VOGONS


First post, by PentAmd

User metadata
Rank Newbie
Rank
Newbie

Hello!

I have a ASUS A7M266-D dual Socket-A motherboard with 2pcs Athlon MP 2000+. The mobo supports only FSB 100 (200) and 133 (266) MHz.
I have 1 GB (2 module) non ECC non REG memory in this mobo.
Memtest86 passed, then I installed a windows XP on it, and in the win task manager I can see the two CPU.
So everything is fine.

I wanted to have faster processor with double of L2 cache. So I checked wikipedia, what kind of Athlon MP processors were produced. I saw, that there was two Bartons, the 2600+ and 2800+ with FSB 266.
In the E-Bay there were not much Athlon MP, so I searched in my collection 2 nearly same Athlon XP 2500+ (333MHz) Bartons. These processors are known for the ability to clocked higher FSB, and can reach the 3200+.
Because my dual processor motherboard supports only 266FSB, I changed the clock multiplier by cutting thru some of the L3 lines on the processor. Thus I made a 2800+ Barton with 266FSB.
Bios post recognised it coreectly and I tested both processors individually with memtest, they passed.
After the successful multiplier modding, I converted both of them to Athlon MP (enabled the multiprocessor support), connecting the last pins of the L5 on the proccessor by using conductive ink.
According to the Bios Post screen it was successful, both of them were MP capable.

My first problem was the Memtest86. For multiprocessor systems I use the "newer" version of memtest V 4.3.7. After start, it showed the following error: "SMP: After STARTUP IPI: err = 0x4"
I looked up this kind of error on the interet, and it is related the reset of the multi processor system, which was somehow not successful. I did not have this error message with the "real" Athlon MP 2000+.

I tried to run Windows XP, but it recognises only one processor.

Everywhere on the internet forums I read that the XP and MPs have the same core. After enabling the MP feature, why it is not working? Maybe during production the Multiprocessor unit of the CPU was faulty after they tested the silicons, and that's the reason why they disabled the feature?

On thing to mention is, that the Athlon XP 2500+ CPUs which I use were made in 2003 week 21 and week 41.

Reply 1 of 9, by swaaye

User metadata
Rank l33t++
Rank
l33t++

Are you sure you got the L5 connected? Connecting the bridges on the XP chips was a lot more difficult than with the older Athlons with ceramic packaging. I used to socket mod them instead of trying to multiplier unlock via the bridges.

Reply 3 of 9, by PentAmd

User metadata
Rank Newbie
Rank
Newbie
swaaye wrote on 2024-10-31, 00:23:

Are you sure you got the L5 connected?

The bios recognises the cpu as MP. Without mod the CPU0 is not showing up at all.

Last edited by PentAmd on 2024-11-01, 10:57. Edited 1 time in total.

Reply 5 of 9, by PentAmd

User metadata
Rank Newbie
Rank
Newbie

I made a last unsuccessful attempt:
I compared the pictures between a MP2800+ and a XP2500+. I saw difference in the L12 jumpers. On the XP only one jumper was opened (cut). On the MP there are two. So I made the second cut, but it changed nothin.
So now I have two theories. As I said before, maybe the SMP was faulty, so AMD sold the processors as stand alone. Or via bounding the SMP releated stuff (pinouts of silicon) are WILLINGLY not connected to the package.

Reply 6 of 9, by PentAmd

User metadata
Rank Newbie
Rank
Newbie

I found the solution:
Bios version 1011 beta 005 from 2004-03-15 solved the problem. It seems that the bios with .003 ending was not that good for my newer Bartons.

I would like to thank you member Darry, who shared some links in this post, which helped me to find the solution:
Beta Bios 101A7MDBBF10.ROM (2010/05/27) for the Asus motherboard A7M266-D?

Reply 8 of 9, by PentAmd

User metadata
Rank Newbie
Rank
Newbie

Hello Zami555!

Because I am a freetime retro proggramer I can tell you, what differece a barton makes.
As I mentioned in my first post the bus speed is 266MHz. This is also the RAM access speed. To convert this speed to trasfer rate it makes 8byte * 266MHz = 2128 MB/sec. (8 byte means, that the CPU can read 64bit=8byte in one cycle)
Compared to the 400FSB 3200+ this is only 2/3 speed.
The "slow" memory access can improved having bigger L2 cache, because the L2 cache is much faster to acces as the RAM.
It is only in theory, the real world programs are not optimized for a random selected CPUs. That means you cannot see much differece in the result.
But there is another thing. Before the Geforce 2 GTS card, the transformation of vertices in a 3D scene was made in the CPU. (this means 10 millions of multiplication, additon in every 16ms - if you want 60FPS). Because the camera position in a FPS game allways changing, the CPU must do a lot of MMX (P2/K6) or SSE (P3/Athlon XP) floating point calculations. So I think with the higher multiplier of the CPU, the transformation will be 28% faster, compared to the Athlon MP 2000+. Of course it depens on the rendered screen. UT99 for example has very much limited triangle count I think 300 triangles/scene, so there is not much gain, but latter games has more and more demand for triangles.
This mobo can accept a Voodoo 5, where the triagle transformation is done in the CPU, so I am going to test some games with it.

Unfortunately non of the games that era was programmed to support multithread (and with that multiprocessor). But during retrocoding of a indy-demo-game maybe I am able to achive some improvement.
One thing I really like with multiprocessor motherboards, that the Windows XP User-Interface is much more responsive, than with 1 CPU. Especially if the CPU is 100% used.

Reply 9 of 9, by zami555

User metadata
Rank Newbie
Rank
Newbie

Wow, excellent analysis. Hats off. Thank you very much for your deep explanation on the benefits of new CPU. I read it with great interest.
As you've mentioned the multi CPU is not a big benefit in terms of games. Anyway in your case the Barton CPU is a great step in direction of maxing out this platform