive a.s.o. it can't keep up.

elianda wrote:

For unoptimized code the K6 shines due to it's short pipeline. There it can be on par with an early Duron at the same clock speed.
Everything else, like compiled with P2 or Athlon as target platform, FPU intensive a.s.o. it can't keep up.

One issue is that all benchmarks are very optimized, people kind of get the wrong impression as alot of other software isnt very optimized. When it comes to games there isnt much saving grace for the K6 though unless the games use 3DNow!

Back in the 98 I remember we choosed for a 350Mhz K6-2, cause the Pentium II line actually was more expensive even at lower freqs in the shop. But this led me to make the error to buy a Voodoo3 that with that cpu was heavily underused equally with the later bought 550Mhz version of the same cpu. This cpu was a BIG disappointment in games/bench. Later going for a Duron 750 with a KT133 based mobo, the difference was like night and day.
But I still like what the K6-2 meant in the cpu history.

Skyscraper wrote:

elianda wrote:

For unoptimized code the K6 shines due to it's short pipeline. There it can be on par with an early Duron at the same clock speed.
Everything else, like compiled with P2 or Athlon as target platform, FPU intensive a.s.o. it can't keep up.

One issue is that all benchmarks are very optimized, people kind of get the wrong impression as alot of other software isnt very optimized. When it comes to games there isnt much saving grace for the K6 though unless the games use 3DNow!

Which is why I don't use benchmarks. How a piece of hardware performs in actual games is the only thing that matters. Benchmarks can be rigged or drivers can be tweaked to give false results or the benchmark itself could be heavily weighted towards functions that you almost never see utilized in actual games.

Back in those days, for me a way to compare two CPUs or even better, two system, was the 2.2 Linux kernel compilation time, it also tested buses and disks a little bit, for a brief time the K6-III was the king indeed until the Athlon, which in turn was destroyed by the EV6 Alpha @ 667Mhz.
The Intels were very good all round , but they were particularly good while executing binaries which took advantage of the deep floating point pipeline, synthetic benchmarks and execution of code FPU intensive reflects that. Most of-the-shelf software products were compiled for Intel, and who can blame the vendors if the CPU is dominant for home users, so fair enough AMD hired ex DIGITAL employees to design the next generation floating point unit which went to the Athlon processor.

Having said that, I'm impress how nobody mentioned SPEC , SPEC is bussines and pretty much serious stuff.

I think one has to be precise and disciplined in this comparison. Otherwise, you start comparing apples and oranges in no time. So, the control in thread title is the K6-III+/550. That's a Sharptooth, .18u core, full speed 256K L2 on-die cache, with MMX, 3DNow, and Enhanced 3DNow extensions.

Keep K6-2's out of it. That's the same as saying you want to talk about a Coppermine P3 and someone brings up a no-cache, Covington Celeron as the example.

Also, comparing a K6-III to a Pentium on steroids?! If the main difference was just clock speed, I might agree. However, the micro-architectures are nowhere near similar. A K6-III has a RISC core with an x86 decoder with out-of-order execution like the P6 microarchitecture. Whereas the Pentium is a traditional CISC in-order micro-architecture with dual integer pipelines.

On business/integer benchmarks, everything I see has K6-III's meeting if not besting all intel offerings clock for clock until the Coppermine core offerings. The tables turn when we see FPU benchmarks and 3D gaming benchmarks. K6-III stays competitive if 3DNow is used. However, P-III jumps ahead if SSE optimized is used. I think that's where one sees the Achilles heal of the K6-III architecture. It's FPU performance is behind that of the P-II/P-III architecture and definitely behind the Athlon/Duron processors.

QBiN wrote:

I think one has to be precise and disciplined in this comparison. Otherwise, you start comparing apples and oranges in no time. […]
Show full quote

I think one has to be precise and disciplined in this comparison. Otherwise, you start comparing apples and oranges in no time. So, the control in thread title is the K6-III+/550. That's a Sharptooth, .18u core, full speed 256K L2 on-die cache, with MMX, 3DNow, and Enhanced 3DNow extensions.

Keep K6-2's out of it. That's the same as saying you want to talk about a Coppermine P3 and someone brings up a no-cache, Covington Celeron as the example.

Also, comparing a K6-III to a Pentium on steroids?! If the main difference was just clock speed, I might agree. However, the micro-architectures are nowhere near similar. A K6-III has a RISC core with an x86 decoder with out-of-order execution like the P6 microarchitecture. Whereas the Pentium is a traditional CISC in-order micro-architecture with dual integer pipelines.

On business/integer benchmarks, everything I see has K6-III's meeting if not besting all intel offerings clock for clock until the Coppermine core offerings. The tables turn when we see FPU benchmarks and 3D gaming benchmarks. K6-III stays competitive if 3DNow is used. However, P-III jumps ahead if SSE optimized is used. I think that's where one sees the Achilles heal of the K6-III architecture. It's FPU performance is behind that of the P-II/P-III architecture and definitely behind the Athlon/Duron processors.

This. 😊

QBiN wrote:

I think one has to be precise and disciplined in this comparison. Otherwise, you start comparing apples and oranges in no time. […]
Show full quote

I think one has to be precise and disciplined in this comparison. Otherwise, you start comparing apples and oranges in no time. So, the control in thread title is the K6-III+/550. That's a Sharptooth, .18u core, full speed 256K L2 on-die cache, with MMX, 3DNow, and Enhanced 3DNow extensions.

Keep K6-2's out of it. That's the same as saying you want to talk about a Coppermine P3 and someone brings up a no-cache, Covington Celeron as the example.

Also, comparing a K6-III to a Pentium on steroids?! If the main difference was just clock speed, I might agree. However, the micro-architectures are nowhere near similar. A K6-III has a RISC core with an x86 decoder with out-of-order execution like the P6 microarchitecture. Whereas the Pentium is a traditional CISC in-order micro-architecture with dual integer pipelines.

On business/integer benchmarks, everything I see has K6-III's meeting if not besting all intel offerings clock for clock until the Coppermine core offerings. The tables turn when we see FPU benchmarks and 3D gaming benchmarks. K6-III stays competitive if 3DNow is used. However, P-III jumps ahead if SSE optimized is used. I think that's where one sees the Achilles heal of the K6-III architecture. It's FPU performance is behind that of the P-II/P-III architecture and definitely behind the Athlon/Duron processors.

What are the "technical" reasons for this different performance of the K6's FPU line? (considering the FPU was not "a new thing" in the late 90s) Could that mean the K6 line was more oriented for office applications?

Depending on your knowledge on how an operation is executed, the process it takes for something to be accomplished by the CPU. Perhaps you can google "pipelined FPU" There are significant differences between the K6 and Pentium FPUs, the most notable is the lack of pipeline in K6, but it's also interesting that for single operation the K6 FPU tends to be faster, which is also true for the 6x86, though the thing turns around when you have a "stream" of calculations to be made, think of a really fast guy (AMD) who is trying to put out a fire and he carries a bucket of water from the fountain to the fire really quick, otoh you have an intel boy who is not as fast, but there's a line of those guys from the fountain to the fire, so for a single bucket of water, the AMD chap can do it quicker, although if you need 15 of those (don't remember the pipeline depth but I think it's 15) the AMD guy needs to go back and forth 15 times, where's the intel team with much less cpu cycles can put out the fire.

QBiN wrote:

It's FPU performance is behind that of the P-II/P-III architecture and definitely behind the Athlon/Duron processors.

And its memory performance. Don't discount that one so easily. Memory performance also became a major bottleneck for K7 and P6 by late 1999 as they approached 1 GHz.

Interesting; I would imagine that single FPU operations where not that usual with 3D games or audio/video encoding/decoding.

386SX wrote:

Interesting; I would imagine that single FPU operations where not that usual with 3D games or audio/video encoding/decoding.

AV encoding and decoding use integer ops.

A notable thing about SSE is its instructions that improve MPEG2 decoding. I had a 300 MHz Katmai on 66 MHz FSB decoding a DVD in software with no assist. Good luck achieving that on a K6-III+ even with its 3DNow Extended.

swaaye wrote:

386SX wrote:

Interesting; I would imagine that single FPU operations where not that usual with 3D games or audio/video encoding/decoding.

AV encoding and decoding use integer ops.

A notable thing about SSE is its instructions that improve MPEG2 decoding. I had a 300 MHz Katmai on 66 MHz FSB decoding a DVD in software with no assist. Good luck achieving that on a K6-III+ even with its 3DNow Extended.

I had no problems to watch DVDs with a K6-2 500 in software mode.

vmunix wrote:

I had no problems to watch DVDs with a K6-2 500 in software mode.

Try it recently?

swaaye wrote:

vmunix wrote:

I had no problems to watch DVDs with a K6-2 500 in software mode.

Try it recently?

What's your point ? what difference will it make the same situation in 2000 and in 2015 ? are you serious ? also are you suggesting I'm making this up? that is not possible to decode DVD with a K6-2 or III ? why are you derailing the thread ?

vmunix wrote:

What's your point ? what difference will it make the same situation in 2000 and in 2015 ? are you serious ? also are you suggesting I'm making this up? that is not possible to decode DVD with a K6-2 or III ? why are you derailing the thread ?

The chatter is about how K6-3+ can supposedly match a Duron. DVDs included. I spent some time with a Aladdin V ASUS P5A and a K6-3+ 600 a few months ago trying to play DVDs and it wasn't very pleasant. It took some AGP GART tweaking to get it to happen without hardware acceleration and that only worked with a Voodoo3 because other AGP cards became unstable.

So my point is K6 has some problems with DVD playback. I'm challenging nostalgia memory used as evidence.

I had a programmer friend who now has moved to the country of burgers and plastic surgery to work for Google. He used a laptop with a K6-2+ CPU to watch all kind of movies including DVD, DivX and Xvid as late as 2007. When asked how he managed to get that slow system to do this he would just give you a stare like you asked something really stupid 😀. He was a Linux person...