K6-III 400 needs 2.4V. TDP is 16W, tops out at 27W.
K6-2+ 500 needs 2V. TDP is 13W, tops out at 16W.

With 2.1V it should be slightly higher, but still better than K6-III. Take a good look at the caps on the board and see if any of them is bulging. If they look in good shape, they usually are fine and you've got nothing to worry. It was only with later CPUs that things got out of hand (fast Pentium III, Athlons).

Thank. I tested both, in 3dmark CPU test the K6-2+ get 300 more points of the K6-3 400 but overall score is similar. It seems that the K6-2+ render faster web pages and faster desktop usage but oevrally I would say that they are similar. I didn't remember that the K6-2+ had 3dnow enhanced istructions. It would be nice to use the Powernow! energy features I don't know if they are used without o.s. supporting.

My K6-III+ needs just 1.6v.

Why some need 2.0v (or more) and other 1.6v at factory default? Considering they had the same process how could they get to stabilize it without 0,4v?

http://www.amd-k6.com/cpu-specs/

Looks like the 1.6v CPUs are for embedded applications. Lower voltage provides for lower power consumption.

philscomputerlab wrote:

My K6-III+ needs just 1.6v.

Phil, does the Socket7 MB you use go that low voltage wise? If so, which do you use? I know many desktop Super Socket7 boards only go down to 2.0V.

swaaye wrote:

http://www.amd-k6.com/cpu-specs/

Looks like the 1.6v CPUs are for embedded applications. Lower voltage provides for lower power consumption.

Yup that's the one!

QBiN wrote:

philscomputerlab wrote:

My K6-III+ needs just 1.6v.

Phil, does the Socket7 MB you use go that low voltage wise? If so, which do you use? I know many desktop Super Socket7 boards only go down to 2.0V.

Not all of them. Some only go down to 2.1. So it depends. But the Gigbayte GA-5AX goes down that low if I remember correctly.

386SX wrote:

Why some need 2.0v (or more) and other 1.6v at factory default? Considering they had the same process how could they get to stabilize it without 0,4v?

There are different models with different voltage requirements. I got mine from a German eBay seller I believe, he had quite a few.

For power/voltage requirements, I find http://pclinks.xtreemhost.com/elec_pentium.htm is a good source.
The thing is that there are different models even of the same speed grade.
Take eg. K6-2 450:
The AFX variant take 2.2V, typically 11.3W (typical power is not the same as TDP, but probably close enough), maximum 18.8W
AHX takes 2.4V, typically 17.05W, maximum 28.4W.

(Those 3 letter shortcuts refer to the rated voltage and maximum temperature.)

DFI K6XV3/66+ supports 1.6V has an excellent layout and has 3 ISA slots.

Trying and trying both cpu it's difficult to see which is faster. Seems they both has their benefits but I would say similar at the end. One strange thing is that at 2.1v the K6-2+ 500 seems to run at higher temperature on the heatsink that the K6-3 at 2.3... I remember tried many years ago a K6-2+ 533 2.0v on a Ali chipset and it ran MUCH cooler...

Is a K6-2 500 drastically slower than a K6-2+ or a K6-3/K6-3+ when it's used with a Revision G Alladin V Chipset?

Revision G on the asus p5a/p5a-b runs very Slow with a + CPU. Normal k6-2 and k6-3 work fine.
This is a bug and there is no fix. Ali rev. G is used on p5a/p5a-b 1.05 and 1.06.

Older revisions work fine as they use older Ali v chipset revisions.

Don't know about boards from other brands.

meljor wrote:

Revision G on the asus p5a/p5a-b runs very Slow with a + CPU. Normal k6-2 and k6-3 work fine. This is a bug and there is no fix. […]
Show full quote

Revision G on the asus p5a/p5a-b runs very Slow with a + CPU. Normal k6-2 and k6-3 work fine.
This is a bug and there is no fix. Ali rev. G is used on p5a/p5a-b 1.05 and 1.06.

Older revisions work fine as they use older Ali v chipset revisions.

Don't know about boards from other brands.

Ok thanks.

What I wanted to know though was in reference to this:

http://www.amd-k6.com/category/hardware/

With these chips and their onboard L2 cache (256 kB for the K6-III(+) and 128 kB for the K6-2+) all these issues magically disappear and the limit is 4 GB! So just install one of the “+” chips if you can and need RAM…

So the question is, with a K6-2/550 and a Revision G chipset (with say 768MB of ram), would there be a significant difference in speed between the vanilla or plus version?

Is this like asking if a + chip is faster in general?

Here's a link that gets posted weekly on here. Includes + vs plain. I don't see info about how much RAM they used with the K6-2. It may be running with uncacheable memory.
http://www.tomshardware.com/reviews/benchmark … hon,590-23.html

swaaye wrote:

Is this like asking if a + chip is faster in general?

Here's a link that gets posted weekly on here. Includes + vs plain.
http://www.tomshardware.com/reviews/benchmark … hon,590-23.html

I guess what I'm asking is how a vanilla K6-2 chip compares to a K6-3 chip when the K6-2 chip is paired with an Aladdin V and maxed out with RAM.

The reason I ask is because that page mentions that earlier Aladdin V chipsets couldn't cache as much RAM, so then the question remains, how much greater is the speed boost from the on-die cache vs. the larger cacheable ram.

I can say with confidence that accessing uncacheable RAM is not fast. 😀 This is how people here simulate 486 and such. Disabling the L2. But if you have a large buffer of cacheable memory like 512MB cacheable and then 256MB uncacheable the loss would be less awful. AFAIK the cacheable memory is used first.

K6 III would cache all of the RAM in its internal L2 and the motherboard would still provide additional L3 up to 512MB.

I really ought to test it. I have an Asus P5A 1.06 in my possession along with a K6-2 500 AFX. Interestingly, the 1.06 version is missing the TAG ram chip, but it is equipped with a "tm" T35L6464A-5Q" (4092KB) for the L2 Cache. Anyone know anything about that?

I'm betting all the benchmarks out there are with only 128MB of RAM cached. In reality, how much faster is the K6-3 or K6-2+ (K6-3+ is just a die shrink of the K6-III, and the K6-2+ is just a K6-III with half the L2 cache) when the K6-2 doesn't have its hands tied behind its back.

What I now understand about the K6-III and the + being able to cache the maximum amount of RAM regardless of the Aladdin revision is that this ability has nothing to do with the L2 cache being on-die per se, but rather it has more to do with how it bypasses the limitation of the chipset by handling it differently (something to do with tag RAM).

Anyone want to loan me a K6-III for some benchmark comparisons? And which benchmarks should I run (memory bandwidth/latency benchmarks would be particularly of interest here)?

In the video I posted earlier, I've benchmarked with 128 MB and 512 MB, comparing the K6-2, K6-2+ and K6-III+ at 550 MHz. Because of popularity, I've also included the K6-III+ @ 100 x 4 = 400 MHz.

I've also included test with motherboard cache disabled.

Under Windows at least, the K6-III+ @ 100 x 4 = 400 MHz is pretty much on par with the K6-2 @ 550 MHz.

philscomputerlab wrote:

raymangold wrote:

Did you try the 'performance boost' from the Win98SE unofficial service pack? It states it requires at least 256 MB of RAM to operate. I wonder if that would increase performance when cacheable memory is changed from 128 MB to 512 MB.

Nope. Never tried the service pack. I don't see a need for it to be honest, but I might do a review one day, comparing before / after. But what would be a good machine to test this?

Yes, but you clearly state in the video that you are only caching 128MB of RAM. That means your GA-5AX doesn't have a Revision G chip.

mockingbird wrote:

K6-3+ is just a die shrink of the K6-III

Not quite. The + chips have 3DNow Extended like the Athlon.