nVidia 7800 and 7900 series

[K404]

This applies to more than just these cards... but this has made no sense for a long time.

 

7800GT: not split by memory amount

7800GTX and GTX512MB: split

7900GS: split by memory amount

7900GT: split by memory amount (coincidentally, 7900GT 512MB can show up as a 7900GTO....)

7950GT: Not split by memory amount.

 

I don't mind/care which way things go....but please can there be consistency here?

 

Thankyou

[ZFeSS]

1 hour ago, K404 said:

7800GT: not split by memory amount

7800GTX and GTX512MB: split

7800GTX 512 and 7800GTX are totally different and produced as different models. Additional 120Mhz on core, different PCB and 136-pin 1.1ns memory instead 144-pin 1.6ns on 256MB card. Also all 7800GTX are reference cards. AFAIK 7800GT 512 is Gainward only. That's why I see the logic here.

 

 

[K404]

They are not totally different. Based on that logic, there should be new categories for Matrix/Lightning/HOF/non-ref cards etc.

 

 

 

[ZFeSS]

 

10 hours ago, K404 said:

They are not totally different.

Different PCB, memory packaging and speed, clocks, model name, memory amount. Is it not enough? They shares same GPU only, everything else is different. Nvidia says that 7800GTX and 7800GTX 512 are different models also.

[K404]

But your reasoning makes no sense and is "dangerous" for hardware classification. Would you use those reasons for any other hardware classification?

Different PCB: Ref Vs OC model

clocks: Ref Vs factory OC

Memory amount: makes no difference to the benchmarks that run on a 7800

 

memory packaging is a little bit interesting, but HWBot do not differentiate on package. Only memory type. They both have GDDR3

 

Based on Device ID, nVidia see both 0090 and 0091 as 7800GTX: https://us.download.nvidia.com/XFree86/Linux-x86/346.47/README/supportedchips.html (7800 and 7900 series are quite far down)

I generally do not like Device ID as a means of seperation, though. It can create problems with rebranded cards/cores.

 

 

My reasoning for possible class combination: Two cards named 7800GTX made with G70-A1 silicon. 24 Pixel, 8 vertex, 16-ROP, 256-bit bus, connecting GDDR3

 

One good (?) reason to not separate the classes: The memory clocking ability of the GTX512 gives it a clear lead over the 256MB version and the 512MB version appears to be hard to find, so leaving them split keeps the 7800GTX category more open to new good scores.

Edited June 28, 2018 by K404

[ZFeSS]

7 hours ago, K404 said:

But your reasoning makes no sense and is "dangerous" for hardware classification. Would you use those reasons for any other hardware classification?

Different PCB: Ref Vs OC model

clocks: Ref Vs factory OC 

Memory amount: makes no difference to the benchmarks that run on a 7800

Your reasoning is even worse. In that case we should merge all RX400 and RX500 series, R9 290 and 390, same for 7000 and R9/R7 series, GTX770 and 680, GTX460 and 560, and more than 50 other models. Same for CPU. Should we merge i7-8700K and 8086K? Maybe all AMD FX series?

Why we not merge cards with same PCB, memory speedbin, memory amount, but should merge cards which differs much more? Also all 7800GTX and 7800GTX 512 are reference cards.

First reason for new category was different model name - http://www.nvidia.com/page/geforce_7800.html Nvidia says that 7800GTX and 7800GTX 512 are different cards.

[K404]

What? What part of my reasoning suggests that?  I don't know enough about AMD cards to comment on that, but there is ONE crossover in core spec between all the GTX460 and GTX560 cards (although different silicon...which completely zeros your comment) .... out of 5 GTX 460 classes and 6...7? GTX560 classes. They have different NAMES and SPECIFICATIONS

 

8700K and 8086K have different names. You are picking bad examples to support your viewpoint.

Is there a language barrier here?

 

 

 

Is nVidia marketing (which sometimes mentions the 512MB separately and sometimes doesn't) important to how HWB classes a card? I don't think it is.

Is it the only time that nVidia have officially made two ref cards with different memory amounts (no other changes?)

Is that important when all we care about are synthetic benchmark results and want (I hope) a standardised consistent platform for hardware?

 

 

 

 

Does anyone else have an opinion on this entire subject? Should there be general rules for adding new classes based on memory amount or should they be dealt with on an individual basis? How many benchmarks should be affected by RAM volume before there should be consideration of a new category?

Edited June 28, 2018 by K404

[ZFeSS]

1 hour ago, K404 said:

8700K and 8086K have different names

7800GTX and 7800GTX 512 have different names too. That's what I try to explain to you. It's first and main reason to divide them. I agreed with you, that 7900 series cards with 256 and 512Mb can be merged, but not 7800GTX.

[K404]

In Rivatuner and GPU-Z they don't, though...

[Strunkenbold]

Well I can just tell you that we started merging cards with different memory sizes last year. Now Christian Ney is gone, Massman is gone, Frederik has no time. I dont have permission.

I link your thread to the internal "merging thread" but I dont think something will happen. If Frederik has ironed out the most serious website bugs sometime, if at all he _might_ feel comfortable doing this. Maybe somewhere next year.