anakin94 wrote on 2024-11-05, 00:41:

Hi, which CPU is faster? I ask because, the P3 has the half cache size compared to the P2. […]
Show full quote

Hi,
which CPU is faster?
I ask because, the P3 has the half cache size compared to the P2.

Btw. is there a benchmark list of CPUs from this era?

That smaller cache is running twice as fast, clockspeed-wise, though and, AFAICR, has a wider data path too.

Coppermine, even with half the L2 cache, but with extra fast L2 cache, was faster in practically all scenarios.

EDIT: The Pentium 3 500E with 256KB L2 cache is significantly faster than even the Pentium 3 500 and its 512KB L2 cache, at Quake 3, for example.

https://www.anandtech.com/show/399/9

Thank you for explaining.

Link to my old test of intel cpu at 450mhz
Intel cpus at 450mhz test

Interesting. Looks like the nvidia driver has an sse codepath.

A more general point: you can only directly compare clock speeds and cache sizes between CPUs of exactly the same architecture and series. In general more=bettter, but more cache and clock won't always compensate a less efficient design, particularly if the cache is run slower.

Case in point: a late socket 7 system with K6-2 450 or 500 and a full 2MB of L2 cache (at 100MHz) would be (slightly) faster than the same system with 512kB of L2 cache, but it would run quite a lot slower in most tasks than either of these P2/P3 CPUs and in fact be soundly beaten by a Mendocino Celeron with 128kB of L2 cache at full speed. In fact, run that Celeron at 4.5x100MHz and it would beat the P2-450 in most tasks, despite having only 1/4 of its L2 cache.

danijelm wrote on 2024-11-05, 07:57:

Link to my old test of intel cpu at 450mhz
Intel cpus at 450mhz test

That really shows, what "we all" knew back in 1999(?) that the Celeron 300A@450 actually kicked the PII's A$$ for most use. Though - probably for other uses, like database applications, etc., the bigger albeit slower cache might have had an edge on the Celeron ...

H3nrik V! wrote on 2024-11-05, 10:15:

danijelm wrote on 2024-11-05, 07:57:

Link to my old test of intel cpu at 450mhz
Intel cpus at 450mhz test

That really shows, what "we all" knew back in 1999(?) that the Celeron 300A@450 actually kicked the PII's A$$ for most use. Though - probably for other uses, like database applications, etc., the bigger albeit slower cache might have had an edge on the Celeron ...

Well, pretty muched kicked the PII's a$$ at least ... But as per bang for the buck - undoubtedly 😁

danijelm wrote on 2024-11-05, 07:57:

Link to my old test of intel cpu at 450mhz
Intel cpus at 450mhz test

Brilliant idea, by the way, using a 600EB to get a CuMine CPU down to a clock speed, that has a PII equivalent. Haven't thought of that before, but was rather speculating in finding an unlocked PII that would overclock to 5x100 for comparison of architecture 🤣

While the Pentium II at best was a great CPU I would expect the P-III at lowest clocks to destroy it not to mention on the FPU side. More or less like an Athlon or Duron would did with any K6s cores at least for test I remember on the MPEG2/DVD decoding apps.

386SX wrote on 2024-11-05, 12:35:

While the Pentium II at best was a great CPU I would expect the P-III at lowest clocks to destroy it not to mention on the FPU side. More or less like an Athlon or Duron would did with any K6s cores at least for test I remember on the MPEG2/DVD decoding apps.

I don't recall that the FPU was faster on the P!!! than the PII per se - but the SSE, where used, probably gave it an edge ...

At least These Days, SIMD is considered to be one of the FPU's responsibilities. And the first iteration of SSE was focused purely on floating point. That alone would make the PIII's peak FP performance much higher than the PII's.

But even inefficient old x87 is quite bandwidth intensive, so I'd expect the Coppermine's faster cache to give it a bit of an edge there as well.

SSE also helps with some media decoding. A Katmai P3 at 300 MHz on a 66 MHz FSB can smoothly play a DVD in software. I ran a P3 450 on 66 MHz FSB to test that.

Something I also wonder is if SSE made soft modems less demanding. I believe that DSP processing was another target. Soft modems were annoyingly demanding with a P2 400.

SSE surely was a media beast.... On paper. It lasted quite some time for developers to implement it widely. By that time any cpu would reach beyond the GHz.

For cache I can only confirm.
P2 450 and P3 450 performed the same, the only difference was SSE which basically was a non event because lack of support.
P3 500 and P3 500E made the difference because of the faster cache on the latter one. SSE still didn't get too much support in these months between.

And, speaking of this, the original P3 should have been called pentium 2 SSE and the coppermine should have been called the original pentium 3. But K6 went for 3 so Intel should have at least a no 3 as well... Marketing.

darry wrote on 2024-11-05, 01:49:

Coppermine, even with half the L2 cache, but with extra fast L2 cache, was faster in practically all scenarios.

Coppermine L2 cache is effectively 8x faster - full speed + 256bit internal bus. It also has improved SSE, compared to Katmai.

The Serpent Rider wrote on 2024-11-06, 00:05:

darry wrote on 2024-11-05, 01:49:

Coppermine, even with half the L2 cache, but with extra fast L2 cache, was faster in practically all scenarios.

Coppermine L2 cache is effectively 8x faster - full speed + 256bit internal bus. It also has improved SSE, compared to Katmai.

It's even faster than that (via wikipedia):

"In performance, Coppermine arguably marked a bigger step than Katmai by introducing an on-chip L2 cache, which Intel names Advanced Transfer Cache (ATC). The ATC operates at the core clock rate and has a capacity of 256 KB, twice that of the on-chip cache formerly on Mendocino Celerons. It is eight-way set-associative and is accessed via a Double Quad Word Wide 256-bit bus, four times as wide as Katmai's. Further, latency was dropped to a quarter compared to Katmai. Another marketing term by Intel was Advanced System Buffering, which encompassed improvements to better take advantage of a 133 MT/s system bus. These include 6 fill buffers (vs. 4 on Katmai), 8 bus queue entries (vs. 4 on Katmai) and 4 write-back buffers (vs. 1 on Katmai).[5] Under competitive pressure from the AMD Athlon, Intel reworked the internals, finally removing some well-known pipeline stalls.[citation needed] As a result, applications affected by the stalls ran faster on Coppermine by up to 30%.[citation needed] The Coppermine contained 29 million transistors and was fabricated in a 180 nm process. "

microprocessor report (which should be a very reputable source) had something different to say about coppermine's cache at launch:

In addition, the new Pentium III increases the width of the L2 bus to 256 bits (plus ECC), supplying an entire cache line in one cycle. This change does not reduce the load-use latency, as previous designs fetched the critical word first, but it allows the cache to service a new request every other cycle, delivering a peak bandwidth of 11.7 GBytes/s at 733 MHz.
Although technically this is a half-speed cache, it delivers twice the bandwidth of Xeon’s external full-speed cache and four times the bandwidth of Katmai’s external half-speed cache. Staying at half the CPU speed in Coppermine still supplies plenty of bandwidth without creating a critical timing path that might limit future speed increases.

waterbeesje wrote on 2024-11-05, 20:26:

And, speaking of this, the original P3 should have been called pentium 2 SSE and the coppermine should have been called the original pentium 3. But K6 went for 3 so Intel should have at least a no 3 as well... Marketing.

Oh I like that, never thought it like this but it makes total sense.