Skyscraper wrote:

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...

It can handle low complexity DIVX / XVID ok. Think low bitrate 512x384 without fancy features like quarter pixel or GMC.

And most (if not all) K6+ notebooks would have a graphics chip with MPEG2 offload for DVD. That was fairly commonplace when K6+ was being sold.

Skyscraper wrote:

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...

I am a linux guy and certainly used Ogle with DeCSS back in those days compiled in my own box and worked fine in the K6-2, but I also have somewhere PowerDVD cd with it's license which I remember the system requirements were: Windows98, PentiumII 350Mhz or better, had a Pioneer slot loading DVD drive and a Voodoo Banshee (may be that was the key I don't know) and the motherboard is a 5EHM, it is still one of my retrogaming rigs, the DVD drive is long gone, otherwise I would upload a video while watching one of my DVDs.
May be I will swap a DVD drive if I get a working IDE.

vmunix wrote:

and a Voodoo Banshee (may be that was the key I don't know) and the motherboard is a 5EHM, it is still one of my retrogaming rig

It would help if write allocation was setup by default. That's an important consideration for K6-2 onward. But with K6+ sometimes boards don't configure things properly, although I do have the ASUS P5A Beta BIOS. It may also be related to the ALI AGP driver and AGP quirks in general on there. See the fun times over in F2bnp's current P5A thread....

I probably should have tried my PCI Voodoo3. But after spending hours trying to get a Rage 128 working without BSODs, I was just sick of ASUS P5A. 🤣

swaaye wrote:

vmunix wrote:

and a Voodoo Banshee (may be that was the key I don't know) and the motherboard is a 5EHM, it is still one of my retrogaming rig

It would help if write allocation was setup by default. That's an important consideration for K6-2 onward. But with K6+ sometimes boards don't configure things properly, although I do have the ASUS P5A Beta BIOS. It may also be related to the ALI AGP driver and AGP quirks in general on there. See the fun times over in F2bnp's current P5A thread....

I probably should have tried my PCI Voodoo3. But after spending hours trying to get a Rage 128 working without BSODs, I was just sick of ASUS P5A. 🤣

Totally different experince here, the VIA ETEQ (Apollo VP3) in my Soyo and also an Epox MVP3-G were not only rock solid but also seemed built for these processors, never tried other graphics cards the Banshee just worked there, and I think it has some kind of mpeg-2 ability.

vmunix wrote:

Totally different experince here, the VIA ETEQ (Apollo VP3) in my Soyo and also an Epox MVP3-G were not only rock solid but also seemed built for these processors, never tried other graphics cards the Banshee just worked there, and I think it has some kind of mpeg-2 ability.

If you stick to PCI things go a lot smoother with Super 7. Voodoo AGP cards usually work fine though.

AFAIK no 3dfx hardware had MPEG2 offload. They only had color space conversion and bilinear scaling. Basic video features.

I'd like to find more details on the CPU<->Northbridge<->DRAM differences of say the 440BX vs. VIA MVP3 or Alladin V. On paper, all have 64bit wide datapaths and 100Mhz bus speed. So if there are memory bottlenecks between the typical K6-3 vs. a P-II/P-III it must be in the timings and efficiencies of the northbridge to DRAM interface... or maybe performances differences between Socket7@100MHz vs. GTL+@100MHz.

As a crude comparison, I ran 3DMark99-Max's "Synthetic CPU 3D Speed" test (and only that test) on the following two systems:

I'm hoping that running just the Synthetic CPU benchmark will avoid the video card discrepencies. I only really care about the CPU/Mainboard performance. 3DMax99 detected both CPU's and did use 3DNow optimizations for the AMD system. All other settings were left default.

System#1:
MB: Asus P2B-F (440BX)
CPU: P-II 450MHz (Deschutes core)
RAM: 128MB PC-100
Video: Riva TNT2 Ultra + 3dfx V1

System#2:
MB: FIC PA-2013 (VIA Apollo MVP3)
CPU: K6-III 450MHz (Sharptooth core)
RAM: 192MB PC-100
Video: Radeon 7200 DDR

Granted, this isn't a very controlled test of comparison whatsoever. I didn't attempt to equalize any bios settings or anything like that. Just a quick ballpark based on two systems I have that I could spin up without any effort as they sat. Your mileage may vary.

The Intel system scored 4233 "CPU 3DMarks"
The AMD system scored 6889 "CPU 3DMarks"

Is SPEC benchmark software freeware? I'd gladly try that next.

Like I said earlier, 3DMark isn't really great for comparing these two CPUs. It is using 3DNow! to really help performance and this doesn't usually happen.

How can one test minimum or maximum frames under Windows 98 benchmarking?

Or should testing be done under XP? But then software like FRAPS might be too demanding...

Or just pick the few benchmarks that do analyse min, max and average...

F2bnp wrote:

Like I said earlier, 3DMark isn't really great for comparing these two CPUs. It is using 3DNow! to really help performance and this doesn't usually happen.

Ok. That doesn't seem rational to me. Shall we disable SSE on P-III's when we benchmark those, too? What the criteria for inclusion? Market share?! I think it's entirely fair to benchmark a CPU as it stands with whatever optimizations are available... as long as they are disclosed in the benchmark and let the results speak for themselves for all to interpret. There are plenty of reviews and benchmarks on record showing the K6-III's for what it was.

For however much romanticism there exists for old AMD processors, it seems there is also a contingent of irrational detractors.

For the record, in 1999, I had a dual P-II, 300MHz and never owned a K6 family proc until recently. So I don't really have a bone to pick here. As an EE, I'm just more interested in the history of how these things played out and the technology behind it more than anything.

QBiN wrote:

Ok. That doesn't seem rational to me. Shall we disable SSE on P-III's when we benchmark those, too? What the criteria for inclusion? Market share?! I think it's entirely fair to benchmark a CPU as it stands with whatever optimizations are available... as long as they are disclosed in the benchmark and let the results speak for themselves for all to interpret. There are plenty of reviews and benchmarks on record showing the K6-III's for what it was.

I believe benchmaks are as valid as they are useful. Not a lot of games were well optimized for 3DNow!, so whatever score it reaches in an optimized benchmark won't represent real world performance and thus not only won't be that useful, but can be misleading even.
I do not recall if SSE was widely used in games that you would want to run on a P3, but if not many games did, then yes, the benchmarks might be more useful if SSE is disabled.

IMHO if you want benchmarks to compare a few processors when intending to game on them, you should use games as benchmarks. A wide range of them is ideal, as they mgiht have varying degrees of optimization for different CPUs.

QBiN wrote:

Ok. That doesn't seem rational to me. Shall we disable SSE on P-III's when we benchmark those, too? What the criteria for inclus […]
Show full quote

F2bnp wrote:

Like I said earlier, 3DMark isn't really great for comparing these two CPUs. It is using 3DNow! to really help performance and this doesn't usually happen.

Ok. That doesn't seem rational to me. Shall we disable SSE on P-III's when we benchmark those, too? What the criteria for inclusion? Market share?! I think it's entirely fair to benchmark a CPU as it stands with whatever optimizations are available... as long as they are disclosed in the benchmark and let the results speak for themselves for all to interpret. There are plenty of reviews and benchmarks on record showing the K6-III's for what it was.

For however much romanticism there exists for old AMD processors, it seems there is also a contingent of irrational detractors.

For the record, in 1999, I had a dual P-II, 300MHz and never owned a K6 family proc until recently. So I don't really have a bone to pick here. As an EE, I'm just more interested in the history of how these things played out and the technology behind it more than anything.

But this is exactly the issue. SSE was widely used, 3DNow! wasn't. 3DMark isn't representing real-world performance, because other than Quake 2, no other game ever used 3DNow! to the extent that you see in 3DMark. This is why it is a completely irrelevant test for comparing these CPUs. It is a great tool to compare performance when you're building a system and want to see if you're not far off similar systems (a.k.a. based on the same architecture).

Most points have been made (FPU performance, memory bandwidth).
The shorter K6 pipeline (maybe even some more parallelism, it has been a while since I last looked into the optimization manual) enables it to execute certain kind of applications faster. Basically programs where the path of execution can't be anticipated as well as in streaming or rendering applications (games, video, audio, ...) run better with a shorter pipeline. Compiling has been mentioned as an example, IIRC chess is also an example. (Found a source for this: K6-2 is faster than P2 at the same clock, K6-3 is even faster.)
Office software (basically legacy stuff with lots of logic, generally not very optimized) is also an example.
See Red Hill: "... But in practice, they overclocked like crazy. We ran a 450MHz K6-III/450+ at 560MHz for a year or so in our main accounting machine, and this was something you could do as mere routine. These were the most overclockable chip since the Celeron 300A. With the 1MB cache and PC-133 RAM clocked up to 112MHz (instead of the design 100) and the multiplier at 5 instead of 4.5 it ran like an express train, and remained perfectly cool to touch. We should really have tried a still higher clocking. No matter. For the purely business tasks that we used it for, it was superior to several of the CPUs we replaced it with (including a Duron 700 and a Thunderbird 900) and roughly the equal of the Thunderbird 1000 that replaced them. ..."

I love my k6-3+, and i really liked the redhill pages about old cpu's (i've read them all).

But i don't really find their stories usefull when it comes to performance... They PRAISED the cyrix well into the p2 era and dismissed that same p2 as a real option. Same goes for the k6-3 and saying their windows performance only got better with a 1ghz cpu 🤣 Talking about VALUE they are right, as the intel's were way to expensive, but performance wise they had no match, atleast in games. But that's the importatnt thing i guess, as they mostly talk about integer and then the cyrix/amd cpu's had great value.

About 3dmark99: my k6-3+@550 or 600mhz absolutely slaughters my p2-450 (no sse) when it comes to cpu marks. But there is not a single game it can keep up and the intel is MUCH faster. So no, 3dmark99 is not a good test to compare these cpu's.

I really like to use Everest cpu benchmarks (old version of Aida64). It has a big database and you can compare to other cpu's. Here you can see that the k6-3 has some strongpoints in a few tests but others and in fpu tests it is a long way from the intel's.

QBiN wrote:

I'd like to find more details on the CPU<->Northbridge<->DRAM differences of say the 440BX vs. VIA MVP3 or Alladin V. On paper, all have 64bit wide datapaths and 100Mhz bus speed. So if there are memory bottlenecks between the typical K6-3 vs. a P-II/P-III it must be in the timings and efficiencies of the northbridge to DRAM interface... or maybe performances differences between Socket7@100MHz vs. GTL+@100MHz.

Yes, all (super)socket7 chipset memory controllers are just slow.

Few Aida64 memory benchmarks:
.................................................MB/s....MB/s....MB/s....ns
platform................................ RAM read / write / copy / latency
PIII 600EB + i820 + PC800 RDRAM.......998......1021...1002....160
Athlon 600 + KX133 + PC133 SDRAM....779.......1027...875....222
PIII 500 + i440BX + PC100 SDRAM........691......778....340....169
PII 450 + i440BX + PC100 SDRAM.........501......192....292....173
K6-III+ 577 + Aladdin V + PC105 SDR....361.......160...207.....210
K6-III 450 + Aladdin V + PC100 SDR......345......130...191.....221
K6-III 450 + MVP3 + PC100 SDR...........261......145...185....291
Pentium 166 + Aladdin V + PC66 SDR.....256......151...185....214

And since you've showed some 3dmark99 CPU benchmarks, why not add total score as well. 😀

3dmark99 MAX, 800x600 (default), Voodoo3 3000 AGP, win XP

PIII 600EB + i820 + PC800 RDRAM.......5016
Athlon 600 + KX133 + PC133 SDRAM....5308
PIII 500 + i440BX + PC100 SDRAM........3896
PII 450 + i440BX + PC100 SDRAM.........3166
K6-III+ 577 + Aladdin V + PC105 SDR....3143
K6-III 450 + Aladdin V + PC100 SDR......2651
K6-III 450 + MVP3 + PC100 SDR...........2809
Pentium 166 + Aladdin V + PC66 SDR.....818

meljor wrote:

I love my k6-3+, and i really liked the redhill pages about old cpu's (i've read them all). […]
Show full quote

I love my k6-3+, and i really liked the redhill pages about old cpu's (i've read them all).

But i don't really find their stories usefull when it comes to performance... They PRAISED the cyrix well into the p2 era and dismissed that same p2 as a real option. Same goes for the k6-3 and saying their windows performance only got better with a 1ghz cpu 🤣 Talking about VALUE they are right, as the intel's were way to expensive, but performance wise they had no match, atleast in games. But that's the importatnt thing i guess, as they mostly talk about integer and then the cyrix/amd cpu's had great value.

About 3dmark99: my k6-3+@550 or 600mhz absolutely slaughters my p2-450 (no sse) when it comes to cpu marks. But there is not a single game it can keep up and the intel is MUCH faster. So no, 3dmark99 is not a good test to compare these cpu's.

I really like to use Everest cpu benchmarks (old version of Aida64). It has a big database and you can compare to other cpu's. Here you can see that the k6-3 has some strongpoints in a few tests but others and in fpu tests it is a long way from the intel's.

I think thats the point tho, interger was king outside of 3D games, remember many, many users just used email, netscape, word 97 and played solitaire, occassionally scanned a photo and played a few Champ arcade games or similar. most of these programs benefited from a fast CPU, but the NPU speed was pretty irrelevant, an awful lot of people never played Quake or similar, this was the playstation era, Tekken, Gran Turismo were king, by 1998 Dreamcast was around too, these were cheap and easy for gaming and the Dreamcast was VERY hard to beat. Come the end of 1998 early 1999 a Cyrix MII 300 could be had for £30/£40, the K6 was perhaps £10/£20 more, The Intel P6 chips on the other hand were a lot more, £150+++ and motherboard costs were way higher too. The K6-2 was made to compete with Celeron anyway not P2 or P3. In using a common standard platform(socket 7) along with Cyrix it made a lot more sense for the average user than Pentium 2 or even Celeron.
Of course today it makes no odds, it fact a Pentium 2/3 CPU is way cheaper than a Faster K6-2/3 on ebay, even slot 1 boards are cheaper today than Super 7's.
But at the time I think both the AMD K6 series and Cyrix 6x86 series & IDT Winchip made a lot of sense. They were more than fast enough for what most people did with a PC.

I have the feeling that the Duron spitfire was perhaps the best CPU ever as far a price/performance is concerned, pretty much ticked all the boxes for gamers and regular users combined.

Havli, I only ran the 3Dmark CPU test because I had very dissimilar systems, and I was trying to isolate the score differences to just those factors relating to CPU and chipset. Though, you and and F2 need to get together and decide when 3Dmark is a valid comparison, apparently. Perhaps only when it agrees with your premise? You can't have it both ways.

As BSA mentions, look, there is no debate that the Intel platforms of the era were better gaming systems than the K6-III (and derivatives). Likewise the same reviews of the time note the impressive performance for the K6-III in average desktop and business software. Nothing more. Nothing less.

However, I would remind all those that can't decide whether 3Dnow! is fair/irrelevant/whatever... in 1998 and much of 1999 there was no SSE enabled processors on the market. In a very real way, 3Dnow was the only steaming SIMD instruction set extension out. It did have real market share even if it did not have a significant impact. At least enough of a market impact that DirectX included optimizations as well as nVidia's drivers of the day (my subtle counter-example to the notion that nothing used 3DNow or that it had no impact). So any snapshot of the late 90's really does deserve to include 3DNow optimizations even if nothing more than to show what it could be capable of -- even if it ended up being just a sidenote in computing history.

BSA also touches on a huge factor for Socket 7 owners back then... for those that did have Super Socket 7 boards, the K6-III represented a great upgrade path at a great price, especially for non-gamers. Back to the premise of the thread, everything after it including the Athlons and Durons handily outmatch it. Dollar for dollar, the Durons were amazing performers.

K6-III+ is more flexible, software multiplier and better for DOS gaming

/close thread

Well, according to my tests, in all applications except CPUMark99 K6-III 450 is slower than PII 450 or even Celeron 433. And that is not just games but rendering, file compression, audio encoding, video encoding, and photo editing. Also all applications I used are much newer than tested CPUs... so SW developers had plenty of time to implement 3dnow support.

I don't know what real life prices used to be back then but official launch price is quite high for K6-III.
K6-2 450 = $203 @ February 26th 1999
K6-III 450 = $476 @ May 13th 1999
Celeron 433 = $169 @ March 22nd 1999
PII 450 = $669 @ August 24th 1998

BSA, QBiN and Phil all three hit the nail on the head.

Likewise, for people still stuck on (non Super-)Socket 7 the K6-2s presented a great upgrade option.

I got a K6-200 with TX chipset in late 1997. Of course, by 1998, it was ridiculously outdated. With the help of a K6-2/300 and a Voodoo (later Voodoo II) I managed to make it last till 2000, when I had the funds to upgrade to a Duron/GeForce MX system (which again was a night and day difference).