VOGONS


First post, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie

A few weeks ago I received my very first Slot A boards, both Gigabyte GA-7IXE, which (I had heard) was supposed to be one of the best Slot A boards.
So naturally I benchmarked the hell out of it and also did some comparisons to one of my Slot 1 builds.

Let me just start by saying that, I'm not really sure how other Slot A boards behave, but this one is probably as close to perfect as it gets for this platform - both in terms of stability (comparable with an Intel 440BX chipset), and in terms of speed (as you are about to see). Throughout all of my extensive testing I didn't encounter any issues - perfect reliability, no freezes & no BSODS.
The only weird behavior that I saw was related to RAM: it's very picky about RAM modules, more picky than all the SS7 boards I own - I literally had to dig through all of my RAM collection and only found very few that work (it seems to love the Kingston 256 MB PC133 memory modules, which work great and it's what I eventually used for this test). The board refuses to boot with most of my other PC133 SDRAM modules, which typically work fine even on very picky SS7 boards.

What was the goal for all these benchmarks (besides just doing it for fun)? Well, I really wanted to see what the sweet spot would be for a Slot A platform, find a perfectly balanced video card - not too slow, not too fast (while also taking into account compatibility). I know that a lot of people would probably go with a 3dfx card without giving it a second thought, but I prefer to weigh all the pros and cons first.
Also, of course, I wanted to see how the Pluto compares to the Thunderbird, and how both compare to a similar Pentium 3 Coppermine CPU (and also to a faster Coppermine CPU).

THE BENCHMARKS
A few notes first:
- the same Kingston 256 MB PC133 memory module was used for both platforms
- the same sound card (Yamaha YMF724F-V) was used for both platforms
- for the nVidia cards I used driver version 7.76 (for the GeForce 3 Ti 200 I forced the installation)
- for the 3dfx Voodoo 3 3000 I used the latest reference driver: v1.07.00
- the nVidia driver actually forces AGP 1x when it sees the AMD chipset (probably because previous AMD chipsets had issues in 2X mode). I used PowerStrip to reactivate AGP 2x, and it worked fine, but as expected, it doesn't make much of a difference (2 - 5 FPS at most for the very high resolutions), so I decided to just leave it at AGP 1x.


1. GIGABYTE GA-7IXE WITH AMD ATHLON 750 MHz (PLUTO)


  • With the GeForce 2 MX card:
Expendable 640 x 480 x 16 ---> AVG: 75 FPS / HIGH: 113 FPS / LOW: 52 FPS
Expendable 1024 x 768 x 32 ---> AVG: 72 FPS / HIGH: 99 FPS / LOW: 48 FPS

GLQuake 640 x 480 x 16 ---> 234 FPS
GLQuake 1024 x 768 x 32 ---> 74 FPS

Quake 2 640 x 480 x 16 ---> 209 FPS
Quake 2 1024 x 768 x 32 ---> 101 FPS

Quake 3 640 x 480 x 16 ---> 81 FPS
Quake 3 1024 x 768 x 32 ---> 56 FPS

3DMark 2000 ---> 4522 points
3DMark 99Max --> 5792 points (12423 CPU 3DMarks)

Interesting, with the GeForce 2 MX, framerate in GLQuake at 1024x768x32 is lower than in Quake 2 at the same resolution and identical to Riva TNT2 Pro. It seems that this game runs into a memory bandwidth bottleneck - since both the GeForce 2 MX and the Riva TNT2 Pro have the same 2.6 GB/s memory bandwidth).

  • With the GeForce 2 GTS card:
Expendable 640 x 480 x 16 ---> AVG: 77 FPS / HIGH: 116 FPS / LOW: 54 FPS
Expendable 1024 x 768 x 32 ---> AVG: 75 FPS / HIGH: 102 FPS / LOW: 51 FPS

GLQuake 640 x 480 x 16 ---> 337 FPS
GLQuake 1024 x 768 x 32 ---> 165 FPS

Quake 2 640 x 480 x 16 ---> 210 FPS
Quake 2 1024 x 768 x 32 ---> 163 FPS

Quake 3 640 x 480 x 16 ---> 81 FPS
Quake 3 1024 x 768 x 32 ---> 78 FPS

3DMark 2000 ---> 5931 points
3DMark 99Max --> 6716 points (12495 CPU 3DMarks)
  • With the GeForce 3 Ti 200 card:
Expendable 640 x 480 x 16 ---> AVG: 75 FPS / HIGH: 115 FPS / LOW: 52 FPS
Expendable 1024 x 768 x 32 ---> AVG: 74 FPS / HIGH: 111 FPS / LOW: 52 FPS

GLQuake 640 x 480 x 16 ---> 341 FPS
GLQuake 1024 x 768 x 32 --> 223 FPS

Quake 2 640 x 480 x 16 ---> 209 FPS
Quake 2 1024 x 768 x 32 ---> 197 FPS

Quake 3 640 x 480 x 16 ---> 79 FPS
Quake 3 1024 x 768 x 32 ---> 79 FPS


3DMark 2000 ---> 6181 points
3DMark 99Max --> 6614 points (12457 CPU 3DMarks)
  • With the Riva TNT2 Pro video card:
Expendable 640 x 480 x 16 ---> AVG: 75 FPS / HIGH: 102 FPS / LOW: 52 FPS
Expendable 1024 x 768 x 32 ---> AVG: 45 FPS / HIGH: 68 FPS / LOW: 18 FPS

GLQuake 640 x 480 x 16 ---> 176 FPS
GLQuake 1024 x 768 x 32 --> 74 FPS

Quake 2 640 x 480 x 16 ---> 128 FPS
Quake 2 1024 x 768 x 32 ---> 58 FPS

Quake 3 640 x 480 x 16 ---> 64 FPS
Quake 3 1024 x 768 x 32 ---> 37 FPS

3DMark 2000 ---> 3225 points
3DMark 99Max --> 6884 points (12378 CPU 3DMarks)

Also interesting, the TNT2 Pro has the highest score in 3DMark99, higher than with the GeForce 2 GTS or the GeForce 3 Ti 200 (maybe a driver optimization for this specific card).

  • With the Voodoo 3 3000 card:
Expendable 640 x 480 x 16 ---> AVG: 76 FPS / HIGH: 101 FPS / LOW: 55 FPS
Expendable 1024 x 768 x 16 ---> AVG: 66 FPS / HIGH: 96 FPS / LOW: 40 FPS

GLQuake 640 x 480 x 16 ---> 199 FPS
GLQuake 1024 x 768 x 16 --> 92 FPS

Quake 2 640 x 480 x 16 ---> 177 FPS
Quake 2 1024 x 768 x 16 ---> 84 FPS

Quake 3 640 x 480 x 16 ---> 66 FPS
Quake 3 1024 x 768 x 16 ---> 52 FPS

3DMark 2000 ---> 2989 points
3DMark 99Max --> 5843 points (12281 CPU 3DMarks)

Obviusly, the 1024 x 768 results are not really comparable to the other cards, since the Voodoo 3 only supports 16 bit color.


2. GIGABYTE GA-7IXE WITH AMD ATHLON 700 MHz (THUNDERBIRD)


  • Default clock speed with the GeForce 2 GTS:
Expendable 640 x 480 x 16 ---> AVG: 68 FPS / HIGH: 108 FPS / LOW: 47 FPS

GLQuake 640 x 480 x 16 ---> 292 FPS
Quake 2 640 x 480 x 16 ---> 184 FPS
Quake 3 640 x 480 x 16 ---> 73 FPS

3DMark 2000 ---> 5694 points
3DMark 99Max ---> 6257 points (11548 CPU 3DMarks)

  • Overclocked to 770 MHz, with the GeForce 2 GTS:
Expendable 640 x 480 x 16 ---> AVG: 75 FPS / HIGH: 113 FPS / LOW: 52 FPS

GLQuake 640 x 480 x 16 ---> 319 FPS
Quake 2 640 x 480 x 16 ---> 204 FPS
Quake 3 640 x 480 x 16 ---> 81 FPS

3DMark 2000 ---> 6079 points
3DMark 99Max ---> 6735 points (12568 CPU 3DMarks)

Well, this is very strange, I did not expect this at all! The Thunderbird is actually slower (even when overclocked) than the Pluto. Whaaaaa?!


3. GIGABYTE GA-BX2000+ WITH PENTIUM 3 1000 MHz (COPPERMINE)


  • Downclocked to 750 MHz (FSB 100) with the GeForce 2 GTS:
Expendable 640 x 480 x 16 ---> AVG: 67 FPS / HIGH: 94 FPS / LOW: 47 FPS
Expendable 1024 x 768 x 32 ---> AVG: 66 FPS / HIGH: 92 FPS / LOW: 45 FPS

GLQuake 640 x 480 x 16 ---> 277 FPS
GLQuake 1024 x 768 x 32 ---> 170 FPS

Quake 2 640 x 480 x 16 ---> 191 FPS
Quake 2 1024 x 768 x 32 ---> 169 FPS

Quake 3 640 x 480 x 16 ---> 77 FPS
Quake 3 1024 x 768 x 32 ---> 76 FPS

3DMark2000 ---> 5900 points
3DMark99Max --> 6670 points (10982 CPU 3DMarks)

Well, I knew the Athlons were fast, but I thought the Coppermine would be a bit faster. Again, I was wrong: definitely slower than the Pluto and it's trading blows with the Thunderbird. I'm still not sure what's going on here...

  • Default 1 GHz clock speed (FSB 133) with the GeForce 2 GTS:
Expendable Timedemo 640 x 480 x 16 ---> AVG: 91 FPS / HIGH: 123 FPS / LOW: 63 FPS
Expendable Timedemo 1024 x 768 x 32 ---> AVG: 89 FPS / HIGH: 120 FPS / LOW: 61 FPS

GLQuake 640 x 480 x 16 ---> 362 FPS
GLQuake 1024 x 768 x 32 ---> 170 FPS

Quake 2 640 x 480 x 16 ---> 253 FPS
Quake 2 1024 x 768 x 32 ---> 182 FPS

Quake 3 640 x 480 x 16 ---> 109 FPS
Quake 3 1024 x 768 x 32 ---> 96 FPS

3DMark 2000 ---> 7361 points
3DMark 99Max --> 8451 points (14589 CPU 3DMarks)

Finally, at 1 GHz, the Coppermine is obviously the fastest, but I would argue that not by much, you'd think that it would be much faster than the Pluto, with it's crippled L2 cache and lower frequency.

So... what are my conclusions?
Well, for starters, I'm still trying to understand why the Pluto is faster than both the Thunderbird and the Coppermine. Honestly, my intention was always to use the Pluto for my new Slot A build (even if I always assumed that it would be slower than the TB) - it just seemed like it would be the more 'period correct' & 'cooler' choice. However, after also seeing these results, I definitely have all the reasons to go with the Pluto.
What about the video card? Obviously, the 'period correct' choice would be either the Voodoo 3 or the Riva TNT2 Pro, but it seems to me like they would severely bottleneck this platform, particularly at higher resolutions, so I'm going with... *drum roll*... the GeForce 2 GTS (which is still semi-period correct, definitely a video card I would have loved to be able to afford in the year 2000, and I'm sure many people who had either a Riva TNT2 or a GeForce 256 upgraded to it)

Finally, here's two pics (still in unfinished state):
- I still have the GeForce 2 MX installed in the case (will be replaced with the GeForce 2 GTS soon)
- Need to also install the DVD-RW
- Need to add a few 120mm fans... because why not? Maybe I'll go crazy with some RGB fans 😁
- Need to decide which sound cards to go with. I'll probably go with a SB Live 5.1 + an ISA card (haven't decided which one yet)
- Yes, those front USB 3.0 connectors are working 😁

Attachments

  • 20201012_171522.jpg
    Filename
    20201012_171522.jpg
    File size
    1.18 MiB
    Views
    1744 views
    File license
    Fair use/fair dealing exception
  • 20201012_171349.jpg
    Filename
    20201012_171349.jpg
    File size
    762.99 KiB
    Views
    1744 views
    File license
    Fair use/fair dealing exception
Last edited by bloodem on 2020-10-12, 16:24. Edited 3 times in total.

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 1 of 14, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

Well, for starters, I'm still trying to understand why the Pluto is faster than both the Thunderbird and the Coppermine.

Because Thunderbird wasn't direct upgrade over Pluto/Orion, it was mostly cost reduction for OEM and future Socket A release.

750 Mhz Pluto has 512 Kb 300 Mhz L2 cache working via internal 64-bit bus.
700 Mhz Thunderbird has 256 Kb 700 Mhz L2 cache working via internal 64-bit bus.

So that's only 2.3x times faster at the cost of halving cache size. And let's also not forget that K7 has gigantic exclusive (doesn't copy data into L2) 128 Kb L1 cache, so faster but smaller L2 cache aren't necessarily the better one.
Compared to that, Coppermine increased L2 bandwidth 8x times, over Katmai core, by going full speed cache with 256-bit internal bus.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 2 of 14, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie

Yes, indeed, those are valid points. Still, it surprised me, I was sure that a full speed on-die cache would translate to better real world performance. Obviously, that isn't the case.

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 3 of 14, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie

Athlons had insane floating point performance, and well, games kinda liked that. 😀

P6 chip. Triple the speed of the Pentium.
Tualatin: PIII-S @ 1628MHz | QDI Advance 12T | 2GB DDR-310 | 6800GT | X-Fi | 500GB HDD | 3DMark01: 14,059
Dothan: PM @ 2.9GHz | MSI Speedster FA4 | 2GB DDR2-580 | GTX 750Ti | X-Fi | 500GB SSD | 3DMark01: 43,190

Reply 4 of 14, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie

The Slot A build now has a brother (with an identical case): yet another Slot 1 PC 😀
Fun fact: through a series of unforeseen events, my wife ended up building 99% of this PC (and this is actually the first PC she ever built entirely by herself) - she even did basic wire management (which... looks better than mine) 😁.
The only thing I did was to mod the USB front panel connector, to make it compatible with the MB connector.
Yep, it's official... 2020 is weird.

Thank you to @aaronkatrini for the GeForce 2 Pro 😀

THE SPECS:
CPU: Intel Pentium 3 Coppermine 1 GHz FSB133
Motherboard: Gigabyte GA-BX2000+
RAM: 512 MB PC133 @ CL2
VGA 1: Asus V7700PRO/32M (nVIDIA GeForce 2 Pro 32 MB) + custom cooler
VGA 2: Righteous Orchid 3D II (3dfx Voodoo 2 12 MB)
SOUND: Yamaha YMF724F-V + SB-Link cable (one of my favorite sound cards, somehow I always end up using it for most Win98 builds where DOS is also an option)
HDD: Hitachi 82 GB
OS: Windows 98 SE

Some pics:
ga-2000bxplus_1.jpg
ga-2000bxplus_2.jpg
ga-2000bxplus_3.jpg
ga-2000bxplus_4.jpg
ga-2000bxplus_5.jpg
ga-2000bxplus_6.jpg
ga-2000bxplus_7.jpg
ga-2000bxplus_8.jpg
ga-2000bxplus_9.jpg
ga-2000bxplus_10.jpg
ga-2000bxplus_11.jpg
ga-2000bxplus_12.jpg

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 5 of 14, by Delerium

User metadata
Rank Newbie
Rank
Newbie

Did you have Super Bypass enabled? That makes quite big difference.

https://www.anandtech.com/show/474/14

Attachments

  • image016.gif
    Filename
    image016.gif
    File size
    5.67 KiB
    Views
    1536 views
    File license
    Public domain

Reply 6 of 14, by cyclone3d

User metadata
Rank l33t++
Rank
l33t++

I came here to pretty much ask the same thing.

As for the different Athlon CPUs, you could also try "overclocking" the Pluto via the 0 ohm resistor locations on the CPU.

I bet if you take the heat spreader off, you will find that you actually have an 800 or 850 core and the L2 cache could probably be bumped up as well.

It has been a few years, but if I remember right, some of the 750s even came with 900Mhz CPU cores.

Yamaha modified setupds and drivers
Yamaha XG repository
YMF7x4 Guide
Aopen AW744L II SB-LINK

Reply 7 of 14, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie

I know about Super Bypass, however the Gigabyte GA-7IXE does not have a BIOS option to toggle it on/off. Anyway, based on the benchmark results that I got, I assumed that it's probably enabled by default, but I will look further into this when I play with one of the boards again.
I remember from back in the day that there were also some Windows utilities that could enable Super Bypass.

Regarding overclocking, I only tried FSB overclocking (112MHz x 7.5), and it worked without a hitch. However, soon I will have a Goldfinger device, which will allow me to really push these CPUs to their limits 😀 And I'm actually curious if the Thunderbirds will start outperforming the Plutos after a certain clock speed has been surpassed.
Once I have new data I will update this thread.

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 8 of 14, by Joseph_Joestar

User metadata
Rank l33t
Rank
l33t
bloodem wrote on 2020-10-12, 14:53:

Well, this is very strange, I did not expect this at all! The Thunderbird is actually slower (even when overclocked) than the Pluto. Whaaaaa?!

Heh, I remember being similarly surprised back in the day when I got my first Socket A system with a Thunderbird 750. Ended up being slightly slower than some Slot A based Athlons. It was a weird time. But then again, I wasn't as proficient at tweaking memory timings and such back then, and I think I only had PC100 sticks.

bloodem wrote on 2020-11-02, 20:05:

SOUND: Yamaha YMF724F-V + SB-Link cable (one of my favorite sound cards, somehow I always end up using it for most Win98 builds where DOS is also an option)

Yeah, the YMF7x4 cards are incredibly versatile. No other PCI sound card can provide the same feature set, without being paired with something else.

On that note, I recently replaced my generic YMF724 with a Guillemot Maxi Sound Fortissimo which uses higher quality components and also has Digital Out (TOSLINK). It's amazing how much sound clarity one can get out of DOS games using that.

PC#1: Pentium MMX 166 / Soyo SY-5BT / S3 Trio64V+ / Voodoo1 / YMF719 / AWE64 Gold / SC-155
PC#2: AthlonXP 2100+ / ECS K7VTA3 / Voodoo3 / Audigy2 / Vortex2
PC#3: Athlon64 3400+ / Asus K8V-MX / 5900XT / Audigy2
PC#4: i5-3570K / MSI Z77A-G43 / GTX 970 / X-Fi

Reply 9 of 14, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie
Joseph_Joestar wrote on 2021-03-23, 09:47:

Heh, I remember being similarly surprised back in the day when I got my first Socket A system with a Thunderbird 750. Ended up being slightly slower than some Slot A based Athlons. It was a weird time. But then again, I wasn't as proficient at tweaking memory timings and such back then, and I think I only had PC100 sticks.

Yes, in my case both the Pluto and the Thunderbird were tested on the same configuration, same memory & timings, same operating system. So the Thunderbird was clearly slower clock for clock. I guess that, as others have pointed out, Thunderbirds were in fact slower, but were able to scale better and hit higher frequencies, so they were a "long term investment" on AMDs part. 😀
Intel also did the same (but on a MUCH bigger scale) with the Pentium 3 --> Pentium 4 jump. The first Pentium 4 CPUs were a joke, but they did catch up quickly, once the architecture evolved and they were also able to ramp up the clock speeds. Of course, nonetheless, the Netburst architecture remains a sad period in Intel's history.

Joseph_Joestar wrote on 2021-03-23, 09:47:

Yeah, the YMF7x4 cards are incredibly versatile. No other PCI sound card can provide the same feature set, without being paired with something else.

Yes, although I did have issues with them on Socket A KT600/KT880 systems (freezes in Quake 2 when changing resolutions or exiting the game). So on these PCs it's still better to go with a Creative Audigy 2 ZS/SB Live as a primary Windows card and keep the YMF724/744 card for DOS.

Joseph_Joestar wrote on 2021-03-23, 09:47:

On that note, I recently replaced my generic YMF724 with a Guillemot Maxi Sound Fortissimo which uses higher quality components and also has Digital Out (TOSLINK). It's amazing how much sound clarity one can get out of DOS games using that.

That card does look very nice, but it seems that it's missing a SBLink header, which is too bad (then again, it can be used on motherboards that don't have a SBLink connector anyway).

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 10 of 14, by Joseph_Joestar

User metadata
Rank l33t
Rank
l33t
bloodem wrote on 2021-03-23, 11:35:

That card does look very nice, but it seems that it's missing a SBLink header, which is too bad (then again, it can be used on motherboards that don't have a SBLink connector anyway).

It has one, but it's not clearly labeled as such.

The JP6 in the bottom right corner is actually SB-Link. I've tested it on my Abit ZM6 and it works just fine. Kinda strange that they didn't label it properly as someone might think that it's a speaker/line out jumper instead.

PC#1: Pentium MMX 166 / Soyo SY-5BT / S3 Trio64V+ / Voodoo1 / YMF719 / AWE64 Gold / SC-155
PC#2: AthlonXP 2100+ / ECS K7VTA3 / Voodoo3 / Audigy2 / Vortex2
PC#3: Athlon64 3400+ / Asus K8V-MX / 5900XT / Audigy2
PC#4: i5-3570K / MSI Z77A-G43 / GTX 970 / X-Fi

Reply 11 of 14, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie
Joseph_Joestar wrote on 2021-03-23, 12:04:

Kinda strange that they didn't label it properly as someone might think that it's a speaker/line out jumper instead.

Ha, that's exactly what I thought! 😁

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 12 of 14, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie

Thanks to the wonderful people on this forum, we now have easy access to Athlon Golden Finger devices, so naturally I just had to revisit my Slot A build 😀

My Athlon "Pluto" Slot A 750 MHz was actually an 800 MHz part under the hood. Furthermore, it turns out that it's a golden sample - it overclocks LIKE CRAZY.
One of the first things I tried was to set the multiplier to 9, while also keeping the voltage at 1.6V. I was sure that the PC would refuse to POST, but not only did it POST, it perfectly booted Windows.
At first I thought that maybe the voltage was higher than I thought, or that the multiplier was not set correctly... but no. My motherboard has voltage monitoring and it was reporting 1.62V for the VCORE, and the frequency was in fact 900 MHz. Ran a bunch of benchmarks, no issues... however it did crash after 15 minutes of Prime 95 (small FFTs). Not a problem! Quickly increased the voltage to 1.65V and the system was rock solid after that. Amazing! Considering the fact that even original 900 MHz and 1 GHz Orion CPUs come with a default voltage of 1.8V, my CPU is definitely a top performer.

Of course, I also had to try 1 GHz and it did work, however the CPU needed 1.75V to be Prime95 stable. This is still below the default voltage for 1 GHz parts, but temperatures are a bit too high for my liking - apparently 45 degrees C idle/50 degrees C load as shown by the motherboard temp monitoring, but using a thermocouple on the back of the CPU, it turns out that actual temperatures are far higher: 63 degrees C. This means that the CPU die is even hotter, and let's not forget the cache chips which are most likely even more sensitive to heat.

So after a lot of testing, I finally settled for FSB110 x 8.5 multi @ 1.65V (935MHz). The speed is very similar to 1 GHz (FSB 100), but the CPU is a lot cooler. The cache runs at 2/5 divider which results in a whooping 374 MHz (higher than any Athlon in its original state).
Of course, I also added a lot of coolers (2 x 50 mm fans on the CPU, 1 x 120mm intake fan in front of the case, 1 x 120mm intake fan on the case side panel directly above the CPU, 1 x 120mm exhaust fan on the back of the case).

Some benchmarks with the 935 MHz overclock (they can be compared to the 750 MHz benchmarks in the first post, it's the same PC):

  • With the GeForce 2 GTS card:
Expendable 640 x 480 x 16 ---> AVG: 95.36 FPS / HIGH: 136 FPS / LOW: 67 FPS (no sound)

GLQuake 640 x 480 x 16 ---> 362 FPS (with sound)

Quake 2 640 x 480 x 16 ---> 227 FPS (with sound)

Quake 3 640 x 480 x 16 ---> 91 FPS (with sound)

3DMark 2000 ---> 6731 points
3DMark 99Max --> 7476 points (13787 CPU 3DMarks)

So, yeah... pretty similar to the P3 Coppermine 1 GHz - in some tests faster, in others slower (even though that one runs at 133 MHz FSB and it also has on-die full speed cache). Well, come to think of it, the Athlon has a dual pumped FSB, so it is higher 😀
This PC (the GA-7IXE motherboard in particular) is rock solid, did not have a single issue, everything just works - very similar to my 440BX builds, so I really like the AMD 751 chipset. Don't have experience with the VIA KX133 chipset, though; as I understand it should be faster, but not sure about stability.

Regarding the Super Bypass discussion: it was enabled. Turns out that on the later BIOS revisions of the Gigabyte GA-7IXE (version F7 in my case), Super Bypass is enabled by default when certain RAM timings are used (in my case it was enough to keep the timings on Auto, switching to lower manual timings actually resulted in decreased performance by 3 - 10%). More info about this can be seen on this German website: https://www.heise.de/ct/hotline/Speicher-Timi … rds-307644.html

Some pics taken during my tests:

Attachments

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 13 of 14, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie

I replaced the original CPU die thermal paste/cache thermal pads with Arctic MX-2 and Thermal Grizzly minus pad 8 thermal pads.
As can be seen in the first photo, the original cache thermal pads in particular were a joke, one of them was barely touching the cache chip (from what I read on Anandtech, this sloppiness was quite common at the time, especially for Athlons up to 900 MHz).

After doing so, temperatures at 1 GHz / 1.75V dropped significantly (55C during Prime95, as reported by my thermocouple).
Unfortunately, even though the CPU is Prime95 stable at 1 GHz, it's actually only 99% stable during normal use (there are a few random BSODs per day). Bumping the voltage even further does not help in any way.
So, my assumption is that it's related to the external cache chips. At 1 GHz (with the CPU's default cache divider: 2/5), the cache is running at 400 MHz, which is definitely pushing its limits. So I need to find a way to force a 1/3 cache divider on my motherboard.

Unfortunately I have yet to find a solution to the cache divider problem. I tried all possible software solutions (those created by H.Oda! 20+ years ago), tried to modify the Gigabyte 7IXE BIOS file with the FixCache utility, nothing worked... The divider is stuck at 2/5. Apparently there was another utility (WCPUA2B) created by H.Oda, and this one had been successfully tested with the F3 BIOS revision of the GA-7IXE. However, I was unable to find this utility anywhere 🙁
If anybody has any suggestions or a miracle workaround let me know 😀

UPDATE: I had to give up on software solutions for the cache divider problem (they are definitely a waste of time / not very reliable). Instead, I just swapped a few resistors on the Athlon PCB and, lo and behold, the CPU is now PERFECTLY stable @ 1 GHz with 1.75V.

Attachments

  • IMG_1000.jpeg
    Filename
    IMG_1000.jpeg
    File size
    663.05 KiB
    Views
    1289 views
    File license
    Fair use/fair dealing exception
  • IMG_1001.jpeg
    Filename
    IMG_1001.jpeg
    File size
    941.9 KiB
    Views
    1289 views
    File license
    Fair use/fair dealing exception
  • IMG_1021.jpeg
    Filename
    IMG_1021.jpeg
    File size
    717.11 KiB
    Views
    1289 views
    File license
    Fair use/fair dealing exception

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 14 of 14, by supercordo

User metadata
Rank Member
Rank
Member

You guys should come play with us!

Slot A: Aiming for the stars

Attachments

  • 9361.jpg
    Filename
    9361.jpg
    File size
    186.22 KiB
    Views
    616 views
    File license
    Public domain