Why don't manufacturers still do this?!

RogueTrip

2[H]4U
Joined
Feb 22, 2003
Messages
2,869
While at work we have been tossing out some old electronics. While creating a bin to recycle some of the computer parts I came across some older videocards like the Trident 1MB with 1MB upgradeable add-on abilities. Even saved a Sound Blaster 32 AWE that has upgradable memory slots.

Why don't manufacturers make add-on memory modules anymore? This could help cut initial cost and give something to upgrade. I see AMD/ATI likes to have GDDR5 memory and then make two models usually like 512MB, then 1GB. They could just sell a 512MB videocard with a upgradable slot for another 512MB. It looks like they have been keeping GDDR 3/5 for most models currently, so I don't see why this couldn't be picked up accross multiple model lines. Like say a 4870 has 512MB, you bought another 512mb chip to make it 1GB, now you pick up a 5870/50 512mb and just move your 512MB over to that. Help save a little amount of money in the process?!

I'll end my rant here, just alway thought it was nice to be given choices :rolleyes:
 
Choices are nice, but no one bothered to take advantage of the upgradability. You could argue that only audio techs would bother using this feature on a card like the AWE32, but then again, no serious audio tech was going to buy an AWE32 when far superior pro-audio solutions existed.
 
Yes, I also remember the days of the peripheral cards with the empty add on memory slots. Vesa Local Bus video cards and such. It's probably just not cost effective anymore to provide that option, with electronics being so cheap.
 
Ahhh those were the days. Man I loved the excitement of the 90's with computer tech.
 
Heat probably. Memory on video cards now a days needs to be agressively cooled and is usually attached directly to the heatsink. It makes upgrading a problem, you'd need to remove and then re-add the heatsink etc...

It's just easier to build seperate revisions of the card like the 4870 512mb and 1gb variants.
 
Adding in a memory slot adds cost, takes away space for installed memory, and adds more support issues. Then there's also the cost of putting GDDR3/5 (or whatever) onto actual memory modules that you can put in the slot.

That's adding a lot of cost and complexity for something that most people would never use. Sure, you and I would buy the cheap version then add memory when games started needing more, but most people would just buy whichever one, then buy a replacement one in five years when their PC is too slow. It's a lot more efficient to just make one PCB and leave half the RAM off on the cheap model.


Wait, we're still not in the 90's?..............................Man I'm getting old.

We're actually almost out of the Oughts and into the Teens now. I'm old too...
 
Using extra connections and lengthening very fast bus signal paths causes many problems.
ie longer traces give extra capacitance, induction and resistance causing crosstalk (interference), different signal reflection problems and signal loss amongst other things.
Extra connections introduce impedance mismatches which reflect the signal back, reducing the amplitude of the signal received and reducing the signal to noise ratio, amongst other things.
Bad connections can occur as well, resulting in higher faulty returns, some unnecessary.

I imagine its not practical to use memory slots on fast gfx cards unless slower speeds are used (with current technology).
Couple that with the need for decent cooling on the added memory chips and it becomes a logistical nightmare fitting everything in the space of a couple of PCI-E slot.
Development, design and cost of extra parts will raise costs too.

If they didnt make cards in different memory sizes I could see the need for this.
imo its not a problem that needs solving, the costs are prohibitive, performance will suffer and it will be less reliable.
 
I also believe the memory configurations are limited based on the memory bus width which may not offer decent upgrade paths.
 
I also believe the memory configurations are limited based on the memory bus width which may not offer decent upgrade paths.

Memory configurations are limited to a combination of bus width, and number of devices per-controller. You add memory (each device is 32-bits wide) until you hit your bus width limit. After that, you can add more memory, but since you're stacking more memory on each controller, performance goes down (just like your desktop with more than two sticks of ram).

The reason devices in the 1990s had memory upgrade slots was because they were cutting-edge (wavetable synth from the AWE32, 24-bit color and 2D acceleration from the graphics cards), and thus didn't sell enough units to justify several SKUs. But since all these fancy features CRAVED ram (seriously, have you HEARD the AEW32's default 512kK wavetable ROM sample set? it sucks!), there had to be an upgrade path for those who could afford it. Thus it was cheaper to get everyone in the door at a reasonable price, and then sell upgrades to get any real usability out of the devices.

Of course, if you had any balls at all back in those days, you made the upgrade slots proprietary to milk every last dollar. I guarantee you RogueTrip, if they made their memory upgradable, you would be buying every last stick direct from ATI or Nvidia for superinflated prices.
 
The AWE32 was a piece of junk.... I have a few of them laying around and they crackle/hiss, and generally just sound horrible.

Back in the day I found a used Opti930 based cards with a 4MB hardware wavetable built on. That card was the bomb.. I still have it in my parts bin. Worked great on just about every DOS based game I threw at it. Best card I ever found as far as SB-Pro emulation went.

But even that card is nothing compared to the AWE64 Gold! I currently have in my retro gaming rig... Although the Opti based card is a lot easier to setup to work with everything, the AWE64 card sounds a lot better.

You guys remember having motherboards that had the L2 cache that was upgradeable? Take out the old chips, put higher capacity ones in, and change the jumpers for the new configuration. I did that to 2 or 3 boards I had back in the day.
 
By the time you'd need to upgrade cards like that with more memory, it would probably be more cost effective to buy another card all together. Next gen tech, better performance etc.
 
While this was just wishful thinking and the fact that many of you have brought valid points to the discussion, I'm sure a few of them could be answered or ironed out to be workable.

*For cooling of the ram, it could have ramsinks and be installed parallel to the videocard along the cooling track to take care of heat issues.

*Cost is just name of the game and I'm sure nVidia/ATi would make the ram proprietary to there brand anyways even though they might share same specs of chips internally :rolleyes:

*Distance would still be short on the board so trace paths would still be decently short to the bus and if the card is shielded then that drops some of the other issues, It could still have issues but enough engineering might iron out most of it.


For the most part I'd seem them making issues of Latencies/Frequency/voltages since I do notice that change with different revisions of models.

Although the soundcard is more offtopic, I didn't realize what I saved at first was a AWE32. Just thought it might be decent for my retro rig :) I also saved a Diamond audio card, and 3 or so Sound Blaster Live! cards of differing models.

Among some of the things I saved (got for free):
ATi Rage 128 AGP
Rockwell Modem (56K?! Dunno why I grabbed the modems, oh yeah, FREE!)
Real 3D AGP P/N SFA-150-A (Model number 44-0016100)
Lucent Modem (wonder is this is the trashy win modem?)
Sound Blaster Live! CT4830
Sound Blaster Live! CT4760
Sound Blaster Live! CT4780
D-Link DFE-530+ Rev D2 (think this is a 10/100 NIC PCI)
Viper V770 32MB AGP T3 (dunno if this is OEM, it doesn't have the AGP 2x/4x switch)
3-Com EtherLink XL PCI (10MBps probably)
Diamond DT-588 soundcard
Kingston KNE111TX (10/100 NIC PCI)

Maybe another good question about Videocards, would it be worth swapping out my V770 32MB 4x AGP for the Rage 128? Never really played with ATI.
 
qft!


also no keep the v770, in that time frame nvidia>ati.......*dons flame retardent suit*

Well, the ATI has mfr code of PN 109-66500-00 which seems to just be a 32MB AGP VIDEO CARD RAGE 128, This seems to be in competition of Riva TNT (or my older Viper 550 16MB). So your statement seems true that V770 is better.

Also looked up that Real 3D Starfighter AGP, looks like its just some Intel740 chipset also not worthy of resurrecting, oh well. It was worth a try.
 
Among some of the things I saved (got for free):
ATi Rage 128 AGP
Rockwell Modem (56K?! Dunno why I grabbed the modems, oh yeah, FREE!)
Real 3D AGP P/N SFA-150-A (Model number 44-0016100)
Lucent Modem (wonder is this is the trashy win modem?)
Sound Blaster Live! CT4830
Sound Blaster Live! CT4760
Sound Blaster Live! CT4780
D-Link DFE-530+ Rev D2 (think this is a 10/100 NIC PCI)
Viper V770 32MB AGP T3 (dunno if this is OEM, it doesn't have the AGP 2x/4x switch)
3-Com EtherLink XL PCI (10MBps probably)
Diamond DT-588 soundcard
Kingston KNE111TX (10/100 NIC PCI)
Wow, all hardware not worth the power it consumes.
 
Lots of reasons:

1) Very little demand for it. For the most part, your GPU gets slow before you end up needing more RAM. As such not many people are interested in RAM upgradable cards.

2) Memory controller limitations. A given controller can only address so much RAM. It only has so many address lines, only so many electrical ranks. As such there are limits to how much RAM a GPU can even have. It isn't as though they could just add more slots and make it take whatever. Generally, the cards on the market have as much RAM as they'll take with double rank modules (since those cost less).

3) Cost. It costs more to make removable RAM sticks than it does. If the feature was free, maybe, but it's not and keeping cost down is important.

4) RAM stick availability. Many of the cards you mention used normal RAM that a computer did. This meant it was easy for people to go and buy it. Not the case with a GPU. They use specialized RAM. This would then have to be produced just for GPU upgrades, and of course would cost more money since it is produced in far less quantity than system RAM.

5) Speed. Video RAM pushes the limits of what we can get speed wise with our current technology. When you do that, things like the length of the wires and losses become important. Putting chips on a stick increases trace length, and also adds loss at the joint between the stick and the socket. So going with socketed RAM would mean slower RAM. Given that memory bandwidth is a major limiting factor in GPU performance, that's a bad idea.

So there is just no good reason to do it, and lots of reasons not to. While more RAM on a video card sounds nice, are you really going to use it? Would you really upgrade an old 7800 to 512MB from 256MB just because you could? Seems unlikely, seems more likely that you'd go and get a new card that has more RAM and more power.

Also, there are cards today with more RAM, for those that need it. The professional cards like Quadros and FireGLs are the same chips, with more RAM onboard. So supposing you have one of the rare applications that needs more VRAM, it can be had.
 
Also don't forget heatsinking issues. Most likely upgradeable memory would require pulling off the heatsink to upgrade. Most likely the specialized memory would be too high priced compared to another card.
 
Because BGA form factor would be almost impossible to make a socket for.
 
The reason devices in the 1990s had memory upgrade slots was because they were cutting-edge (wavetable synth from the AWE32, 24-bit color and 2D acceleration from the graphics cards), and thus didn't sell enough units to justify several SKUs. But since all these fancy features CRAVED ram (seriously, have you HEARD the AEW32's default 512kK wavetable ROM sample set? it sucks!), there had to be an upgrade path for those who could afford it. Thus it was cheaper to get everyone in the door at a reasonable price, and then sell upgrades to get any real usability out of the devices.

Of course, if you had any balls at all back in those days, you made the upgrade slots proprietary to milk every last dollar. I guarantee you RogueTrip, if they made their memory upgradable, you would be buying every last stick direct from ATI or Nvidia for superinflated prices.



Back in the day, it wasn't just memory add-ons, but co-processors and full "daughter-board" add-ons that improved or added capabilities. For example, a 3D daughter board could be clipped on to add 3D capabilities to a 2D board. Not quite the same thing, but you could even buy the Math Co-Processor chip separately from the CPU prior to the 486s.

This was less about "future proofing", but more about mass producing and distributing a "core" SKU and allowing the minority of power-users to buy any add-on modules that they needed. Sure sometimes, a upgrade board / memory module wasn't available until after launch, but it wasn't like there were "new technology" add-in boards coming out every year that made your product perpetually upgradeable.

Agreed, the cost / performance hits of taking features off-die and making them discretely "add-able" components just doesn't work today like it did back then. The advances in silicon integration eliminated the cost advantage of selling these optional add-in boards. And in many cases, it's impossible to adapt circuit-board techniques to meet the time-delay, power consumption, and radio-frequency interference standards required at gigahertz-range clock speeds that we operate at today.
 
Back in the day, it wasn't just memory add-ons, but co-processors and full "daughter-board" add-ons that improved or added capabilities. For example, a 3D daughter board could be clipped on to add 3D capabilities to a 2D board. Not quite the same thing, but you could even buy the Math Co-Processor chip separately from the CPU prior to the 486s.

This was less about "future proofing", but more about mass producing and distributing a "core" SKU and allowing the minority of power-users to buy any add-on modules that they needed. Sure sometimes, a upgrade board / memory module wasn't available until after launch, but it wasn't like there were "new technology" add-in boards coming out every year that made your product perpetually upgradeable.

Agreed, the cost / performance hits of taking features off-die and making them discretely "add-able" components just doesn't work today like it did back then. The advances in silicon integration eliminated the cost advantage of selling these optional add-in boards. And in many cases, it's impossible to adapt circuit-board techniques to meet the time-delay, power consumption, and radio-frequency interference standards required at gigahertz-range clock speeds that we operate at today.

this.

In an attempt to increase our computers capabililties back in the day we tried to find the memory modules mentioned by the OP to see if it was tangible. We found out that no store had them in stock (manufacturer never shipped these extra modules out) and they were like $50 a peice for EDO memory from the manufacturer!

Although a good idea, no real benefit comes this kind of SKU anymore.
 
Back
Top