VOGONS


AMD K5-200

Topic actions

Reply 20 of 28, by SavantStrike

User metadata
Rank Member
Rank
Member
sliderider wrote:
sgt76 wrote:
Tetrium wrote:

Making CPU's is not about being the fastest or making the most efficient chip, it's about making money and designing a K5-FPU + nx686 CPU wasn't the money way.

Hit the nail on the head...

Oh, and what about all the sales they lost to Intel because their floating point performance sucked? Wouldn't there have been more money in grabbing those sales away from Intel by including a better FPU even if it took a little more time to incorporate it into the design?

Long term it might have been better, if they could have even made it work. There may have been engineering constraints we are unaware of.

But short term, The K5 was late compared to the Pentium, the K6 was late compared to the Pentium MMX, and if AMD didn't release something relevant fast, they were going to get steamrolled by the PII and Celeron. They didn't have another year or two to spend in development, they didn't even have 6 months.

Reply 21 of 28, by swaaye

User metadata
Rank l33t++
Rank
l33t++

Another aspect to those days was that everyone underestimated the value of the FPU. You see, before 3D games, the FPU was barely used outside of scientific and engineering apps. Then in ~95, the 3D game revolution started. And since CPUs spend years in design, there was no way to react to this change. Most of these CPU companies were designing CPUs for office "productivity" apps.

Intel was the only x86 manufacturer that spent any effort on making x87 somewhat fast. But even their FPU sucked compared to the various alternatives like MIPS, Alpha, Sparc and PowerPC. Even P6 was slow compared to those until we saw PII and PIII really crank the clock speed.

Reply 22 of 28, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
swaaye wrote:

Another aspect to those days was that everyone underestimated the value of the FPU. You see, before 3D games, the FPU was barely used outside of scientific and engineering apps. Then in ~95, the 3D game revolution started. And since CPUs spend years in design, there was no way to react to this change. Most of these CPU companies were designing CPUs for office "productivity" apps.

Intel was the only x86 manufacturer that spent any effort on making x87 somewhat fast. But even their FPU sucked compared to the various alternatives like MIPS, Alpha, Sparc and PowerPC. Even P6 was slow compared to those until we saw PII and PIII really crank the clock speed.

Good post swaaye 😉

I kinda forgot about the non-x86 architectures. Now that I think about it, I still have 2 Alpha AT boards with CPU's that I haven't been able to get to post.

I mean, if you first compare a DX4 with a Pentium Socket 4, the Pentium looks very impressive!
But compare a Pentium s4 and one of those Alpha's, the Alpha's look like a Pentium XL and are running at a whopping 166Mhz to boot!!
(Disclaimer:Those particular Alpha's were made in late 1994 though).

I posted a topic on VCF about those boards, will edit this post with a link to that particular thread in a moment...

Edit: Link here http://www.vintage-computer.com/vcforum/showt … y-info-about-it

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 23 of 28, by F2bnp

User metadata
Rank l33t
Rank
l33t

AFAIK you can't just get the FPU from one CPU and put it on another. Not without problems anyway. And problems equal time and money that AMD couldn't possibly afford. The K6 series didn't just give AMD some money to work on the Athlon, it also saved them from bankruptcy.

Reply 24 of 28, by ggsociety

User metadata
Rank Newbie
Rank
Newbie

First post here, and I would like to add my 2 cents!

There has been debate to the blunder that AMD made by not getting the K5 FPU into the K6. Plain and simple, it would have been practically impossible and useless! I recall that by that time, FPU and (integer) CPU were no longer discrete entities running at asynchronous clock rates; The chips had had highly integrated designs that meant the FPU and the rest of the CPU were designed ground up to work together to minimize any bottlenecks in order to get peak performance. As people have stated in thread before: The K5 and K6 are very different designs in that the K5 had a pipelined FPU while the K6 had a low latency non pipelined FPU. It would not have made any sense, and is practically impossible. Add to that a few other points. 1) The K5 FPU would likely not scale to the clock rates necessary 2) During the K6's time, the Pentium II came along along and its FPU performance crushed anything before it making the K5 FPU performance a moot point.

In any case, during the times of the K5 and K6 were released, things were sure interesting in terms of CPU development.

At the time, only Intel was really focusing on FPU performance. Everyone else focused on integer performance. And then Quake 1 came out, followed by the slew of FPU dependant games ( I remember playing quake 1/quakeworld on a 486 DX4 120). Intel's Pentium, and later Pentium 2 really shone during that time. Everyone was caught off guard, and most CPU manufactorers began to fade away after that. AMD had a good chip with the K6 though. It was able to scale to high clock speeds (later as K6-2, K6-3, K6-2+) and it was a monster in terms of integer performance (The K6-3 could best the P2 and early P3 clock for clock in integer performance). In order to hold the fort untill it's own FPU monster was released, the K7, AMD developed the 3DNow! instructions.

Now I really don't know if the K6 was designed with future 3Dnow! instructions in mind... But it really seems like it did. No other CPU showed such a dramatic gain in performance with properly implemented SIMD instructions. 3DNow! brought K6 (2 or 3) performance parity with the P2 and P3 in some cases, and at a much lower cost, and it was not multiplier locked to facilitate easy overclocking! The Quake series of games good to the K6-2/3 in terms of performance. Recall that 3D accelerators came into prominance, and it was through 3DNow! optimizations of the video drivers that made the K6 line of CPU competetive with the P2 and P3 (Recall the miniGL driver for 3dFX, or the first Detonator drivers for the Riva TNT).

In any case, that is what I remember. A lot of factors and new technologies came into prominance. These factors helped the K6 series without the pipelined FPU survive untill the K7 was released.

Reply 25 of 28, by swaaye

User metadata
Rank l33t++
Rank
l33t++

Quake 2 was pretty fast on K6 w/3DNow thanks to the AMD written 3DNow patch. But Quake 3 was another story because it wasn't particularly 3DNow optimized. By that time though the FPS people were interested in Athlon instead.

It's also probably best to stick with 3dfx cards with a K6, particularly Voodoo2 with the Quake2 3dnow patch. I don't know if NV and ATI got more serious about 3DNow but they behind on that for awhile. This also avoids AGP complications on those Super 7 boards.

Reply 26 of 28, by ggsociety

User metadata
Rank Newbie
Rank
Newbie

If I remember correctly, what mattered the most during that time was the video card's 3dnow! implementation which determines how well it can feed the video card with data. Remember that hardware transform and lighting wasn't around yet!

I remember buying a Voodoo3 2000 for the quake 3 beta (140$ Canadian I believe, damn those were the days for a lower clocked top of the line card). The beauty of the Voodoo3 was that it too could use the 3Dfx 3dnow! optimized MiniGL driver that was developed for Quake2 originally (So could Half Life/Counter strike).

This miniGL driver was a key piece in enabling the K6 to remain competitive (3Dfx was the most popular 3D accelerator manufactorer at the time, they released the first decent 3D card after all) and luckily,the major games of that day (Quakeworld, Quake2, Quake3, Half life, Counterstrike) all worked with it flawlessly. What was interesting was that when 3Dfx released a full OpenGL driver performance actually dropped, so a lot of people kept using the stripped down but optimized Quake2 miniGL driver for everything.

Nvidia continued with good 3dnow suppport after the first detonator drivers, which is nice because it worked out of the box.

ATI had the worst performance discrepancy between AMD and Intel. It's lack of 3Dnow! optimizations were swept under the rug when the K7 arrived, and K6-x was retired.

Reply 28 of 28, by Arioch

User metadata
Rank Newbie
Rank
Newbie

Bumping old thread.

1) about close integration between CPU and FPU. Cyrix told to use CRX core from separate x387 chip to MediaGX/Geode cores. So it was not fused into CPU core. NexGen also made a separate chip, that they distributed in multichip package (Pentium Pro like) - http://coprocessor.cpu-info.com/index2.php?ma … id=Manu&tabid=8). So while Nx686 had to implement the Nx587 custom bus, it probably had no other binding and the bus could be changed - if they only had time to do it.

2) Socket 7 compatibility was not AMD influence, but was decided and probably partially complete before the acquisition ( http://www.businessweek.com/1997/06/b351386.htm ). If Nx686 did not offered P.MMX pins compatibility, then AMD would probably not be so interested in NexGen

3) K5 core had (according to the link above) potential to speed up. Who knows what could be if NexGen top managers made their way into AMD year earlier to change priorities and to iron out bottlenecks.

I wonder what information exists about pre-NexGen native K6 attempt ? It seems there is just no informatiiioon at all, even rumours are only fragments.