gbeirn wrote:

I also have a rare mobile Duron that is aactually a palomino core and you can adjust the l2 cache from 64k>128k>256k using a pencil across the bridges.

I've actually at one time tried to find out where the cache on the die is with my main motivation to resurrect damaged Athlons which had part of their cache part broken of, but this idea never made it to a point where I would actually try it out.

Could you please let me know what you know about the bridges and the en/disabling of L2 cache?

Tetrium wrote:

gbeirn wrote:

I also have a rare mobile Duron that is aactually a palomino core and you can adjust the l2 cache from 64k>128k>256k using a pencil across the bridges.

I've actually at one time tried to find out where the cache on the die is with my main motivation to resurrect damaged Athlons which had part of their cache part broken of, but this idea never made it to a point where I would actually try it out.

Could you please let me know what you know about the bridges and the en/disabling of L2 cache?

Here is the information you seek concerning the bridges on these CPUs! 😀

http://fab51.com/cpu/barton/athlon-e23.html

Skyscraper wrote:

Tetrium wrote:

gbeirn wrote:

I also have a rare mobile Duron that is aactually a palomino core and you can adjust the l2 cache from 64k>128k>256k using a pencil across the bridges.

I've actually at one time tried to find out where the cache on the die is with my main motivation to resurrect damaged Athlons which had part of their cache part broken of, but this idea never made it to a point where I would actually try it out.

Could you please let me know what you know about the bridges and the en/disabling of L2 cache?

Here is the information you seek concerning the bridges on these CPUs! 😀

http://fab51.com/cpu/barton/athlon-e23.html

Omg I had actually forgotten about that page! Cheers for the link! 😁

I'm really experiencing exactly how quirky and flakey hardware used to be 15 years ago, but I'm still having fun! 😀

I ran in to an irritating shortcoming of the ECS Photon AF1 KT600 board while I was testing some Bartons in-between Thunderbird CPU changes. The motherboard will not make it though 3Dmark03 "Battle Of Proxycon" with a Barton CPU clocked higher than ~2000 MHz.

At first I though it was an issue with the Cheiftek 350W PSU lacking amps on the 12V rail just as member kanecvr had foreseen so I tested some other PSUs (Seasonic 500W 30A-5V / 33A-12V, FD Newton R2 1000W 30A-5V / 75A-12V and Cooltek 500W 45A-5V / 36A-12V). I think we can exclude the PSU from the list of probable causes.

The issue must be that the Photon AF1s AGP port has a hard time handling the X1950 Pro eventhough the Radeon card takes most of it's power from the two 12V molex jacks. I tested running AGP4X and turning off fast writes and such but no change. As soon as a Barton CPU is cloked at ~2 GHz or higher regardless of multiplier and voltage the compuer will reset during the "Battle Of Proxycon" test.

So I'm tinkering with other K7 stuff this weekend.

Here are two motherboards I should recap... when I have managed to order caps and when I do not work 60 hours a week.

First an Abit KV7 VIA KT600 board which would solve my benching issues, or so I hope. It has a cap blown open but it dosn't really show on this picture but the missing chipset cooler is easy to spot! 😜

And also an Abit NF7-S V2.0, one of the better nForce2 motherboards. The bad cap is easier to spot on this board, I'm thinking of stealing a cap from some other motherboard... 😀

Today I have been testing a board I thought I would use in my "Ultimate Year 2000" build, the Asus A7M266 AMD 760 chipset motherboard. I have now found a better motherboard to use in the Y2K build, I will show that board later.

But lets start with some background, perhaps as much legend or lore as history.

The AMD 760 chipset was released in October 2000 along with the Athlon 1200 even if the availability of motherboards was poor the first months, especially outside Asia. The motherboard you actually could get was AMD's own Corona EVT8 AMD 760 motherboard but most of these motherboards went to system builders.

The seemingly launch ready Asus A7M266 motherboard was shown to the public in October 2000 and the 1.01 revision was perhaps sold in Taiwan but never reviewed and definitely not sold outside Asia. These first non preproduction motherboards were at least in ciculation in one way or another in Asia as it was this version which had the multiplier block actually present on the motherboard and not only silk sceened like on all later revisons.

According to the legend the multiplier block was removed because of demands from AMD. AMD was at the time more than a little bit irritated over the fact that most motherboard makers had show very little interest in promoting the AMD 750 chipset the year before and thus made the removal of the multiplier block a condition for delivering any more chipsets to Asus. This could also be the main reason the Asus A7M266 was a no-show in the first months after the AMD 760 release.

The multiplier block can be solderd in on all revisions but needs some hard to solder and hard to find 8 legged resistor clusters to function. It's a well documented mod but nothing I will try and this is also the reason I will not use this motherboard in my "Ultimate Year 2000" build.

Late 2000 the Asus A7M266 was already at revison 1.03 and these were the first motherboards put on the slow banana boats heading for the US and Europe. The boards would reach retailers and reviewers late February with reviews staring to pop up shortly after. Revision 1.04 started showing up in Asia in January 2001 with rumors starting to spread on forums that this revsion in fact had a functioning multiplier block, the rumors were of course false.

How the overclocking community even knew the multiplier block was missing from rev 1.02 and 1.03 I dont know as these posts in January 2001 were among the the first forum posts in English about the Asus A7M266 I could find other than speculation about the Asus A7M266 and AMD 760 in general prior to the AMD 760 release. The chipset and a preproduction sample of the A7M266 had been shown on some event during the summer 2000.

Revision 1.02 of the Asus A7M would also show up in Europe and USA but from what I can gather all reviwed motherboards were revison 1.03, 1.04 and 1.05. For some reason (cough cough Intel) the AMD 760 chipset just like it's predecessor the AMD 750 chipset was not promoted much by the motherboard manufacturers. The fact that AMD wanted a pretty penny for every sold chipset and the high prices of DDR memory probably also played parts in the lack of interest from the board makers.

Many hardware sites diddn't even get review samples of AMD 760 motherboards at all and only reviewed motherboards during the summer 2001 when they could borrow motherboards from retailers. Some even wrote that the chipset was old technology on it's way out soon to be replaced by VIAs KT266A in their first AMD 760 reviews but at the same time all was impressed by the chipsets performance.

This is my Asus A7M266 running a Thunderbird AXIA 1000 and a Radeon X850XT video card!

This is the final BIOS, it do not recognise ... well hardly any CPUs correctly. Not even models released early summer 2000 running at stock speed... Asus just didn't care! 😁

Oldskoolmaniac wrote:

Anyone else have any weird flukes with win98 and a nforce2 chipset board, driver issues?

Yup. Not the best chipset for win98. I prefer the KT880 myself. Good brand KT400 or KT600 boards also work well in a pinch.

@Skyscraper - two issues here... first off, the KT600 is not a very good chipset for high performance parts. Then again, I'd advise against using the X1950 - or a 7800 AGP for that matter - on a socket A machine - the architecture just can't feed the card enough power or data and it will perform like crap or cause instability. Second issue is the KT600 and most socket A boards take CPU power from the 5V rail - this includes most nForce 2 boards, even if they have an auxiliary 12v CPU power connector - so it's a good idea to use a PSU with at least 20A 5V rail for something like a 3200+. The CPU does not use even close to that much power (20ax12v=240w) but it can have a 1-2ns power spike when in heavy load when it can draw close to 15A - at least that's what my equipment discovered while running benchmarks.

This holds true for video cards as well - especially more modern ones, but it includes the X1950. This is the reason most X1950 cards use a 6pin PCI-E connector or 2x Molex - current spikes. You can measure this yourself if you have the equipment.

kanecvr wrote:

Yup. Not the best chipset for win98. I prefer the KT880 myself. Good brand KT400 or KT600 boards also work well in a pinch. […]
Show full quote

Oldskoolmaniac wrote:

Anyone else have any weird flukes with win98 and a nforce2 chipset board, driver issues?

Yup. Not the best chipset for win98. I prefer the KT880 myself. Good brand KT400 or KT600 boards also work well in a pinch.

@Skyscraper - two issues here... first off, the KT600 is not a very good chipset for high performance parts. Then again, I'd advise against using the X1950 - or a 7800 AGP for that matter - on a socket A machine - the architecture just can't feed the card enough power or data and it will perform like crap or cause instability. Second issue is the KT600 and most socket A boards take CPU power from the 5V rail - this includes most nForce 2 boards, even if they have an auxiliary 12v CPU power connector - so it's a good idea to use a PSU with at least 20A 5V rail for something like a 3200+. The CPU does not use even close to that much power (20ax12v=240w) but it can have a 1-2ns power spike when in heavy load when it can draw close to 15A - at least that's what my equipment discovered while running benchmarks.

This holds true for video cards as well - especially more modern ones, but it includes the X1950. This is the reason most X1950 cards use a 6pin PCI-E connector or 2x Molex - current spikes. You can measure this yourself if you have the equipment.

The reason for using for example a Radeon X1950 Pro with an Athlon XP3200+ is to see how new games can be made to work decently. My thought was to use the same video card for all my basic performance testing so results can be compared between CPUs and motherboards. Perhaps I need to back down to a Geforce 6800GT because of compatibility but that will seriously limit what kind of 2006-2008 games the faster CPUs and motherboards can run.

I did not save many screenshots from the Barton testing as I only double checkled that these CPUs still worked. I did save this one with the 3dmark 2001 result for an Athlon XP 2500+ @3200+. The 3dmark03 score at stock 1866 MHz which works fine was pretty decent but a Barton needs to run at 2200+ Mhz to make good use of a video card like this. The Athlon XP 2500+ CPU is an "I stepping" so the CPU stability is not an issue at 2200+ MHz even at 1.6V. The 166 MHz FSB version of the Athlon XP 3000+ would also cause restarts in 3dmark03 until I backed the frequency down to under 2000 MHz.

Barton 2500+ @ 3200+, ECS Photon AF1, Radeon X1950 Pro 512 AGP: 3dmark 2001.

I ran into the samme issue more or less with the Radeon X850XT when doing some test runs with the Asus A7M266. It would work great with a Thunderbird even overvolted and overclocked but it freezes with an Athlon XP 2100+ Palomino at stock 1733 MHz. I have still not tinkered with the AGP settings on this motherboard so perhaps it can be solved but I doubt it.

Even my MSI Master 2 Far from mid 2003 can run the Radeon X1950 Pro without issues and it's older than the ECS KT600 motherboard. That the Asus K7M266 has these issues I can understand but motherboards from late 2003 or early 2004 should be able to run a video card from 2006 as that is pretty much the normal upgrade cycle.

I have a feeling that the Abit NF7-S V2.0 won't have these kind of issues and I can't remeber running into these kind of problems with the Asus A7N8X eventhough it lacks auxiliary 12v CPU power connector. The Asus might be an Barton killer but it's at least stable as a rock at stock speeds (and at moderate overclocks using less than 1.75V).

Edit

Yay! I solved the issue with the Asus A7M266 motherboard, this board holds up better than the crappy ECS Photon AF1. It was either "Fast Writes" or some AGP memory precaching setting that solved the issue. 😀

/Edit

Edit2

It seems the isse with the X850XT on the Asus A7M266 wasn't fast writes or any other AGP settings but more about getting the total performance level down to a certain point.

This issue is now fully solved it seems, the solution was a bit odd. Up until now I have been using a single 1GB DDR stick capable of running ~235 MHz CL2-3-2-6 at 2.5V. This stick usually works great with older motherboards at a lower speed even with motherboards not liking some other 1GB sticks and it seemed to work perfectly in the Asus A7M266. I even Primed for an hour this morning and also did a quick test with HCI memtest and everything checks out good.

The reason I started to doubt the 1GB sticks stability in the Asus A7M266 was that when I tested using two Samsung 512MB sticks late last night I could suddenly turn the awesomesauce knob to maximum without the system locking up during heavy 3D loads. It seems that the Asus A7M266 do not like running this 1GB stick in combination with a really fast CPU and video card in 3D even if it works fine in 2D and with slower CPUs and/or video cards...

I kind of doubt the issue is the same with the ECS Photon AF1 KT600 board but it's worth a try since I have used the same 1GB memory stick in that motherboard aswell.

/EDit2

Edit3

Progress!

I tested the Radeon X1950 Pro 512 AGP with the A7M266 to see if the combination would work but I ran in to the exact same issue as I did with the ECS Photon AF1. The computer resets during heavy 3D load with a fast CPU... This is a bit strange I thought as it works with the Radeon X850XT which probably uses just as much power and performs on about the same level on a system like the Asus A7M266 with a Palomino Athlon XP 2100+...

What was the difference? The difference was the driver, with the X850XT I used Catalyst 7.1 and with the X1950 Pro I used Catalyst 9.3 to be able to compare the results with my KT600 numbers. It seems Catalyst 9.3 isn't 100% stable with this X1950 Pro card, at least not with the A7M266 and a fast CPU eventhough I have used Catalyst 9.3 with other X1950 Pro cards running on Intel and K8 boards.

Now I hope that Catalyst 7.1 will solve my issues with the X1950 Pro running on the KT600 board. I will have to redo all 3D/game benchmarks but I would have to do that if switching to another video card anyhow.

/Edit3

... I've never experienced those issues with any of those drivers. I suppose it could be an issue with the chipset / driver combination. Try setting the x1950 up on the Nforce 2 board.

kanecvr wrote:

... I've never experienced those issues with any of those drivers. I suppose it could be an issue with the chipset / driver combination. Try setting the x1950 up on the Nforce 2 board.

I will later. 😀

Now I want to mess with the other chipsets! 😁

The main point of this thread isn't to build the best stable socket A gaming system but to tinker, bench, compare and find just these kinds of odd quirks, and their solutions.

The Internet is full of information about the nForce2 chipset, its performance and quirks but when it comes to the AMD 760 chipset there isn't much information at all. The gaps need to be filled! 😁

I don't know how good your cooling is on that x1950 Pro, but I recommend checking temps on GPU and it's Rialto PCI-e to AGP bridge chip.
Increase graphics card fan speed manually with RivaTuner, and see if that will help with stability during heavy load.
Adding one fan (on side panel for example), to increase airflow over chipset and top of the graphics card would also be a good idea.

agent_x007 wrote:

I don't know how good your cooling is on that x1950 Pro, but I recommend checking temps on GPU and it's Rialto PCI-e to AGP bridge chip.
Increase graphics card fan speed manually with RivaTuner, and see if that will help with stability during heavy load.
Adding one fan (on side panel for example), to increase airflow over chipset and top of the graphics card would also be a good idea.

You can see the setup a few posts up even if it's the X850XT pictured and not the X1950 Pro. 😀

But looking over the cooling on the Radeon X1950 Pro cooling is probably a good idea, I did that with the X850XT because of a irritating fan noise and the heatsink was clogged with dust.

The main differece I see between Catalyst 7.1 and 9.3 is that there is a much more even frame rate with the 7.1 driver while the 9.3 one spikes and drops the FPS more which probably causes spikes in the power consumtion. Catalyst 9.3 seems to be ~3% faster in 3dmark 2001 and 03 on avarage though.

I managed to get a Barton running on the Asus A7M266 but the motherboard did not seem to like my multiplier locked T-bred 2200+. The issue seems to be that the motherboard gets confused with unsupported multipliers but as the Barton I tried is "super locked" it works. I think the Barton XP-M could work without modifictions eventhough it's unlocked as it normally defaults to 6x on desktop motherboards, the issue seem to be with higher multipliers.

Edit

The reason for the Bartons working and the T-bred CPUs not working did not have anything to do wil locked CPUs, see next page for details.

/Edit

Not much tinkering today but I got a couple of new CPUs.

Two Athlon XP-M 2500+ (1.45V) from France and also a CPU lot including two Slot A CPUs.

I have already posted these two pictures in the "Bought this crap thread" but I might aswell post them here too as they are on topic.

I won this CPU lot with an opening bid of £20! The auction had been in my watch list for a long time and no one else seemed interested in these CPUs. I took a closer look a couple of weeks ago and then decided to place a bid eventhough the shipping cost was pretty steep as you can see. As I now post these two pictures in this thread I think we can assume it's not the boring Slot-1 CPUs that cought my attention!:D

The sellers picture displaying the gods (it was the only one).

What was it I wanted to confirm by looking really close at the picture?

In fact the seller had got something wrong but not at all in the way I first suspected, lets see if you can figure it out. Resize, enhance, sharpen!

As a hint I can say that it's something about the cooling that just dosn't add up with the specifications of the CPU it's mounted on if we go by the model numbers the seller has listed.

Haha didn't buy cause I trusted the description 😁
GG.

sunaiac wrote:

Haha didn't buy cause I trusted the description 😁
GG.

When I googled those model numbers and saw that it was a Thunderbird 650 and not an Argon or Pluto I got more interested in the auction as I own a Gold Finger Device.

But I also noticed that the Slot A Thunderbird 650 dosn't use much power at all and that cooler looked like something you would mount on an Argon 650, not a Thunderbird 650 (at least not unless you are overclocking but the CPU looked untouched).

I thought perhaps the seller had mislabled it and the CPU in fact was an Argon, I'm not really in need of one of those hot bastards. I wanted to take a closer look at the picture to make sure it was a Thunderbird 650 but I was at work and only had MS Paint at hand...

When I resized the picture I could not really believe what I thought I saw.

Today I can provide a much sharper picture and now I know that it's in fact is a Thunderbird! 😁

Skyscraper wrote:

When I googled those model numbers and saw that it was a Thunderbird 650 and not an Argon or Pluto I got more interested in the […]
Show full quote

sunaiac wrote:

Haha didn't buy cause I trusted the description 😁
GG.

When I googled those model numbers and saw that it was a Thunderbird 650 and not an Argon or Pluto I got more interested in the auction as I own a Gold Finger Device.

But I also noticed that the Slot A Thunderbird 650 dosn't use much power at all and that cooler looked like something you would mount on an Argon 650, not a Thunderbird 650 (at least not unless you are overclocking but the CPU looked untouched).

I thought perhaps the seller had mislabled it and the CPU in fact was an Argon, I'm not really in need of one of those hot bastards. I wanted to take a closer look at the picture to make sure it was a Thunderbird 650 but I was at work and only had MS Paint at hand...

When I resized the picture I could not really believe what I thought I saw.

Do I see what I think I see.jpg

Today I can provide a much sharper picture and now I know that it's in fact is a Thunderbird! 😁

wa-wa-wee-wa.jpg

Is that the Thunderbird CPU of the listing, acording to the model number, is an Athlon 1000 Thunderbird Slot A acording to the model number on the CPU, the fatest Slot A Thunderbird

model number: AMD-A1000MMR24B A

http://www.cpu-world.com/CPUs/K7/AMD-Athlon%2 … MMR24B%20A.html

Skyscraper wrote:

When I googled those model numbers and saw that it was a Thunderbird 650 and not an Argon or Pluto I got more interested in the […]
Show full quote

sunaiac wrote:

Haha didn't buy cause I trusted the description 😁
GG.

When I googled those model numbers and saw that it was a Thunderbird 650 and not an Argon or Pluto I got more interested in the auction as I own a Gold Finger Device.

But I also noticed that the Slot A Thunderbird 650 dosn't use much power at all and that cooler looked like something you would mount on an Argon 650, not a Thunderbird 650 (at least not unless you are overclocking but the CPU looked untouched).

I thought perhaps the seller had mislabled it and the CPU in fact was an Argon, I'm not really in need of one of those hot bastards. I wanted to take a closer look at the picture to make sure it was a Thunderbird 650 but I was at work and only had MS Paint at hand...

When I resized the picture I could not really believe what I thought I saw.

Do I see what I think I see.jpg

Today I can provide a much sharper picture and now I know that it's in fact is a Thunderbird! 😁

wa-wa-wee-wa.jpg

Got one of my 1000s in the same kind of lot for 19$ 😁
Now good luck finding a MB for it.
"Tbirds work on all amd750" is urban legend 😜

sunaiac wrote:

Got one of my 1000s in the same kind of lot for 19$ 😁
Now good luck finding a MB for it.
"Tbirds work on all amd750" is urban legend 😜

Perhaps it works in this quality motherboard costing 9€ new! 😁

http://www.computerstunt.nl/onderdeel/1587/mo … nu-maar-12.html

They are almost double the price on Ebay, 17.5 euro. 😀

I'm looking at the Asus K7M on Ebay but the ones availible at the moment are pricey, I'm not in any hurry though.

When researching Slot A motherboards you find more opinions and cheeky remarks than useful facts. 😀

"I wasn't nitpicking, I was pointing out the humor of "best slot a motherboard." it's quite similar to "most prestigeous eastern european car," or perhaps, "best tropical disease"

😀

I haven't tried the K7M.

What I can tell you is that :
- my GA-7IX works fine (It's a non super-byapss capable revision of the chipset) with a 1000 tbird.
- my GA-7IXE works fine (super-bypass and seemingly stable AGP2x revision of the chipset) with a 1000 tbird.
- None of my 3 K7 pro works, despite the 1000 tbird being on the compatibility list, even with the last BIOS. But a 700 tbird OC'd to 833 works. And the 1000 works with a 95 FSB at 950MHz. So it might just be some tuning missing in the BIOS compared to gigabyte.

I'll try to find 768 MB of memory this evening for less than an arm on ebay, so that I can get up and running.

Thanks for the information about motherboards working with the Thunderbird 1000. On "The Internet" most people have heard that this or that will work but as the Slot A Thunderbirds are rare few people seem to know anything for sure.

There is not really any hurry when it comes to the benchmarkng as I seem to have a few issues to work out on my end as well. 😁

256MB PC133 ram sticks are good to have anyhow though and they can usually be had for next to nothing. At least they used to be cheap, I rarly check Ebay memory prices as I kind of have "enough". 😀

sunaiac wrote:

I haven't tried the K7M. […]
Show full quote

I haven't tried the K7M.

What I can tell you is that :
- my GA-7IX works fine (It's a non super-byapss capable revision of the chipset) with a 1000 tbird.
- my GA-7IXE works fine (super-bypass and seemingly stable AGP2x revision of the chipset) with a 1000 tbird.
- None of my 3 K7 pro works, despite the 1000 tbird being on the compatibility list, even with the last BIOS. But a 700 tbird OC'd to 833 works. And the 1000 works with a 95 FSB at 950MHz. So it might just be some tuning missing in the BIOS compared to gigabyte.

I'll try to find 768 MB of memory this evening for less than an arm on ebay, so that I can get up and running.

What about VIA chipsets based mobos like the VIA KX133?, the VIA counterpart for the AMD 750, it supports AGP 4x with the universal slot, up to 2 GB PC133 (512 MB per stick), ATA-66 (686A southbridge)

I only own a NOS Asus K7V which I did not get time to test yet (I'm not interested in the KX133, my build is made around the AMD750 because it's voluntarily a full AMD build)
This thread will "force" me to finally test it, I'll report once it's done.
But I don't have much hope, I think only the K7V-T is "supposed" to be thunderbird compatible with models over 700MHz.