Kann ich morgen nachmessen, da bastle ich eh an einem Board weiter und eine völlig defekte 3000er liegt quasi daneben. Ich tippe auf einen 100nF.
Beiträge von Löschzwerg
-
-
http://www.3dfxzone.it/dir/3dfx/index.php
Da kann man auch auf die Suche gehen
-
Ob billiger und andere Größen (also kleiner wie auch größer) hängt auch von der jeweiligen Generation der Fertigung bzw. dem "Mother Glass" und der Schnittausbeute ab.
Hier kann man sich ein gutes Bild zum Thema verschaffen:
https://wikimovel.com/index.php/Mult…s_display_sizes
An den Beispielen für 8.5 (aktuell) sieht man auch warum 77" immer noch so teuer ist. Die Lage wird sich erst mit Gen 10.5 bessern.
-
Ein wichtiger, wenn nicht sogar der wichtigste Punkt ist der FSB. Hie einfach bei 66 oder 100MHz bleiben ist die halbe Miete.
-
-
Danke für die Infos Geoman Falls noch nicht probiert, bitte mit einer anderen Karte ins Bios wechseln und dort die AGP Geschwindigkeit auf 1x setzen und Sideband Adressing ausschalten. Tut sich dann etwas mit der SnowWhite? Du kannst auch mal schauen was passiert wenn du eine PCI Grafikkarte zusätzlich steckst, ob dann diese als Master läuft usw. (hierzu gibt es evtl. ebenfalls Einstellungen im Bios und es kommt auch auf den Slot an).
-
-
-
i guess reversing the ic chip is not like having verilog code or other fancy stuff. i presume it is like having all the layers scans in proper resolution appropriate to making new lithography.
But without knowing which function block is doing what and how the timings correlate to each parts within the chip, you are having a hard time. It would be so much better to have a higher level starting point where you can build upon and have the proper chance to actually test some things (maybe even on an FPGA basis).
Otherwise I just see a reproduction possible, if at all (since you need exact the same manufacturing process and tooling that was used).
-
DerBiber You get nice theoretical values for very specific tasks. Those aren't necessarily relevant in games, if at all. I don't disagree. They are however relevant to see every bit of system performance, no matter practical useful or not. Why? Why not! Is there anything else to see or test what we haven't seen already? UT @ 640x480 and a highend CPU to see the benefit? Sure, why not
And to be honest, we all have seen so many game tests and also 3DMark values aside... If someone runs 3DMark and gets a nearly identical score, one can probably estimate that certain game will run fine as seen in the test/review. That's the overall benefit of a synth benchmark.
-
What do you guys understand as "blue prints"? Do you mean the register-transfer level or the finished GDS data?
-
PCB Layout never was the problem, we just were too lazy. Chip design on the other hand is a totally different beast. Generate Verilog code from a die-shot... Nope, that's not how it's gonna work
Regarding "should not be too hard" part: It's not simply done by just increasing frequency. The I/O probably needs to be tied to base clock and constantly controlled by a phase lock loop. Otherwise the bus system would run out of sync. Without proper EDA and hardware for simulation/verification I don't see this happen.
-
Einfach mal nach "Computer Dust Cover" oder "UV Schutzhaube" suchen, da gibt es schon ziemlich viel. Blödes erstes Beispiel: https://www.ebay.de/itm/331015543771
-
Well that's totally missing the point about the initial question from Raff. The whole idea about a "modern" Voodoo system is getting more performance in scenarios where a beefy CPU is of benefit. And this if when a tool with specific tests like 3DMark comes in handy. Simple as that.
-
You need the HDL code of the chip, something like Verilog for example. I don't see any chance to get this data (if it still exists at all).
Oh I am very sure it exists, it's just getting it is the biggest question, if it didn't exist, well none of the VSA based cards would of been a thing.
HDL code existed, that is 100% safe, but the big question is if they still do. I don't think that any of the known ex-3dfx engineers do have those data on a backup medium, because what would you do with it anyway if you have no massive chip-simulator at home. That's not something like driver code you can update, compile and run everywhere.
Maybe data exists on a long-term archive at Nvidia (try getting it from them ;)), but even if we had access to it, setting up a FPGA solution is probably the best a hobby project could do. This HDL code needs to be translated to modern tools for chip manufacturing, then get them produced and assembled... Bring the chip to life, write VBIOS and so on.
It's ok if you all like to dream and share ideas, but can we please be somewhat realistic?
-
You need the HDL code of the chip, something like Verilog for example. I don't see any chance to get this data (if it still exists at all).
-
The key values of 3DMark00/01 lie within the feature tests (fillrate, pixel/vertex...), that's where you can see limits/bottlenecks. If you know how to handle these values 3DMark is a very nice and convenient tool. It easily shows if you are in the ballpark when it comes to performance.
-
Backfire Ich habe noch Entlötzlitzen und Flussmittel in meiner Liste oben ergänzt. Hat sich gerade ergeben als ich in meiner Werkstatt die Bestände kontrolliert habe
-
-
Löschzwerg -> Ahh ok... nee ich mache das für Dich mal eben schnell - wenn es Deine Zeit nicht hergibt. Alles gut.
Das wäre dann
470µF: 74-293D477X96R3E2TE3 ( 2x )
100µF: 74-293D107X0016D2TE3 ( 2x )68µF: 667-10TPC68M ( 7x )
Bitte Prüfen
- Backfire -
Danke
Bitte bei den 68µF 10x und dann noch 40x 10µF 74-293D106X0016B2TE3
Liste:
2x - 470µF: 74-293D477X96R3E2TE3
2x - 100µF: 74-293D107X0016D2TE3
10x - 68µF: 667-10TPC68M
40x - 10µF: 74-293D106X0016B2TE3
2x - Chemtronics 40-3-10: 5168-40-3-10
2x - Chemtronics 80-3-10: 5168-80-3-10
2x - MG Chemicals 8341-10ML: 590-8341-10ML