It is currently Sat Oct 21, 2017 6:10 pm

All times are UTC - 7 hours





Post new topic Reply to topic  [ 16 posts ]  Go to page 1, 2  Next
Author Message
PostPosted: Sat Nov 30, 2013 11:36 pm 
Offline

Joined: Wed May 19, 2010 6:12 pm
Posts: 2295
It was brought up in another thread how Nintendo had to sacrifice VRAM access slots, and a more flexible oam for BG3. It makes me wonder how much it would've cost if Nintendo used 2x faster vram.


Top
 Profile  
 
PostPosted: Sun Dec 01, 2013 1:05 am 
Offline
User avatar

Joined: Wed Feb 13, 2008 9:10 am
Posts: 578
Location: Estonia, Rapla city (50 and 60Hz compatible :P)
I looked into my SNESs' and all of them have real SRAM as VRAM, 100ns flavor. Sound RAM is 100-150ns PSRAM.
My google-fu seems to suck, I could only find prices of DRAM...

_________________
http://www.tmeeco.eu


Top
 Profile  
 
PostPosted: Sun Dec 01, 2013 1:07 pm 
Offline

Joined: Sun Sep 19, 2004 11:12 pm
Posts: 19115
Location: NE Indiana, USA (NTSC)
Assuming a 945/44 MHz master clock, and 4 clocks per dot, there are about 186 ns per dot. If you want to take it out to half dots, you'd need 70 ns memory and a lot more pipelining in the tile reader logic.


Top
 Profile  
 
PostPosted: Sun Dec 01, 2013 8:02 pm 
Offline

Joined: Wed May 19, 2010 6:12 pm
Posts: 2295
Would it still require more logic if they cut down on the amount of graphic modes? Or are the two completely unrelated?


Top
 Profile  
 
PostPosted: Sun Dec 01, 2013 8:07 pm 
Offline

Joined: Sun Sep 19, 2004 11:12 pm
Posts: 19115
Location: NE Indiana, USA (NTSC)
You'd still need modes for bit depth (8bpp for everything has disadvantages), offset per tile, affine mapping, etc. But perhaps they'd be set per layer, much as 4/8 bpp mode is set on GBA.


Top
 Profile  
 
PostPosted: Sun Dec 01, 2013 9:09 pm 
Offline

Joined: Wed May 19, 2010 6:12 pm
Posts: 2295
Here's a simple approach:

mode 0: 4 4bpp BGs
mode 1: 2 4bpp BGs and 1 8bpp affine layer

Every non-affine layer will have offset-per-tile available, and disabled backgrounds will be replaced with open DMA access slots. Not only would it have time to fetch sprite patterns during h-blank, but it also would have time to fetch sprite attributes too.

I don't really care that much about 8bpp other than it made mode-7 possible. 4bpp just has a more authentic 16-bit feel to it. Now that I'm at it, why not change it to packed pixel format?


Top
 Profile  
 
PostPosted: Mon Dec 02, 2013 6:15 am 
Offline

Joined: Sun Sep 19, 2004 11:12 pm
Posts: 19115
Location: NE Indiana, USA (NTSC)
Now that's looking very, very GBA, if I do say so myself.


Top
 Profile  
 
PostPosted: Mon Dec 02, 2013 11:42 am 
Offline

Joined: Wed May 19, 2010 6:12 pm
Posts: 2295
Is the price of ram usually proportional to the speed of ram? Or is it exponentially more expensive? Were there any computers around at the time that used 70ns or faster RAM? I know that the Sega Saturn had 28Mhz VRAM accesses, but that was 4 or 5 years later.


Top
 Profile  
 
PostPosted: Mon Dec 02, 2013 1:36 pm 
Online

Joined: Sun Apr 13, 2008 11:12 am
Posts: 6293
Location: Seattle
Hm, 1990 predates DEC's Alphas. Everyone else (POWER, SPARC, i860, 88k, ARM) seemed to be pretty comparable in 1990 (ca. 30MHz).
I do remember some systems using funny RAMs, but for the most part, UNIX workstations were mostly using stock DRAM, usually just put in enough parallel to fetch a 32-bit or 64-bit word all at once. (So, not faster)

For speed, I suspect it's just economies of scale, with a lower sales opportunity and stricter requirements for faster parts.


Top
 Profile  
 
PostPosted: Tue Dec 03, 2013 3:47 pm 
Offline

Joined: Wed May 19, 2010 6:12 pm
Posts: 2295
So there were computers available that used 30Mhz RAM? How much more expensive were they compared to the SNES? We could probably estimate how much extra money a 2x speed VRAM would've cost.


Top
 Profile  
 
PostPosted: Tue Dec 03, 2013 3:53 pm 
Offline

Joined: Sun Sep 19, 2004 11:12 pm
Posts: 19115
Location: NE Indiana, USA (NTSC)
A lot of these faster RAMs had a penalty for accesses to non-consecutive addresses. This "seek time" was common in "fast page mode" DRAMs, which loaded a whole word-line into a small buffer; other accesses within the same word-line had less penalty.


Top
 Profile  
 
PostPosted: Wed Dec 04, 2013 4:18 am 
Offline
User avatar

Joined: Wed Feb 13, 2008 9:10 am
Posts: 578
Location: Estonia, Rapla city (50 and 60Hz compatible :P)
SRAMs used for cache chips don't have any speed penalties, not on the async ones which were ofter 10, 12 or 15ns. Later synchronous SRAMs do have "latency", it takes one cycle to set address, and remaining cycles can be used as reads or writes.
I tried to find any pricing info on cache chips you'd use on a 486 or mayhaps a 36 mobo but could not find any... those are vanilla async SRAM, but much faster than what you would find in a game console, but still may be useful for some ideas.

_________________
http://www.tmeeco.eu


Top
 Profile  
 
PostPosted: Fri Mar 07, 2014 8:23 pm 
Offline

Joined: Wed May 19, 2010 6:12 pm
Posts: 2295
I have a another question about memory latency. Would 6.136Mhz accessing with 150ns ROM possible, or would that need 120ns ROM? I'm asking because it would've made sense to design a system with every chip clocked to the same frequency, including the dot clock, and 6.136Mhz would've gave you square pixels.


Top
 Profile  
 
PostPosted: Fri Mar 07, 2014 9:37 pm 
Offline

Joined: Sun Sep 19, 2004 11:12 pm
Posts: 19115
Location: NE Indiana, USA (NTSC)
You'd still need a master clock of 945/44 = 21.48 MHz to generate the 3.58 MHz color subcarrier. You'd divide it by 7 instead of 6 or 8 to get the 3.07 MHz CPU clock, but you'd need to divide it by 1.75 to get the 12.28 MHz half-pixel clock used by hires mode. And to avoid unsightly side borders, you'd need to redesign nametable layouts around a 288px wide display plane, which means nametables would have to be 64x32 cells.

Arcade and other RGB systems have a lot more freedom with their dot clock rates because they aren't tied to crystals that are multiples of 3.58 MHz.


Top
 Profile  
 
PostPosted: Fri Mar 07, 2014 11:58 pm 
Offline

Joined: Wed May 19, 2010 6:12 pm
Posts: 2295
I know how the clock frequency would be derived, I was asking if 150ns ROMs were fast enough for a 6.136Mhz clock. Theoretically it should because 1000/150=6.667Mhz.


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 16 posts ]  Go to page 1, 2  Next

All times are UTC - 7 hours


Who is online

Users browsing this forum: No registered users and 5 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB® Forum Software © phpBB Group