VOGONS


First post, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie

I haven't done any real benchmarking in quite some time. Now that my PII is up and running, I thought I'd share results from some of the Intel P6-based machines I've built over the years.

So here are some performance numbers spanning nearly the entire P6 family, from Klamath to Yonah.
Update: I've added some non-P6 processors (K6, K7, K8 and Core 2) to serve as additional reference points.

WinXP-SP3 used on all systems except the K6-2, which ran Win2k SP4.

P6 Processor Summary:
92VYHHn.png

Non-P6 processor summary:
5fH24WM.png

m0nIN2N.png

SFdv3K2.png

umiqV3S.png

GEAjxhG.png

xiMMNac.png

au8zIpk.png

Last edited by Standard Def Steve on 2017-01-20, 23:29. Edited 10 times in total.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 1 of 22, by Tetrium

User metadata
Rank l33t++
Rank
l33t++

Interesting (and maybe a bit disappointing) to see Tualeron 1400 being more similar in performance to a Coppermine 1000 instead of the Tualatin 1400.

I did notice similar (poor) performance when comparing my Celeron 800 (coppermine) to both Coppermine 1000 and Pentium 3 550 (probably Katmai, but may have been an early Coppermine).

Drats, I want to like those Celerons 🤣! PPGA Celeron is still kinda neat though, I think. And these earliest s370 Celerons aren't getting a lot of love, even though they are actually not that bad at all 😜

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

Reply 2 of 22, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

Well..... the p3-s 1400 @ 100fsb will run at 1050mhz.

1050mhz, 100fsb, 512kb cache will be slower than 1400mhz, 100fsb, 256kb cache. So the Celeron 1400 is the fastest processor for a non overclock 440BX system.
Enough reason to like the Celeron Tualatin! 😁

We can be picky now so the p3 will be favored but back in the day i had a 333mhz celeron and later a 466mhz. They were a lot of bang for your buck.
The bad pcchips board i used with them made it a bad experience but the Celeron was fast for it's price.

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 3 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

It would really help the reader to place those results on a chart, although it would be nice to use the same GPU in each. From my experience, people like charts, and not so much tables.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 4 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie
Tetrium wrote:

Interesting (and maybe a bit disappointing) to see Tualeron 1400 being more similar in performance to a Coppermine 1000 instead of the Tualatin 1400.

I did notice similar (poor) performance when comparing my Celeron 800 (coppermine) to both Coppermine 1000 and Pentium 3 550 (probably Katmai, but may have been an early Coppermine).

Drats, I want to like those Celerons 🤣! PPGA Celeron is still kinda neat though, I think. And these earliest s370 Celerons aren't getting a lot of love, even though they are actually not that bad at all 😜

Yeah, FSB seems to matter A LOT to these processors, especially when they're clocked higher than ~800MHz. Even Deschutes @ 533MHz with a 133FSB managed to beat Katmai @ 550MHz in the non-SSE optimized tests, which I thought was interesting. I always assumed that Katmai was a bit faster clock for clock than Deschutes.

meljor wrote:

Well..... the p3-s 1400 @ 100fsb will run at 1050mhz.

1050mhz, 100fsb, 512kb cache will be slower than 1400mhz, 100fsb, 256kb cache. So the Celeron 1400 is the fastest processor for a non overclock 440BX system.

Way back in the day I did test PIII-S at 1050MHz on 440BX. There are some applications where the Celeron-1400 outperforms it, for sure, but the underclocked PIII-S still manages to outdo the Celeron in many gaming/3D applications. Not only is there twice the cache, but some low level testing revealed that Tualatin-512's cache also operates at a lower latency than Tualatin-256. In gaming workloads, that large, low latency cache more than makes up for the 350MHz clock deficit.

feipoa wrote:

It would really help the reader to place those results on a chart, although it would be nice to use the same GPU in each. From my experience, people like charts, and not so much tables.

Noted. If I find my old PPGA Celeron results, I'll put all of the data in a chart. 😀

As for my GPU choices, I made sure that each processor was completely outclassed by the video card. That PII will not do any better in 3DMark even with a 9800 Pro or 6800GT.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 5 of 22, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
Standard Def Steve wrote:

Way back in the day I did test PIII-S at 1050MHz on 440BX. There are some applications where the Celeron-1400 outperforms it, for sure, but the underclocked PIII-S still manages to outdo the Celeron in many gaming/3D applications. Not only is there twice the cache, but some low level testing revealed that Tualatin-512's cache also operates at a lower latency than Tualatin-256. In gaming workloads, that large, low latency cache more than makes up for the 350MHz clock deficit.

I suppose the Tualatin (the non-S one) also uses faster latency cache compared to Tualeron? Iirc the Tualerons received a bit higher latency from Intel in order to cripple its performance.
Might be interesting if a Coppermine 1000 with 100MHz FSB was included in the charts to compare to Tualeron 1000.

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

Reply 6 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie
Tetrium wrote:
Standard Def Steve wrote:

Way back in the day I did test PIII-S at 1050MHz on 440BX. There are some applications where the Celeron-1400 outperforms it, for sure, but the underclocked PIII-S still manages to outdo the Celeron in many gaming/3D applications. Not only is there twice the cache, but some low level testing revealed that Tualatin-512's cache also operates at a lower latency than Tualatin-256. In gaming workloads, that large, low latency cache more than makes up for the 350MHz clock deficit.

I suppose the Tualatin (the non-S one) also uses faster latency cache compared to Tualeron? Iirc the Tualerons received a bit higher latency from Intel in order to cripple its performance.
Might be interesting if a Coppermine 1000 with 100MHz FSB was included in the charts to compare to Tualeron 1000.

It's not just Tualeron; even the regular 256K Tualatin PIII has a higher cache latency compared to the 512KB version.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 7 of 22, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
Standard Def Steve wrote:
Tetrium wrote:
Standard Def Steve wrote:

Way back in the day I did test PIII-S at 1050MHz on 440BX. There are some applications where the Celeron-1400 outperforms it, for sure, but the underclocked PIII-S still manages to outdo the Celeron in many gaming/3D applications. Not only is there twice the cache, but some low level testing revealed that Tualatin-512's cache also operates at a lower latency than Tualatin-256. In gaming workloads, that large, low latency cache more than makes up for the 350MHz clock deficit.

I suppose the Tualatin (the non-S one) also uses faster latency cache compared to Tualeron? Iirc the Tualerons received a bit higher latency from Intel in order to cripple its performance.
Might be interesting if a Coppermine 1000 with 100MHz FSB was included in the charts to compare to Tualeron 1000.

It's not just Tualeron; even the regular 256K Tualatin PIII has a higher cache latency compared to the 512KB version.

Thanks for clearing this up 😀

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

Reply 8 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie

Updates:
-Added beautiful graphs!
-Added PII-300 (Klamath)
-Added Celeron 300A @ 450MHz
-To keep the GPU consistant with all 440BX systems, I replaced the Katmai's Radeon 9800 Pro with a GF4 Ti4400. The 3DMark01 score only decreased by 35 points, and Doom3 actually increased by a whopping .1 fps; from 7.8 to 7.9 fps.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 9 of 22, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Did you have to increase the voltage to run the Tualatin at 1575 MHz?

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 10 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie

No, it was fully stable at 1.45v

The only overvolting I did was on the P2-400 and Yonah processors, but they were much higher overclocks; from 400->533 and 2166->3083, respectively.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 11 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie

-Added a Celeron-266 (Covington) to the main table.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 12 of 22, by Kamerat

User metadata
Rank Oldbie
Rank
Oldbie
Standard Def Steve wrote:

-Added a Celeron-266 (Covington) to the main table.

It would have been fun to see the Covington @ 4x100MHz. 😈 Maybe you should play around with the FSB and multiplier on the Klamath, see how much it gaines from 100MHz FSB.

DOS Sound Blaster compatibility: PCI sound cards vs. PCI chipsets
YouTube channel

Reply 13 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie

Updates:
-Overclocked Covington Celeron to 448MHz (FSB112)
-Retested PII-300 with 100MHz FSB
-Added Celeron-766 with FSB66
-Added some non-P6 processors (K6, K7, K8, and Core 2) to serve as additional reference points.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 14 of 22, by Kamerat

User metadata
Rank Oldbie
Rank
Oldbie

Thanks!
Great overclock on that Covington, but still a poor thing.

DOS Sound Blaster compatibility: PCI sound cards vs. PCI chipsets
YouTube channel

Reply 15 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie
Kamerat wrote:

Thanks!
Great overclock on that Covington, but still a poor thing.

Yeah, the performance was worse than I thought it would be. Even at 448MHz/FSB112 it got its ass handed to it by the K6-2/500 in both 3DMark01 and 7-Zip.

I'm starting to think that WinXP is responsible for Covington's poor performance. I remember reading somewhere that Windows NT is a lot more sensitive to cache performance than Win9x. I chose XP to keep things consistent; it's the OS that I used to test all of the other machines here. I'm considering retesting the Covington with Win98SE to see if performance picks up (normally, XP tends to bench a little higher than 98, but on a cacheless system, who knows...)

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 16 of 22, by Palladium

User metadata
Rank Newbie
Rank
Newbie

Look at how close the P3-S is to the 2083MHz Barton despite being 500MHz slower and hampered by a 150MHz FSB, and also how Yonah is destroying the K8 X2 in Doom3 while easily clocked to 3GHz before C2D even showed up.

AMD was only lucky in 2001-2005 when Intel screwed around with Netburst, otherwise it would just to be as bad as a decade of Intel domination since C2D except starting in 2001.

Reply 17 of 22, by SPBHM

User metadata
Rank Oldbie
Rank
Oldbie
Palladium wrote:

Look at how close the P3-S is to the 2083MHz Barton despite being 500MHz slower and hampered by a 150MHz FSB, and also how Yonah is destroying the K8 X2 in Doom3 while easily clocked to 3GHz before C2D even showed up.

AMD was only lucky in 2001-2005 when Intel screwed around with Netburst, otherwise it would just to be as bad as a decade of Intel domination since C2D except starting in 2001.

AMD showed competence with the K7 before Netburst existed, and K8 was great on it's own, I don't agree with your view.
Intel was doing what they could developing Netburst AND the other design at the same time, that's how they got to C2D and the domination in the end.

let's not forget that Yonah is from 2006 and Tualatin is also not to far from the Barton release, also his Barton representation is average, while his Tualatin PC is pretty special (with DDR and OC), if you run the Barton at the same level (DDR500, 2.5GHz) you will see something more interesting from there... if you run the Athlon XP Palomino from the same era as the 1400S with OC and fast ram it will also look interesting I suppose.

anyway thanks for the results, I also tried to run D3 on my PII and it failed, guess it needs SSE

Reply 18 of 22, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie
Palladium wrote:

Look at how close the P3-S is to the 2083MHz Barton despite being 500MHz slower and hampered by a 150MHz FSB, and also how Yonah is destroying the K8 X2 in Doom3 while easily clocked to 3GHz before C2D even showed up.

Yeah, I find that, with an overclocked bus, PIII-S paired with the Apollo Pro 266 platform is just as fast as Barton clock for clock. I think it's the scary fast cache that does it. Cache throughput tests show that Tualatin's L2 is much quicker than Barton's. It's a shame that this motherboard (QDI Advance 12T) limits me to FSB150. The CPU is running at its default voltage, indicating that there's still plenty of OC headroom.

As for Yonah, keep in mind that mine is highly overclocked. Stock frequency for the T2600 is only 2.16GHz. At that speed, I'm almost certain that it would have been slower than even a stock-clocked (2.6GHz) Opteron. Also, I had to really overvolt Yonah to stabilize it at 3083MHz. Opteron, on the other hand, had no problem hitting 3GHz at stock voltage.

Ten Gigahertz
5 Groovy GHz: Ryzen 9 5900X | GTX 1080 Ti | 32GB DDR4-3600 | 2TB NVMe, 8TB HDD | Win 10
5 Troll GHz: AMD FX-8350 | Radeon R9 Fury | 16GB DDR3-1866 | 500GB SSD, 2TB HDD | Win 8.1

Reply 19 of 22, by Palladium

User metadata
Rank Newbie
Rank
Newbie
Standard Def Steve wrote:
Palladium wrote:

Look at how close the P3-S is to the 2083MHz Barton despite being 500MHz slower and hampered by a 150MHz FSB, and also how Yonah is destroying the K8 X2 in Doom3 while easily clocked to 3GHz before C2D even showed up.

Yeah, I find that, with an overclocked bus, PIII-S paired with the Apollo Pro 266 platform is just as fast as Barton clock for clock. I think it's the scary fast cache that does it. Cache throughput tests show that Tualatin's L2 is much quicker than Barton's. It's a shame that this motherboard (QDI Advance 12T) limits me to FSB150. The CPU is running at its default voltage, indicating that there's still plenty of OC headroom.

As for Yonah, keep in mind that mine is highly overclocked. Stock frequency for the T2600 is only 2.16GHz. At that speed, I'm almost certain that it would have been slower than even a stock-clocked (2.6GHz) Opteron. Also, I had to really overvolt Yonah to stabilize it at 3083MHz. Opteron, on the other hand, had no problem hitting 3GHz at stock voltage.

Keep in mind for Yonah it's a 31W mobile-oriented CPU vs the Opty which is a 110W desktop CPU. Needless to say the stock voltage for the former would be tuned way lower.

I would love to see an alternate history where an 2 GHz P3-S with SSE2/400FSB/DDR existed, that would have been a beast in 2001.