thank you very much for Komponent!!

thank you for your help!!!

I must buy another V4 4500 PCI , and compare with This!

After  i had compared 。i will send you the pictures ！

OK then. Sorry for the delay, Ssstjy.
I guess that I will do my best to try to help our friend from Asia, allthough I do not know if my English is any better than his...
Remember, dudes, a short time ago, Ssstjy managed to get a Voodoo4 PCI that reported 64Mb of video memory, both at boot-screen and in diagnose programs. In an effort to find if the card has 64Mb RAM for real, and not due to a BIOS hack, he saved it's BIOS for further study and then he even flashed the card with the orriginal 1.15PCI 3DFX BIOS. After the flash, the Voodoo4 continued to report the same amount of memory at post-screen and in diagnose softwares too. The study of the saved BIOS revealed no hack; in fact it was just the default that can be found on Voodoo4 PCI cards with 32Mb.
You should allready know by now that Ssstjy's 64Mb Voodoo4 PCI has serials on the memory modules that indicate 8Mb chips for a total of 32Mb RAM and it has no visible traces for the A11 line adress required for 16Mb memory modules to function at full capacity. Also, one very important fact is that there are no visible signs of any rework on the card, it looks genuine, factory default, but yet it reports 64Mb of video RAM.
Realy not your usual experience, right dudes? Well, there is more to it... And now I am even more puzzled.
But first, let me say that the obvious path to try to solve the mistery with this unusual Voodoo4 PCI was to compare it's performance to another regular 32Mb one. If benchmarked against each-other on the same test-system and conditions, the reported 64Mb card would have showed a boost in performance compared to a regular 32Mb Voodoo4 PCI, then this would be the proof that extra-memory realy is present on the card and plus it is working. And for what we knew about 3DFX'es products it would also be a great leap from faith to fact.
So, Ssstjy was lucky enough to be able to get a second Voodoo4 PCI, this time one that reported the usual 32Mb; and the testing could now begin.
Here is a picture with the two cards -they look identical to my eyes- :

http://img374.imageshack.us/img374/3536/dscf0018pn5.jpg

And here we can see a very nice close-up in great detail showing the RAM chips soldered onto both PCB's:

http://img374.imageshack.us/img374/6739/dscf0023zy8.jpg

Before we go into testing results, I ask you for a litle memory training, he-he...
If you dudes still remember it, it happened a long time ago, in a galaxy far, far away... I was able to mod a regular Voodoo4 AGP from 32Mb to 64Mb RAM by soldering 4 extra memory modules into the empty places on the PCB and after flashing the card with a modded BIOS. To confirm that the memory was fully used, at your request I benchmarked the Voodoo4 AGP with 64Mb against a Voodoo4 AGP with orriginal 32Mb and against a Voodoo5 5500 AGP and a Voodoo4 PCI too. Thanks for theyr help into this project are renewed to Mikulaish and HankSemenek and the old topic regarding my greatest success can be found here:

http://www.falconfly-central.de/cgi-bin/yabb/YaBB.pl?board=general;action=displa...

I only mentioned this sort of off-topic story for a reason: the results after testing the Voodoo4 AGP with 64Mb against the Voodoo4 AGP with 32Mb in 3DMark 2001 were:

Voodoo 4 AGP 64 Mb AGP4x : No FSAA : 1576 3D marks

http://img174.imageshack.us/img174/5339/v4agp64mbnoaa8cb.jpg

Voodoo 4 AGP 32 Mb AGP 4x : No FSAA : 1471 3D marks

http://img172.imageshack.us/img172/19/v4agp32mbnoaa6lq.jpg

That stands for arround 100 3DMark points in performance boost for the extra memory on the 64Mb card.

Now, let us go back to Ssstjy's Voodoo4's PCI and the results after testing both cards with the same 3DMark2001:

Voodoo 4 PCI 64 Mb: 1294 3D marks

http://img241.imageshack.us/img241/1667/64mpci2of8.jpg

Voodoo 4 PCI 32 Mb: 1210 3D marks

http://img241.imageshack.us/img241/2227/32mpci2pi5.jpg

That stands for 84 3DMark points in performance boost for the reported 64Mb card against the regular 32Mb one; and this is very close to the results that I got with my modded 64Mb Voodoo4 AGP against the default 32Mb Voodoo4 AGP; allthough the difference is just a little smaller, it can be justified by the fact that Ssstjy's cards were runing at 33Mhz PCI bus that does not scale as well as the AGP 4X 266Mhz.

Now, can you understand why I feel puzzled? There were important clues indicating for a hoax, but all the datas provided by Ssstjy appear to be the confirmation for a real Voodoo4 PCI with 64Mb, fully working.
I know, there is a old say, "seeing is beliving" and I was not able to test the card myself to cast away any doubts, but we have no reason not to take for granted and true the info's received from Ssstjy; and the conclusion that I see after all the proofs has shifted to accepting his Voodoo4 PCI for a real 64Mb card.
It is only my oppinion based on the facts that I am aware of. Dudes, what do you think about the subject?

Were these benchmarks scores obtained from running 3DMark once? Or the average of several?

3DMark doesn't produce the same score everytime, which can explain the difference.

And what processor was used?

Quote:

No, no; normaly that option should not even be plausible for a fully working 64mb card out of chips of 16Mb RAM each of them; you know it as well as as know it too.
Even if the memory modules were of 16Mb but printed with the wrong serials, for the 8Mb variant, that does not change the fact that in order for the RAM chips to function at theyr full capacity it is necesary for the supplementar A11 line adress to be connected to the memory controller -and this one is a part of the GPU. If this is not accomplished, the memory modules would be seen at best as 8Mb chips or not at all.
BUT, on the standard Voodoo4 PCI, the A11 line adress is NOT wired; there is no trace from the M_A11 ball under the GPU to the coresponding pin solder pads. If one would want to use higher density memory modules onto the Voodoo4 PCI PCB, it would be necesary to wire the line with an external conductor.
I had done a study and research to find the connecting points for the missing traces onto a Voodoo4 PCI PCB, in the perspective of a future modding project and I found the following result - see the next picture:
http://img145.imageshack.us/img145/2703/v4pcivsa100modgw6.jpg
God, I would love to put my hands on Ssstjy's Voodoo4 PCI with 64Mb to get to know for sure how was it realised...

There is something wrong with your V4 ipchipang!
I got this with V3 3000 AGP(P3 800MHz, 256Mb):


can you tell what CPU and how much memory you got in that system. CPU affects a lot in that test!

it is the same with you
------------------------------

P3 800 SLOT 1
kingmax PC150 128M
kingmax pc133  128M
MSI BX master

-------------------------------
windows 98se chinese version

3Dmark 2001

Some more news: Our chinese buddy, Ssstjy sent me a couple of new screenshots of testing his 64Mb Voodoo4 PCI against the regular 32Mb Voodoo4.
OK, we allready know that the performance boost offered by the extra memory becomes important when the cars is pushed to it's hardware limits, for example when Anti-Aliasing is applied. The procedure practicaly doubles the amount of video memory used for rendering the scene, because each frame has to be super-sampled; so more RAM onto the video card gives more performance.
This point was confirmed when I tested my Voodoo4 AGP with 64 MB, modded by adding extra memory modules to a regular Voodoo4 32Mb AGP:

Voodoo 4 AGP 64 Mb AGP4x : 2x FSAA : 803 3D marks

http://img173.imageshack.us/img173/3887/v4agp64mb2xaa5uc.jpg

Voodoo 4 AGP 32 Mb AGP 4x : 2x FSAA : 661 3D marks

http://img171.imageshack.us/img171/8343/v4agp32mb2xaa2dv.jpg

That stands for arround 140 3DMark 2001 points in performance boost for Voodoo4 AGP 64Mb against 32Mb.
So, when Anti-Aliasing is applied, the difference becomes more important if the card has more memory installed.

Back to Ssstjy's cards. He tested both Voodoo4 PCI cards -with 64Mb and 32Mb- with 3DMark 2001 when FSAA was activated, so next are the results:

Voodoo 4 PCI 64 Mb : 2x FSAA : 765 3D marks



Voodoo 4 PCI 32 Mb : 2x FSAA : 607 3D marks



That stands for arround 160 3DMark 2001 points for the performance boost of the 64Mb Voodoo4 PCI against the regular 32Mb one. So the extra memory does make an important enough difference...
Am I wrong, or Ssstjy's card is the only confirmed 64Mb Voodoo4 PCI in the whole world?
Thanks for sharing the testing results, buddy.

Now if only would Ssstjy agree to follow my instructions, we could have a good chance to find how does the card works with 16Mb chips with a PCB that looks identical to the default design of the 32Mb Voodoo4 PCI, and with no visible reworks... you know, the A11 adress line's trace problem...