Dear Voodoo(²) Driver Hackers ??? ,
well, i took this adress because i think that nobody else may have interest in this ;) . The Voodoo² driver (latest reference) sets both K6-2 (those with the newer CTX-CORE, i think this applies to K6-2+, K6 III and K6 III+ as well) MemoryTypeRangeRegisters to allow WriteCombining access to the Voodoo² framebuffer. MTRR1 is set to 16MB WC at D5000000 (this adress may vary depending your configuration). MTRR2 is set to 0,1MB UnCached at  D5000000 because the Voodoo² has it's IO also in the linear memory instead via the real IO port and the K6-2 MTRRs can be set to 2 digits (D5 for example) only. The reason because MTTR2 must be set is that if the IO part of the linear memory is set to WC the Voodoo² crashes. Now it would be interesting how the driver handles two Voodoo² in SLI Mode because the K6-2 has only 2 MTTRs. By the way does anyone know how Intel handles MTTRs?

I also checked 3DMARK99 MAX how much effect the MTTRs have on Performance. On my ATI Xpert@Work (RAGE PRO 4MB SGRAM AGP2X) there was nearly no effect in all tests (0,2% advantage in some tests and also some disadvantages). On my Voodoo² there were also nearly no effects in the practical benchemarks (0,4% in some tests and also some disadvantages) but some big difference in the theoretical KPolygons/s tests. The 6/25/50 tests were 14% (individual) and 18% (strips) slower without MTTR's. The 250 and 1000 tests had no difference. I havn't checked specific games for benchmarking yet but the effect in the practical benchmarks of 3DMARK is so minor that there won't be big differences.

Last but not least i can't check 3DMARK with the original Voodoo Graphics because there are only DirectX 5 drivers for that card.

Regards,
Andreas

... nearly forgot. Anyone who want's to experiment with MTTRs on K6-2. Here you can find a program to set these mysterious registers ftp://ftp.heise.de/pub/ct/ctsi/setk6v3.zip ...

It seems the K6 series has many performance functions that are not widely used. I also have a tweak program called CTU by Rob Muller that can do the same things as SETK6 and is a real windows application. It also shows an option called Write Ordering that i never heard of. But these in-processor tweaks and Super 7 bioses in general are nothing compared to the mass of options, especially for rams, that a a modern Athlon board has. I  ::) and  :o whe i saw the bios of a XP1600 from a friend of mine!

Regards,
Andreas

Yep, there's alot of Stuff that can be done for those older Super Socket7 Systems.

VIA MVP3 have the hidden 4way Memory Interleave, and for the ALI V, there's a Truckload of very effective Registry Tweaks, that will boost performance alot.

Most interesting are also the advanced features of the K6-2+, K6-III and K6-III+ Variants. Besides the onchip L2 Caches, I think those also support the updated MMX+ ...

Too bad they were not sold in large quantities, back in those days one could only dream of such a CPU for the Socket7 System :)

I still keep one Socket7 System in service, as a heritage to the Socket7 CPU's I still have :
K6-2 550MHz, 2.3Vcore
A rare K6-III+ 500MHz, 2.0Vcore
A very rare K6-III+ 550MHz, 2.0Vcore (currently running in the Network)  ;D

WHOA! :o

Falconfly has a K6-III+ 500MHz 'AND' a K6-III+ 550MHz!

Damn, now that's RARE! ;D

I recently got a k6-III+ 450MHz for my old laptop (luvverly stuff!) along with a combo drive. The laptop runs a lot better now! :D

Another cool feature of the k6-2/3+ CPU's is software overclocking! ::)

Now WHY didn't they keep that feature?  ???

Congratulations to this great AMD collection. You prevented the Intel-Sockel-Chaos, the Celeron faster as a same clocked PII and PII300 sold as PII266 and all the other kwirks (all fantastic portraied in the c't Prozessorgeflüster) by staying on the AMD line :)

Also in the c't was written that the K6III was planned as a competitor to the Pentium III. Why has AMD forced the Athlon on SlotA to make it a SockelA a short time later  :-/

By the way the 4way Memory Interleave wasn't so secret. At least a friend of mine has an Sockel7 NMC ATX board that has this option in BIOS (2way, 4way selectable).

Regards,
Andreas

P.S.
VIA-VP3 Bioses let you select between CL3 and CL2 for SDRAM but they only work as 2-3-3 with the CL2 setting. You also have to set Timing from 10NS-SDRAM (other options in my bios are 8NS slow medium fast) to turbo to run the SDRAM in 2-2-2 mode. This gives another small boost. I confirmed this with SANDRA.

Hm, I'll check this out to make sure (although performance is set to Turbo, and PC133 SD-RAM is in that Machine)

One great part of the Socket7 Architecture should have been taken to the Athlon Platforms though :

The Cache :)

Really nothing beats the nice boost of 1MB, or even 2MB L3 Cache ;)

A pity that useful (and not too expensive) performance booster was abandoned :(