Dumb question -- does adding VRAM to a Nubus video card or Mac onboard video make video rendering faster? Or does it just give you more colors and larger screen support?
VRAM - more speed, or just more colors?
- Thread starter misterg33
- Start date
Solution
I have an 8*24 GC in my MacIvory system and took some benchmarks a while back.
oldvcr.blogspot.com
Refurb weekend: the Symbolics MacIvory Lisp machine I have hated
Every collector has that machine, the machine they sunk so much time and, often, money into that they would have defenestrated it years ago...
oldvcr.blogspot.com
I have not done the testing in practice.
In theory, it depends on the card. While one would expect it to just give one more capacity and therefore more colors and screen size/resolution, there is a way it could improve performance.
If the added memory actually widens the data bus (goes from 32 bits wide to 64 bits wide, e.g.) there are some fancy muxing/demuxing tricks that video cards can do with memory to gain performance.
I don't know whether any of them behave that way in practice. It makes the design more complicated, and for un-upgraded cards, the mux/demux circuitry is just sitting there (expense) doing nothing unless the upgrade is installed.
In theory, it depends on the card. While one would expect it to just give one more capacity and therefore more colors and screen size/resolution, there is a way it could improve performance.
If the added memory actually widens the data bus (goes from 32 bits wide to 64 bits wide, e.g.) there are some fancy muxing/demuxing tricks that video cards can do with memory to gain performance.
I don't know whether any of them behave that way in practice. It makes the design more complicated, and for un-upgraded cards, the mux/demux circuitry is just sitting there (expense) doing nothing unless the upgrade is installed.
No increase in rendering/blitting performance that I know of. Just increased resolution and color depth.
More pixels/greater workspace = better user efficiency in task performance = higher speed.
So yes, indirectly, I'd say adding VRAM increases workflow efficiency . . . or it just looks prettier.
So yes, indirectly, I'd say adding VRAM increases workflow efficiency . . . or it just looks prettier.
Normally, adding more VRAM will enable higher resolutions and/or bit depths. But in my experience with vintage Macs, a switch to higher bit depths like from B&W to 8-bit (256 colors), or from 8-bit to 16 or 24, tends to reduce machine performance. The same is true for the switch to higher resolutions.
IF you add more VRAM but make no changes to bit depth or resolution, I’ve not seen any noticeable performance boost myself.
Also note that even the fastest NuBus video card is not a speedy as Quadra onboard video.
IF you add more VRAM but make no changes to bit depth or resolution, I’ve not seen any noticeable performance boost myself.
Also note that even the fastest NuBus video card is not a speedy as Quadra onboard video.
AFAIK, the only NuBus Card to ever outperform onboard video (which is the rough equivalent of PDS video) would be the bit bangin' monster that @Melkhior developed.
NuBus Cards, VRAM, high resolutions and color depths ARE NOT for greater speed out of a computer, ever. High resolution and color depth make the system work harder to fill the larger, deeper color frame buffers. But there are handsome tradeoffs involved.
The gain in performance would be at the user level's I/O, interacting with hardware and software. The user would be the ultimate limiting factor when it comes to performance of a SYSTEM outside of things like rendering when that became a thing. Performance improvement is seen in the user working faster/more efficiently within many applications, where larger amounts of information on screen interacts with the Mk.1 eyeball. Some such would include:
- Graphics programs of course, less time scrolling across a large document speeds things up very nicely, color depth needed as well
- Desktop Publishing, where the performance hit of filling a full or two page display's frame buffer pays off tremendously, even 1bit on 128K
- Spreadsheets as well are much faster to use when more cells are visible, especially if two related cells can't be seen without scrolling.
- Database - same deal, putting a larger data set onscreen pays big dividends
Can't think of incremental or even great improvements or deficits in the inner workings of a SYSTEM. The rubber meets the road at the user level.
p.s. When folks talk about accelerated graphics cards, they fail to realize that it's QuickDraw that's accelerated. Users of the apps above throw acceleration right out the window when they press the spacebar and use the hand too to scroll. Page navigation keys and manipulating the scroll bars themselves are accelerated in QuickDraw. The hand tool is not.
NuBus Cards, VRAM, high resolutions and color depths ARE NOT for greater speed out of a computer, ever. High resolution and color depth make the system work harder to fill the larger, deeper color frame buffers. But there are handsome tradeoffs involved.
The gain in performance would be at the user level's I/O, interacting with hardware and software. The user would be the ultimate limiting factor when it comes to performance of a SYSTEM outside of things like rendering when that became a thing. Performance improvement is seen in the user working faster/more efficiently within many applications, where larger amounts of information on screen interacts with the Mk.1 eyeball. Some such would include:
- Graphics programs of course, less time scrolling across a large document speeds things up very nicely, color depth needed as well
- Desktop Publishing, where the performance hit of filling a full or two page display's frame buffer pays off tremendously, even 1bit on 128K
- Spreadsheets as well are much faster to use when more cells are visible, especially if two related cells can't be seen without scrolling.
- Database - same deal, putting a larger data set onscreen pays big dividends
Can't think of incremental or even great improvements or deficits in the inner workings of a SYSTEM. The rubber meets the road at the user level.
p.s. When folks talk about accelerated graphics cards, they fail to realize that it's QuickDraw that's accelerated. Users of the apps above throw acceleration right out the window when they press the spacebar and use the hand too to scroll. Page navigation keys and manipulating the scroll bars themselves are accelerated in QuickDraw. The hand tool is not.
Last edited:
NuBusFPGA: HDMI on NuBus Macs
Hello, I've mentioned my crazy project of interfacing a FPGA (board) with a NuBus Mac in another thread. I'm opening a new dedicated thread as there is a significant update now: This show the naked PCB on the left, the populated version without the required daughterboard in the middle, and the...
68kmla.org
As everyone has mentioned so far, adding VRAM generally increases the maximum number of colors and resolution only. However, on accelerated NuBus/PCI/AGP cards, adding VRAM can speed up graphics. For example, the Apple 8*24 GC card can vastly speed up offscreen graphics with additional VRAM.
"When a drawing operation that involves a GWorld occurs, GC QuickDraw immediately caches the complete GWorld structure in the card's memory if the structure has not yet been cached and if sufficient memory is available. (The card's optional DRAM kit is an important addition for applications that work with large GWorlds.) When CopyBits is called to display the results, the transfer of pixels to the screen driven by the 8*24 GC card is therefore really fast because there is no NuBus transfer. Even displaying the image into other monitors benefits, especially when the other cards can accept block transfers. Drawing operations to and from GWorlds can be executed in parallel. This is not the case when drawing to or from old-style offscreens"
"When a drawing operation that involves a GWorld occurs, GC QuickDraw immediately caches the complete GWorld structure in the card's memory if the structure has not yet been cached and if sufficient memory is available. (The card's optional DRAM kit is an important addition for applications that work with large GWorlds.) When CopyBits is called to display the results, the transfer of pixels to the screen driven by the 8*24 GC card is therefore really fast because there is no NuBus transfer. Even displaying the image into other monitors benefits, especially when the other cards can accept block transfers. Drawing operations to and from GWorlds can be executed in parallel. This is not the case when drawing to or from old-style offscreens"
I’ve long wanted to obtain one of those cards to do my own testing. Alas, they are harder to find than a vintage Xceed grayscale kit (and I have one of those, so I know). But if I recall correctly, at lower bit depths, that Apple card beats out any other vintage NuBus card, including my Radius Thunder 24/GT, which is about the fastest you can get on NuBus for higher bit depths.
Speaking of performance comparisons, the following site, which shockingly went offline a couple years ago for reasons unknown but thankfully was backed up on Wayback Machine, is one of THE best performance comparisons out there for newest video cards that I’ve come across:
I have an 8*24 GC in my MacIvory system and took some benchmarks a while back.
oldvcr.blogspot.com
Refurb weekend: the Symbolics MacIvory Lisp machine I have hated
Every collector has that machine, the machine they sunk so much time and, often, money into that they would have defenestrated it years ago...
oldvcr.blogspot.com
@ClassicHasClass
I used Gemini to glean relevant info from your URL about the Apple Display Card 8•24 GC....
I used Gemini to glean relevant info from your URL about the Apple Display Card 8•24 GC....
1. The Hardware Under the Hood
- A CPU on a GPU: The 8•24 GC isn't just a dumb frame buffer. It features its own onboard RISC processor—an AMD Am29000 running at 30MHz (with 64K of SRAM cache).
- VRAM: It packs 2MB of VRAM by default, but it's expandable up to 10MB using 64-pin IIfx-style SIMMs. (Note: The extra 8MB doesn't unlock higher resolutions or deeper colors; it just acts as additional GWorld space for offscreen rendering).
- How it Works: A control panel (CDEV) loads an Am29K-optimized "GC QuickDraw" kernel onto the card. It then intercepts standard Mac QuickDraw calls and offloads them to the AMD chip, which draws directly to the card's VRAM rather than bottlenecking the Mac's CPU over the NuBus architecture. It will even accelerate standard 8•24 and 4•8 Apple cards if they are installed alongside it!
2. Snooper Benchmark Results
"Baseline" below means a stock Macintosh IIci:- Unaccelerated (GC Off): 65% of Baseline Without the AMD chip engaged, the card runs at about 65% the speed of the IIci's onboard video. Keep in mind, this dip is largely because the NuBus card is pushing a much larger screen resolution than the IIci's onboard video can handle.
- Accelerated (GC On, Stock CPU): 169% of Baseline When the Am29000 chip is activated, video performance jumps to 169% of the IIci baseline. That is roughly 2.5x faster than the card unaccelerated. While this falls massively short of Apple’s bombastic 1990 marketing claims ("5x to 30x faster"), it provides a remarkably snappy, noticeable speedup for a 68030 Mac.
- Accelerated + CPU Upgrade (GC On + PowerCache L2): 253% of Baseline When paired with a DayStar PowerCache CPU accelerator, video performance skyrockets. The beefed-up Mac CPU can muscle through the specific QuickDraw operations that the Am29000 chip can't effectively offload, yielding a blistering 2.5x speed increase over the stock IIci video.
3. The "Gotchas" (Crucial for NuBus Comparisons)
- The 24-Bit RAM Bottleneck: The GC control panel absolutely refuses to run in 32-bit addressing mode. This means your Mac is strictly tethered to 24-bit mode, severely limiting your usable system RAM to just 8MB. If you have a machine stuffed with 64MB of RAM, the vast majority of it will be completely wasted.
- OS Limitations: You really need to run System 7.1. While the card can technically run accelerated under System 7.5 or 7.6, you'll likely encounter graphical glitches, particularly with pull-down menus tearing or artifacting.
- No '040 Macs: Got a Quadra or another 68040-based Mac? Forget it. The 8•24 GC is notoriously incompatible with the '040 architecture.
For the LC models in particular:
- The LC and LC II (and Color Classic) come with 256K stock. This is good for 256 colors on a 12" RGB (or the CC's internal display) at a 512 x 384 resolution. Note that this is taller than the B&W compacts, which are at 512 x 342. (Proof of this is the top and bottom windowing of a program like Shufflepuck Cafe). On a 13" monitor like the HiRes RGB or Macintosh Color Display, this will be good for 16 colors (at 640 x 480). These computers have a single VRAM socket and can be upgraded to 512K, which ups the 512 x 384 to Thousands and 640 x 480 to 256 colors.
- The LC III is worth mentioning since it comes with 512K, giving it the same capabilities as the upgraded LC/LCII/CC, but it has a unique feature described well in earlier copies of the Pogue/Schorr Mac Secrets book: there's an option nestled in the Monitors control panel that can have these machines run in 640 x 400 mode at Thousands on a 13" monitor. That's the same resolution as some of the PowerBooks had, but it does look kind of strange with the big blank area on the monitor.
- The LC 475 has two VRAM sockets and can go up to 1MB total, which ups the capabilities by another notch on all monitor types.
As for speed...I have nine LCs in my lab plus the one on my desk at home. All are the original LC or the LC II and all have the 12" monitor now that I moved my 13" monitor to a IIci. A few of them have the extra VRAM (512K) and none of those machines seem any faster than their 256K counterparts. I actually keep all of them at 256 colors though for compatibility reasons (unless I'm running Kid Pix and we have children who lack experience with the mouse, as the larger color swatches are easier to click).
If you want more speed on an LC, lower color depth will work (they fly in black and white mode), but an even bigger speed boost comes from running System 6.
- The LC and LC II (and Color Classic) come with 256K stock. This is good for 256 colors on a 12" RGB (or the CC's internal display) at a 512 x 384 resolution. Note that this is taller than the B&W compacts, which are at 512 x 342. (Proof of this is the top and bottom windowing of a program like Shufflepuck Cafe). On a 13" monitor like the HiRes RGB or Macintosh Color Display, this will be good for 16 colors (at 640 x 480). These computers have a single VRAM socket and can be upgraded to 512K, which ups the 512 x 384 to Thousands and 640 x 480 to 256 colors.
- The LC III is worth mentioning since it comes with 512K, giving it the same capabilities as the upgraded LC/LCII/CC, but it has a unique feature described well in earlier copies of the Pogue/Schorr Mac Secrets book: there's an option nestled in the Monitors control panel that can have these machines run in 640 x 400 mode at Thousands on a 13" monitor. That's the same resolution as some of the PowerBooks had, but it does look kind of strange with the big blank area on the monitor.
- The LC 475 has two VRAM sockets and can go up to 1MB total, which ups the capabilities by another notch on all monitor types.
As for speed...I have nine LCs in my lab plus the one on my desk at home. All are the original LC or the LC II and all have the 12" monitor now that I moved my 13" monitor to a IIci. A few of them have the extra VRAM (512K) and none of those machines seem any faster than their 256K counterparts. I actually keep all of them at 256 colors though for compatibility reasons (unless I'm running Kid Pix and we have children who lack experience with the mouse, as the larger color swatches are easier to click).
If you want more speed on an LC, lower color depth will work (they fly in black and white mode), but an even bigger speed boost comes from running System 6.
@trag already mentioned bus width; with more formalism (because it's what I do
)Adding VRAM can...
- increase capacity (virtually always, unless using very unusual technique such as shadowing)
- increase bandwidth (almost never, typically by bus widening as mentioned by @trag)
- decrease latency (almost never, typically by additional interleaving)
Increasing bandwidth will help with volume-intensive operations, such as blitting (copying data from one place to another, for e.g. scrolling things). As @trag mentioned, this is typically done by bus widening, though interleaving can also help in much less significant way. Basically the additional chips are accessed "in parallel" to the already present ones. This is why you want to populate evenly the memory of machines with multiple memory "channels" (i.e. why you wanted set of three DIMMs on the original Intel i7).
Decreasing latency will help with most operations, but is mostly visible when doing large number of operations on a small volume of data each. It's hard to decrease latency by adding similar RAM to what already exists, but interleaving can help to some extent by "filling" wait cycles of multi-cycles accesses using additional bank(s) of memory. Of course adding memory to help latency with different memory than what is already installed is just adding (a) cache(s).