A comparison of Intel IPC over the last 24 years

[ninaholic37]

Interesting. Is this the same program as the one posted here (CPUMark99) ?
http://forums.anandtech.com/showthread.php?t=2057154

I wonder if the OS you're running each test on makes a difference? On my Pentium M 1.6ghz with XP SP3, I got 165, then after closing my browser and it's tabs, I got 195. Changing the power scheme didn't make a difference though (still 195), it was always hitting 100% CPU so I guess it was always using the full 1.6Ghz. If they are the same program, I guess that would mean I got a score of 8.2?

 

[JimmiG]

If I had to guess the jump from K6 to K6-2 was the result of the execution engine being increased from 7 to 10 units and the FSB speed increasing from 66MHz to 100MHz. These CPUs generally had 512Kb of L2 cache on the motherboard and since CPUmark99 doesn't fit in the L1, going to the L2 produces a big hit in performance. Speed up L2 access and CPUmark99 performance increases.

For the K6-3 the L3 was 256Kb but it moved to the chip and ran at full speed. Another boost in CPUmark99 performance, and integer performance in general if you look at old Winstone benches.

The K6-3 had great integer performance, actually better than the original Athlon and Pentium 3 (which had a slower L2 cache). However it had the same weak, non-pipelined FPU as the original K6 making it less than ideal for games at the time (remember this was before the "GPU").

 

[Hulk]

The K6-3 had great integer performance, actually better than the original Athlon and Pentium 3 (which had a slower L2 cache). However it had the same weak, non-pipelined FPU as the original K6 making it less than ideal for games at the time (remember this was before the "GPU").

I remember patiently waiting for Anand's 1998 review of the K6-3 and being kind of blown away by the integer performance and then underwhelmed by the weak floating point unit. But man, if you looked at those Winstone scores, which actually meant a lot back then, the K6-3 looked amazing.

 

[Techhog]

Wow, the Pentium Pro was a champ.

 

[Blitzvogel]

Interesting as we gave up expensive core development for moar cores, but it made sense of course.

I'd like to see a chart comparing TDPs of mainstream CPUs from the past 20 years or so.

 

[SlowSpyder]

Interesting as we gave up expensive core development for moar cores, but it made sense of course.

I'd like to see a chart comparing TDPs of mainstream CPUs from the past 20 years or so.

Finally a chart I can be at the top of with my 220 watt TDP. :awe:

 

[NTMBK]

I'd love to know where the Big Iron chips fall on there. Things like the Power8, and their equivalent from yesteryear.

 

[Blitzvogel]

Finally a chart I can be at the top of with my 220 watt TDP. :awe:

LOL You have that FX-9xxx system in your sig ready?

 

[SlowSpyder]

LOL You have that FX-9xxx system in your sig ready?

Define 'ready'. It's 90% done, I just haven't had the urge to work on it lately (I've been playing an old game, that's been using up my 'computer' time lately), but I have all of Saturday left wide open to FINALLY get her running.

 

[glugglug]

I'm kind of surprised to see so much variance between the K6, K6-2, and K6-3. I thought those were more or less all the same CPU core, just with differing amounts of L2, SIMD instructions, and maybe built on a different process. It'd be interesting to see where Phenom and FX fit into the mix, too.

Regarding Intel, Prescott scored somewhere slower than the original Pentium but faster than the 486. Pretty crazy!

K6 and K6-2 were totally bottlenecked by L2 cache bandwidth (otherwise you are right, besides 3dNow instructions, cache integration, and clockspeed, the K6 and K6-III are identical).

K6-2 shows a lower IPC because the clockspeed went up by a higher factor than the L2 cache bandwidth, with the L2 being on the motherboard and accessed through the same slow FSB as RAM. Even though by then the standard was 100MHz FSB instead of 66MHz. K6-III moves the L2 on-chip eliminating that FSB bottleneck for cache access. Which is also the primary advantage of Pentium Pro vs. original Pentium. Also, the K6 seems to access L1 at close to register speed, getting an absolutely ridiculous boost whenever your working set gets down below that 64KB. It will beat a Pentium II running at double the clock speed on such an app.

Also, Pentium is showing close to double the IPC in this bench of a 486 primarily because the FSB went from 32-bit to 64-bit (the same bandwidth bottleneck again). I bet if you were to find a benchmark that fit in the 486's 8KB L1, eliminating this bottleneck, you would see the 486 with a higher IPC than the Pentium 4.

 

[Exophase]

I bet if you were to find a benchmark that fit in the 486's 8KB L1, eliminating this bottleneck, you would see the 486 with a higher IPC than the Pentium 4.

Unless that benchmark had a particularly poor branch predictions rate on the P4 or badly hit some of its other more obscure glass jaws then I doubt it.

This particular benchmark probably does so badly precisely because it's dependent on L2 latency and Prescott has such poor (in clocks) L2 latency. Which is why it does so much worse than Northwood here, but even Northwood had pretty high (again, in clocks) L2 latency.

But the L1 latency is quite a lot better, especially prior to Prescott where it was only 2 cycles. So if the 486 gets a big benefit from running from L1 cache so would the Pentium 4. Then it'd also get the benefit of having fairly deep OoOE with a lot more execution width, not having conflicts between instruction and data accesses in cache, and having better timing for things like (predicted) taken branches.

 

[glugglug]

Unless that benchmark had a particularly poor branch predictions rate on the P4 or badly hit some of its other more obscure glass jaws then I doubt it.

This particular benchmark probably does so badly precisely because it's dependent on L2 latency and Prescott has such poor (in clocks) L2 latency. Which is why it does so much worse than Northwood here, but even Northwood had pretty high (again, in clocks) L2 latency.

But the L1 latency is quite a lot better, especially prior to Prescott where it was only 2 cycles. So if the 486 gets a big benefit from running from L1 cache so would the Pentium 4. Then it'd also get the benefit of having fairly deep OoOE with a lot more execution width, not having conflicts between instruction and data accesses in cache, and having better timing for things like (predicted) taken branches.

Seeing as Prescott L2 is on-die, Prescott L2 latency is absolutely phenomenal compared to K6, K6-2, or Pentium P54 or P55C, which are not only leaving the die, but going to the motherboard to access L2. And yet all of these still blow it away.

And the 486 doesn't even have L2. In that case you are talking about on-die L2 VS RAM.

 

[Blandge]

SPEC is a system benchmark and won't measure IPC well. Dhrystone or Coremark is what you're looking for.

 

[Exophase]

Seeing as Prescott L2 is on-die, Prescott L2 latency is absolutely phenomenal compared to K6, K6-2, or Pentium P54 or P55C, which are not only leaving the die, but going to the motherboard to access L2. And yet all of these still blow it away.

And the 486 doesn't even have L2. In that case you are talking about on-die L2 VS RAM.

I did outright say it was poor in clocks, not in absolute time. That'd be the relevant comparison since we're looking purely at IPC. Although in terms of absolute time, Prescott's L2 latency was quite bad compared to later Core 2 systems that even had more L2 cache.. Intel got bit really badly making a CPU that was supposed to clock higher than it ever did.

486 has L2 the same way Pentium and K6 had it, on the motherboard. Yes, it was optional, but not uncommon. Hulk didn't specify if the one he tested had it or not.

 

[Blue_Max]

Daaaang! I really underestimated the Pentium Pro back in the day! It was also one of the earliest ways to do DUAL processors!

I sure missed out...

 

[SlowSpyder]

Daaaang! I really underestimated the Pentium Pro back in the day! It was also one of the earliest ways to do DUAL processors!

I sure missed out...

When I got my first IT job working on a Helpdesk I upgraded my machine from a Pentium 66MHz to a Pentium Pro 200MHz/1MB... wow. One of my friends who also worked there, him and I would sneak friends in and play Doom II and Command & Conquer on the weekends sometimes. Playing LAN games in ~1997 was awesome... those were the days. :awe: