feipoa wrote:

Vetz - Yes, we need a PCI Matrox G200. All 177 CPUs have been tested using the same hardware. The exception with the hardware […]
Show full quote

Vetz - Yes, we need a PCI Matrox G200. All 177 CPUs have been tested using the same hardware. The exception with the hardware is that a different motherboard was needed when testing different socket CPUs, e.g. socket 7, socket 8, slot 1, etc. The motherboards employed thus far have been,

1 Biostar MB-8433UUD (UMC 8881F/8886BF) - socket 3
2 Biostar MB-8500TTD (Intel 430TX) - socket 7
3 FIC PA-2013 (VIA Apollo MVP3 - 598AT / 586B) - super socket 7
4 Asus P3V4X (VIA Apollo Pro 133A - 694X / 596B) - slot 1
5 Asus TUSL2-C (Intel 815EP) - socket 370
6 Gigabyte 6VX7-4X (VIA Apollo Pro 133A - 694X / 686A) - socket 370
7 MSI MS-6167 (AMD-750 Irongate - 751 / 756) - slot A
8 Asus P/I-P65UP5 with C-P6ND CPU Card (Intel 440FX) - socket 8
9 ECS P5GX-M (MediaGX + CX5530) - socket 7 GXM
10 Xeon motherboard - slot 2

I had to use two socket 370 boards only becuase the VIA Nehemiah CPUs only worked well on a particular motherboard, whereby the Samuel-Ezra cpus also only worked well on a particular motherboard. In no instance did any one of my socket 370 motherboards work with the full spectrum of VIA socket 370 CPUs.

Luckybob - could you fill in the blanks for the Xeon motherboard you used?

I thought I had finished up the charts, but forgot about the 133 MHz Challenge charts. I guess I have more work to do.

Also, for 177 CPUs on a single chart, it requires a lot of scrolling and can be somewhat difficult to get a good sense of the comparison. I wonder if there is a better way to represent this data, like with a single sheet of text, or similar. Open to suggestions. I've included a sample of this dilemma.

A pdf would be a lot more readable.

sliderider wrote:

I think you need to patch in 3DNow! optimizations for Q2. […]
Show full quote

I think you need to patch in 3DNow! optimizations for Q2.

Here's the press release from AMD.

http://www.amd.com/us/press-releases/Pages/Pr … elease_911.aspx

Try to find the 3DNow! patch then test again. The Winchip2 will probably beat it. The only reason the MP6 does so much better without it is because it actually has 3 MMX units in it so it can execute 3 MMX instructions per cycle.

Unfortunately, I found out about this 3DNow! patch way too late in the game. I would need to retest all AMD 3DNow! enabled CPUs as well as all Winchip2's. That's 48 CPU frequencies. I've had to store away most of my equipment to make space, so it won't be me testing this.

If someone else wants to test for the percent increase with and without the patch, we could make a footnote of it. An AMD K6-2, K6-2+ and K6-3+ at 300 MHz; Winchip2 at 233 MHz; and AMD Athlon Slot A at 500 MHz should be sufficient.

feipoa wrote:

Unfortunately, I found out about this 3DNow! patch way too late in the game. I would need to retest all AMD 3DNow! enabled CPUs as well as all Winchip2's. That's 46 CPU frequencies. I've had to store away most of my equipment to make space, so it won't be me testing this.

If someone else wants to test for the percent increase with and without the patch, we could make a footnote of it. An AMD K6-2, K6-2+ and K6-3+ at 300 MHz, and Winchip2 at 233 MHz should be sufficient.

Should be quite significant in that game. I went from 40FPS to 57 FPS at 1024x768 in Quake 2 on my Compaq (400mhz K6-2 running at 66mhz bus, Voodoo2 SLI) by activating 3DNow!

Hey feipoa,

It would be great if you could release the results in an excel format. It would then make it easier to sort and search for certain CPUs etc.

Really looking forward to reading what you've done!

mwdmeyer wrote:

It would be great if you could release the results in an excel format. It would then make it easier to sort and search for certain CPUs etc.

That will come last, but I do plan on making it available if there is sufficient interest. At current, too many changes are being made to the Excel sheet to warrant distribution. If I share it now, then there could be a lot of these sheets floating around on the internet with false data. It took several months before I stopped correcting the 486 data.

Vetz wrote:

Should be quite significant in that game. I went from 40FPS to 57 FPS at 1024x768 in Quake 2 on my Compaq (400mhz K6-2 running at 66mhz bus, Voodoo2 SLI) by activating 3DNow!

So there is about a 40% speed boost for and AMD K6-2 at 400 MHz with the 3DNow! patch. I wonder if it is 40% for 300 MHz and 500 MHz as well... What about the K6-2+ and K6-3+? Anyone with a system setup want to test these?

Vetz, what's the speed boost in software mode? Is it still 40%?

Edit: I did record some 3DNow!-dedicated benchmark scores.

PassMark - 3DNow!
AMD K6-2-233 : 88.7
Winchip2-233 : 51.6

PassMark - MMX
AMD K6-2-233 : 38.4
Winchip2-233 : 16.9

Winchip seems to be no match for the K6-2 w.r.t 3DNow! and MMX. These results are for both chips running a 66 MHz FSB.

Sure, I can do that. I've got a K6-3+ 400 which I'm running at 5x110MHz (can't do 6x100, but 5x110 is just as good).
I have a Voodoo 3 though not a G200.

F2bnp wrote:

Sure, I can do that. I've got a K6-3+ 400 which I'm running at 5x110MHz (can't do 6x100, but 5x110 is just as good).
I have a Voodoo 3 though not a G200.

Sure! I'm just looking to get a ballpark idea.

Alright, I'll see if I can get them running and I'll be back in a couple of hours with some demo1.dm2 and crusher.dm2 scores 😀.

feipoa wrote:

Based on the feedback I received, I will include extra charts for Quake 1, Quake 2 - Software, Quake 2 - OpenGL, DOOM, 3DMark99Max, and PassMark MMX. These charts, as well as averaged ALU, FPU, Overall charts have now been completed.

For Quake 2 OpenGL, you've used a card so fast you won't be GPU limited ever ? Because else, I'm guessing you get to the roof quickly with a G200, don't you ?

Tests are almost done, took more time than I expected because I had to do some other stuff first. Tests are done at 1024x768 and 640x480 for Software and Default OpenGL with and without 3DNow!.
Would you like me to try other speed configurations as well and post them later on? I could do 4x100 and 6x66 to see how much the FSB factors into this 😁.

And done. Full system specs:
K6-III+ 550 (using a 5x multiplier and 110MHz FSB)
Gigabyte GA-5AX
128MB SDRAM 100
3dfx Voodoo 3 3000 AGP
Windows 98SE

demo1.dm2 results:

640x480:
Software: 22.3 fps
3Dnow! Software: 27.3 fps
OpenGL: 70.9 fps
3Dnow! OpenGL: 80.4 fps

1024x768:
Software: 11.2 fps
3Dnow! Software: 13.7 fps
OpenGL: 67.5 fps
3Dnow! OpenGL: 71.9 fps

Crusher.dm2 results:

640x480:
Software: 16.3 fps
3Dnow! Software: 21.2 fps
OpenGL: 34.2 fps
3Dnow! OpenGL: 42.3 fps

1024x768:
Software: 8.9 fps
3Dnow! Software: 11.3 fps
OpenGL: 33.3 fps
3Dnow! OpenGL: 41.6 fps

So it seems like it's not anywhere near 40%. A lot of things could be the reason for this.
FSB, cache and speed
Vetz is using a 66MHz FSB, no ondie cache and a 400MHz CPU.

Like I said in the post above, I'm thinking of retesting using at 4x100 and 6x66 and see what happens 😀.

That is interesting - you only got about a 20% increase.

sunaiac, I don't really know what is the GPU/CPU-limiting factor when using a Matrox G200 and a Super 7. Maybe someone else can answer this?

feipoa wrote:

That is interesting - you only got about a 20% increase.

sunaiac, I don't really know what is the GPU/CPU-limiting factor when using a Matrox G200 and a Super 7. Maybe someone else can answer this?

I thought you used the G200 on all systems.
If only super7, maybe it's not a limitation then.

That's simple, if your scores stop scaling with more powerfull CPU, and you're confident it should (for example, it does in software mode), then you're reaching a GPU limit.

I'm not entirely sure what conclusion can be drawn from this, but I suppose the AMD K6-3+ from 500-600 MHz could be GPU limited. Take a look at the attached chart. The OpenGL slope of fps/freq. for AMD K6-3+ 500,550,600 MHz is 3x less than that for an AMD K6-2 from 233-300 MHz. The slopes for software mode of both the K6-2 and K6-3+ are the same.

sunaiac, is it the case that the slope should be 0 for a GPU-limited system? I would think there would be just a really small slope. I just don't know how small of a slope constitutes GPU-limited.

EDIT:
For the K6-2 in the 233-300 MHz range, Quake II-GL yields 1.8 fps per every 33 MHz increase.

For the K6-3+ in the 500-600 MHz range, Quake II-GL yields 0.56 fps per every 33 MHz increase.

Have you tried plugging in a faster card with the faster CPU's to see if you get a bigger improvement with 3DNow! turned on?

feipoa wrote:

Unfortunately, I found out about this 3DNow! patch way too late in the game... I've had to store away most of my equipment to make space, so it won't be me testing this.

I can offer my services for the cause 😜. What graphics card would you like me to try out with the K6-3+?

F2bnp wrote:

I can offer my services for the cause 😜. What graphics card would you like me to try out with the K6-3+?

Anything fast enough to remove the bottleneck completely just to see how fast the K6-III+ really is with 3DNow! enabled would do.

I think a GF2/4 MX should suffice. Will give it a try tomorrow 😁

feipoa wrote:

I'm not entirely sure what conclusion can be drawn from this, but I suppose the AMD K6-3+ from 500-600 MHz could be GPU limited. […]
Show full quote

I'm not entirely sure what conclusion can be drawn from this, but I suppose the AMD K6-3+ from 500-600 MHz could be GPU limited. Take a look at the attached chart. The OpenGL slope of fps/freq. for AMD K6-3+ 500,550,600 MHz is 3x less than that for an AMD K6-2 from 233-300 MHz. The slopes for software mode of both the K6-2 and K6-3+ are the same.

sunaiac, is it the case that the slope should be 0 for a GPU-limited system? I would think there would be just a really small slope. I just don't know how small of a slope constitutes GPU-limited.

EDIT:
For the K6-2 in the 233-300 MHz range, Quake II-GL yields 1.8 fps per every 33 MHz increase.

For the K6-3+ in the 500-600 MHz range, Quake II-GL yields 0.56 fps per every 33 MHz increase.

It looks like there's still some margin.
With a true GPU limit, it wouldn't go up anymore (http://media.bestofmicro.com/I/C/318180/origi … %202%202560.png)

Here, it looks like the FSB becomes the limit.
If you want to know the limit of your G200+driver couple, just fit it in a tualatin, athlon 64, or whatever overpowered CPU you have compared to the card 😀