How PC Memory has Evolved [Byte Size] | Nostalgia Nerd
Вставка
- Опубліковано 12 вер 2024
- Computer memory has come a long way, and the IBM PC Compatible has been at the forefront of memory technology since it's conception. Whether it's DIP memory, SIMMs, DIMMs, EDO RAM, Fast Page RAM, EDO Fast Burst RAM, SDRAM, DDR SDRAM, DDR2, DDR3, DDR4 of an other memory type. The PC has seen almost all of them, in an ever ongoing bid to increase speed. More RAM equals a faster system after all, and this was especially true in the 90s.. a period when everything was getting faster, operating systems were getting more bloated, and memory space was crucial for a smooth ride. We're now in an era with lightning fast RAM space which can hold many times the capacity of my first hard disk drive. So let's explore this wonderland of memory modules and find out how the world of PC RAM has evolved from the 1980s up until present day.
☟Share☟
Please consider sharing this video on forums, websites, social media, etc,. It helps tremendously.
☟Subcribe☟
www.youtube.co...
✊Support Me! ✊
Please consider supporting me through Patreon: www.patreon.co...
Visit my eBay Shop: ebay.to/1QQpYyy
Buy From Amazon (through my affiliate account): amzn.to/1OzCQWR
★Nerd Social★
Twitter: / nostalnerd
Face: / nostalnerd
Instagram: / nostalgianerd
Web: www.nostalgiane...
★Equipment★
Lumix G6 with Vario 14-42mm Lens
Nikon D3200 with 40mm Macro
Corel Video Studio Ultimate X7
Corel Paint Shop Pro X6
Blue Snowball Microphone
♜Resources♜
Some images sourced from Wikimedia/Wikipedia.
Music sourced from UA-cam audio library, on a non attribution basis.
![What are IRQ, DMA & Address Ports? - Soundcard Memories [Byte Size] | Nostalgia Nerd](http://i.ytimg.com/vi/aPKXmGkRxQ8/mqdefault.jpg)
![What are IRQ, DMA & Address Ports? - Soundcard Memories [Byte Size] | Nostalgia Nerd](/img/tr.png)
![Why are Floppy Cables Twisted? [Byte Size] | Nostalgia Nerd](http://i.ytimg.com/vi/CIfS6YKIf-U/mqdefault.jpg)
Feels like you kinda skipped over DDR. It gets twice the data rate by transferring data on both the rising and falling edges of the clock signal. It gets the same advantage of twice the clock speed without actually having twice the clock speed. I remember DDR being a *big deal* at the time.
Not to mention I've heard rumors(I wasn't around at the time so I don't actually know) of PC X being very static sensitive while DDR having the advantage of being so static resistant that you really didn't need to give a crap when you were putting it in.
The difference between SIMM and DIMM is not whether the chips are on both sides, its about the different contact points on the module
Yes, contact points on either side, and in the case of early DIMMS, this generally meant the chips were on both sides too.
Nostalgia Nerd, Typically only with higher capacities (depending on the density of ram chips used). As SwitchingPower said, what made a DIMM and a SIMM diferent was a SIMM used the same pin for both sides of the board. A DIMM however used separate pins on both sides. It had nothing at all to do with memory placement as prior to the DIMM, SIMMs (72-pin mainly) could have ram chips on both sides as well. www.integralmemory.com/faq/dram-what-difference-between-simm-and-dimm
I was going to say the same thing, But I thought id see if someone already did first.
@@Nostalgianerd Simms could also have chips on both sides, but always had a single row of contacts, even with pads on both sides of the PCB. You showed but didn't mention the predecessor to the 30 pin SIMM - the SIPP or Single Inline Pin Package. Later SIPPs were just SIMMs with soldered on pins but the early SIPPs had plain square ends without any notches since the SIMM socket didn't exist yet. Before all that the most common DRAM package was the DIP, but to save space for some uses there was the ZIP or Zig-zag Inline Package (you also showed some of them) that moved the pins to straight out one side of the package in two close staggered rows.
To get real nostalgic, how about a video on SIMM Stackers and SIMM PCBs to solder DIP chips onto? If you didn't want to DIY there were companies that would take your old DIP chips, solder them to SIMM boards (for a price) and send them back to you. The problem there was it only worked with the fastest DIPs since most computers with SIMM sockets needed faster RAM. Some of those companies would also strip surface mount chips from small SIMMs then solder them onto boards supporting more chips. That avoided speed issues but then your computer had to support SIMMs with a lot of chips. ISTR this may have been offered for some DIMMs, but likely only for FPM and EDO types.
One benefit of SIMMs and DIMMs with 16 or 32 chips is they could allow increasing RAM past the official maximum on *some* computers. This was especially true of many Macintosh models. Except certain ones like the early LC series, and the MacTV where Apple deliberately hobbled their RAM capacity because Apple was afraid of it's own computers 'stealing sales' from it's own higher priced models. So when the choice was between the crappy but cheap LC or the "I could buy a car for that!" IIfx, some buyers bought a PC instead.
@Kevin Counihan Google found funny looking contraptions. I never knew that they could be stacked so...
I remember my first PC in 1998. 16MB memory modules and 2GB hard drive. How things have moved on since.
now 2GB is considered low-end...
@@christiankm8820 no 2 go is on a micro sd card
@@scottschoppert9149 no i meant ram
@@scottschoppert9149 bruh 2GB sd cards aren't even being manufactured anymore
"square chips", image shows (just like most ram chips) very non-square chips
What about Rambus RDRAM RIMMs? I remember having those inside of my old, P2, Win 98 PC. As well as those weird "empty", chipless modules. Maybe they weren't relevant enough, but I've read that there was quite a bit of unfair competition in favor of Rambus fueled by Intel themselves in the early days of Pentium chips.
They were a bit of an oddball. Kind of a backward step, I thought it best just to brush them aside for now. Might make an interesting separate video.
I wouldn't call RDRAM a step backwards, at the time it was faster (and more expensive)
At 0:55 the second one down was a SIPP a variant of the 30 pin DIMM.
I remember in 88 that a friend of mine bought a 030 accelerator for his A500 and I had to insert all the DIPP memory chips into the sockets. The board would flex while inserting the chip and you would hear a cracking sound, he walked back and forth like an expectant father with worry. 32 chips later and he had a very fast Amiga for raytracing.
I think when it was introduced it had a faster clock rate, but bottlenecks in practice led to it being slower than SDRAM. I believe this improved with faster clock rates, but speed differences were never enough to justify the price difference.
Nostalgia Nerd RDRAM was a case of application appropriate usage. As a desktop memory it really didn't stretch it's legs as most applications jump around in memory and the latency was the issue. However if you did rendering, large databases or anyother application that used large blocks of memory the bandwidth becomes king.
Cost was too high for it's limited use on a desktop, but I know guys that did 3D rendering loved it (but even they complained about the cost)
you also had to pair these even if you had to use a "blank"
That Uplink music, I must have spent thousands of hours on that game.
You missed a very important principle with the DDRX memory types: multiplexing. DDR2 with an external clock of 800 MHz doesn't really clock at 800 MHz internally. The DRAM cells themselves clock at 400 MHz, but the output is multiplexed. This has the effect of doubling the bandwidth, but the latency doesn't improve.
DDR3 takes this one step further and multiplexes data from 4 cells, and DDR4 multiplexes from 8 cells.
The DRAM cells are getting smaller, but their characteristics aren't improving much; typical values for good memory:
DDR1, 400 MHz CAS-latency of 2
DDR2, 800 MHz CAS-latency of 4
DDR3, 1600 MHz CAS-latency 8
DDR4, 3200 MHz CAS-latency 16
HBM introduced for GPUs by AMD and Samsung doesn't use multiplexing. Instead it stacks memory chips and uses SIVs and and silicon microbumps on a silicon interposer to create an extremely wide memory interface cheaply (e.g. 4096-bit). A 4096-bit bus on a PCB would be hopeless; but here you never leave silicon. Both AMD and Nvidia will use HBM2 in the future to reach some obscene memory bandwidths at low power and unchanged latency.
And GDDR4 & GDDR5 are branched off (basically) from DDR3, with GDDR4 being barely out of prototype before being replaced with GDDR5 which has been the dominant VRAM type for ~90% of video cards since (really old/cheap cards like the GT210 being in DDR3 versions), which is slowly being replaced by HBM/2 & GDDR5X on the higher-end new offerings from AMD & Nvidia (though we'll likely still see GDDR5 on cards like Nvidia's *60 and below lineup for a fair while).
Precisely and AMD is about to launch their Vega graphic cards within the next two months which uses HBM memory and OMG, they are Terabyte fast.
I really like your videos. brings back so many memories. it's so odd remembering how expensive and important memory was back then. we literally chose to buy computer parts that took money out of our food and clothing budgets.
FYI the song is Sad past - Silent partner
I remember when EDO came out. Got some. Loved it.
My grandfather's 1995 pentium 133 still used edo ram. Sd ram was expensive for many years. Even the earliest 3dfx voodoo graphics cards had edo ram. Even voodoo 2 from 1998.
I know your videos since yesterday: I became a patron and bought your book. Man you are a walking encyclopedia!
Ooh, remembers the bad old days of single sided vs. double sided, high densisty vs low density, and the 2 chip / 3 chip SIMMs that some systems just couldn't get on with.
Another trick, used on a 20MHz 386, was page interleave mode, where the slowness of DRAM was compensated by alternating the two banks, repeated with Intel's "800 FSB" using dual DDR400
Awesome video, really like your content!
What exact song did you used from the UA-cam Audio Library because it's awesome.
One of my first jobs was with the Amstrad User Group. We offered an upgrade path from a PC1512 to a PC1640HD. As well as fitting a 10Mb Hard card into one of the 3 8bit expansion slots, an EGA graphics card into another, we fitted 18 8kb chips onto the motherboard. Later ones used 6 32kb chips.
Trip down memory lane there :P I remember putting an extra 256k on my old OAK 067 VGA card in my 386SX-25. Changed days
Taking the CompTIA cert (which is currently versions 901 & 902) and here is my two cents:
The video was fast but helpful. I definitely watched certain parts more than once. Unfortunately, our official exam guide does include at least a bit about Rambus. I guess if anyone else is watching this to study, don't complete ignore Rambus because it might actually be on the test. Reading the other comments is pretty informative as well.
lov you nerd, ty for teaching us so much cool stuff
SIPPs were a thing! Why neglect that nifty step between DIPPs and SIMMs.
Also, it would have been nifty to include what the terms meant, also. Even if mentioning it as dual or single inline at one point, and memory module or pin package in a different sentence.
I think you mean SIP and DIP, not SIPP and DIPP. SIP and DIP were chip package layouts, while SIMM and DIMM were about the printed circuit cards that the chips were mounted to. Yes, there were SIP packaged chips. They just weren't that big a deal. When the DIP was popular, chips were typically socketed because they were costly and failed more often than you'd want a computer to fail. The SIP package was almost always soldered directly to the PCB, and was seen more often in non-computer applications. As competition grew in the computer market, manufacturers started eliminating the DIM sockets, and started using less expensive SIP chips.
Putting chips on daughter cards was as old as the microcomputer. The Altair 8080 used expansion cards for RAM, and so did many 8086/8088 computers with just an 8-bit bus. The big change was moving the daughter cards off the main system bus and connecting them to the memory controller directly. This only worked on computers that had memory controllers though.
I love your videos, man.. keep up the good work.
No, DIMM is not because chips are both sides, it is because the gold CONTACTs are both sizes, while in SIMM was 30 or 72 pins (IIRC) single size, or both sizes with the same electric contact. You said it wrong in the video.
About 72-pin SIMM - on Pentium (Socket 5/Socket 7) - it was needed to install in pairs but to be clear - on 486's platforms - 72-pins SIMM can be installed 1,2,3 or 4 SIMM and "pairing" was not necessary
The background music (along with the mostly-static diagram) was giving me flashbacks to middle school science class. Felt like one of those old school VHS presentations about the miracle of life. Fortunately, your narration is much more interesting and you didn't include microscope footage of embryonic cells dividing.
0:40 "WE" is a "Wilk Electronics" sticker, a former name of "GoodRam" company. "Wilk" stands for "wolf" in polish.
Nice detail! Also whats the song used in the video?
Fascinating. I had no idea memory worked like that - that it works by grabbing nearest neighbors (and sequentially).
The music is amazing in this episode, do you have the track name?
When you were playing with those ram sticks falling off your head all i had going through my mind were the random times in pc history where a natural disaster or some other event caused a ram/memory shortage and prices sky rochetted or simply couldnt buy ram.
Also, why do people call the hard drive "memory"?
What you call bank (8 or 9 chips) is actually rank.
Bank is internal part of chip, not part of memory module.
I've _never_ heard the term rank. Although I suppose it must have been used somewhere along the line.
I remember using only a single simm stick with Pentium Motherboards and had no problems. So there is a performance hit if you don't use two sticks as a pair?
Excellent explanation.
I had a pc with the RDRAM sticks back in the 1990s. it was bloody hard to find upgrades for it.
A friend of mine has about 8GB of RDRAM. Little good that it does now. He also has some oddities such as 512MB to 2GB PC-100 memory.
having a large cache hdd helps with the limited spinal speed. having a ssd drive with a cache drive by SanDisk absolutely rules, not really necessary but very effective.
Thanks, this helped me study for a test :)
A trip down memory lane...
Nope Static RAM was used on some of the very first home computers EG the ZX80 and ZX81 used two 2114 (1kx4) SRAM chips. When the 486 was around there were Cache RAM blocks that were Static RAM devices too. Static RAM will always outperform dynamic RAM for speed. FPGA's implement pretty much everything in SRAM and multiplexers with the exception of I/O D-types for making everything synchronous (there are some families with function blocks built in but these are not purely FPGAs). There are FPGAs now that can exceed 50GHz on PAM4 transceivers and that gets clocked into RAM. DRAM at the moment just could not keep up with that. The Cache RAM on the processors is SRAM too.
thats some chunky looking memory!
small correction
bank is internal to chip, bank never spans multiple chips
rank is what you describe here
rank is 16/18 (SIMM) or 32/36 (DIMM) or 64/72 (SDR or DDR DIMM) bit wide ALWAYS with no exception
each single chip have its own banks (with advent of DDR4 it may be even 32 banks, if I am not mistaken)
no DIMM ever will go any further then 8 ranks per DIMM as it is already huge load on memory controller
there was a time when rank was called side and you had single sided ram and double sided ram, while actual co arrangements could have been opposite to what people were calling them
those sides were ranks
You showed SIPs but didn't mention them - Single Inline Pin RAM came between chips and SIMMs
you had a picture of it, but you glazed right over SIP or single in-line pin memory
There is also some ZIP packaged RAM at about 3:03. I wonder if any static RAM was pictured?
is that phoenixsc outro
[5:34] not Gigabytes per second, it's Gigabit per second.
Example with 1.1 GBps: 1.1e9 / 8 / 1024**2 is ~ 131 MiB/s.
Great video!
this is very informative.
Back in the day I paid £130 for 1MB (in a 4 board config) from Scan Computers. That was a lot of money for me at the time. I used to live 5 minutes from Scan.
this increments are not banks, but ranks
I refer you to any DataSheet of memory controller (like AMD761 or AMD K8 CPU)
why this choice? you may ask
Both AMD761 and K8 and even K10 support both non-ECC and ECC RAM, while AMD761 supports even both Buffered (registered) and unbuffered (unregistered) RAM
there evening is explained
rows, columns, banks, ranks
just as a warning there is 2 rules of rows
row in actual deep inside RAM, which was just 11 bytes long at best in SDR
and row as seen by row buyer which is 4096 bytes long per each 32bits of RAM width (reason for meltdown - 64bit wide RAM have 8192byte long row, spanning 2 pages of 4KB)
I think DIMM and SIMM has to do with the pins on the bottom. With SIMMs both sides of the module has contact points. but they are simply connected to each other. With DIMMs I think the pins on one side are no longer connected to the pins on the other side. I could be wrong though.
No, you are absolutely right, sir.
I feel like you missed mercury delay line technology.
Back to the days of Alan Turing, I believe. I have also seen the delay line trick done with an acoustically-coupled spring.
you showed a picture of one, but you did not mention SIP memory, single inline pin. that was the precursor to sim.
also was kind of hoping you'd have started with core memory
well it is a byte size video. If we really wanted to start, we could go back to mercury delay line technology and also the Williams tube. Funnily enough magnetic core memory is non-volatile and was really hard to break. It was used in the Space shuttle program even through the challenger explosion in 1986. Magnetic core memory had started to fade away in popularity after 1975.
I did not know that! :)
Pulling data off the hard drive is like going to the store to get food. Pulling it from ram is taking it out of the fridge, and CPU Cache is what's on your plate :)
That's L2 cache. L1 cache is what's in your mouth. 😂
This is a very interesting and educational video! I also like the music used although I'm unfamiliar to it. Mind letting us know what's the song's title and artist?
Please!!
And what is the difference from RAMBUS?
I remember it was supposed to be faster than the faster RAM (pre-DDR that is) and then… just disappeared…
Excellent video. ty
I would say to those chip manufacters, don't bother, just make SRAM chips with the fastest leading conductive meterial possible to get fast speed, it's expensive but it will pay off because speed is now a huge demand these day's.
Great video.
Was it possible to buy true SRAM SIMMs or DIMMs back then?
Silent partner sad past
what I hate the most about memory is the limited amount of conventional memory at 640k with extended memory being ram. having to use emm386.exe frame cc00-d000 tricky to set if you have a uma shared graphics memory.
Gid work on the photo work and edit.
If tis doesn't mention bubble memory I'll be sincerely disappointed.
OobiDoobScoobyDoobyBenubi In my limited experience with bubble memory on mid 80s vintage cnc controls, it's used as a storage device so you don't have to load programs from tape after every startup.
AngeredKabar I know that there was at least on computer (an early laptop if I recall correctly) that used bubble memory as the primary storage device. Always been interested in the technical details of it since I learned of it, though they're few and far between nowadays.
bubble memory was supposed to be really fast, but then the technology got quickly surpassed by semiconductor memory, in both speed and cost. It dead after that.
The second gen simm was called a sipp
I was hoping for a deeper history of computer memory, like, going all the way back to magnetic core memory or core rope memory like I've stumbled across on wikipedia, like, how the fuck does that trippy shit work, lol!?
DDR stands for DRAM DRAM Revolution.
the music in this video is too loud... sounds like host is shouting over it most of the time
Nothing about rambus or SIPs?
what was that website from 5:29 to 5:39?
I now have a new respect for SD RAM
I'm not doubting what you say but I've never seen a Pentium that used DIMMs
Many boards had 72 Pin and 168 Pin slots.
And then you got machines like my desktop, with it's weird fully buffered ECC DDR2 memory.
Compaq by any chance?
daishi5571 No, 2008 Mac Pro workstation, now used as a Windows 10 desktop. Has 8x 4GB memory modules, but the PCI-E slots are loaded with PC only hardware now, so OS X is no longer usable.
Brandon MacEachern Interesting, did you get the system to run Windows on it or was it just something that happened.
daishi5571 I got tired of the way OS X was going, and Linux just wouldn't cut it for me, as far as games were concerned (or the sound card I wanted to use), so I chose Windows 10 (as it was the one I could buy legally). I wanted to start playing more games too, so I got a modern GPU to slap in it, but I'm finding my shakes in my hands, are starting to be a bit of a disability sadly.
I have a 2008 motherboard still chugging away as a Linux machine with 20 hotswap SATA bays using ECC DDR2 also. It has 16x4G in there, and dual Opteron 2384 quad core CPUs :)
i used to have a old 72 pin sim on my key chain, lol
And the Ram pairing requirement does not apply when you use dimms instead of simms?
It depends on the computer. My Mac Pro is very picky about the placement of memory modules (because it uses a Xeon chip). The largest memory modules need to be in the "lower" memory slots, smaller memory modules in the upper slots.
So two 8GB DIMMs in slots 0 and 1, two 4GB DIMMs in slots 2 and 3.
Other processors, like the Pentium, i5 etc. don't need pairing, as a DIMM is actually a pair of SIMMs on a single circuit board.
Thank you, but I was referring to memory used in vintage 1990's 486 and Pentium computers.
geo58impala Ah! Yeah those need pairs of SIMMs spaced evenly. So if a SIMM was in slot 0, another would need to be in slot 2. Slot 1 and slot 3, etc.
Yep, those are sure some square rectangles. That's a weird script oversight. Especially because some chips ARE square, so this is not great for educational purposes.
Completely skipped rambus?
you forgot ddr5 for gpu's
The chips themselves are RAM, not the small circuit boards.
Shh, he thinks those rectangular chips are square, so he also might think the chips are for decorative purposes.
:-) Yes, those are memory chips on memory modules.
What about MRAM?
how graphics card has evolved?
song?
You forgot Windows Vista ReadyBoost ;)
the first theme music (not the intro) makes me cry what's it's name?
😭
I remember it from a Minecraft mods pack
It's called "Sad Past" > ua-cam.com/users/audiolibrarymusic
Nostalgia Nerd thanks
PhoenixSC uses it
But what about core memory? You dirty core!
don't you mean gigabits instead of gigabytes?
Hell nope! - en.wikipedia.org/wiki/DDR4_SDRAM#JEDEC_standard_DDR4_module
DDR4 2400MHz (aka: PC4-2400) on the JEDEC standard has a Peak Transfer rate of 19200MB/sec, or 19.2GB/sec, which is pretty damn close to the "about 20 Gigabytes per-second" he said.
And even going to the damn slow PC4-1600 standard (DDR4 1600MHz), you're still looking at about 13GB/sec transfer rates.
Keep in mind this is for RAM that comes in sticks of up to 16GB for the consumer-grade Non-ECC type (ECC = Error-Correction Code. Not used for DDR4 on Skylake chips on the LGA-1151 socket for consumers, nor on Haswel-E/Broadwell-E i7 chips on LGA-2011v3 for enthusiasts, but *IS* used by Haswell-E/Broadwell-E Xeon chips on LGA-2011v3 for servers).
my bad
***** S'alright man, s'alright.
I feel kinda like a god by having 16 gb
janno288 what about when you got your first 32mb, 64, or 128? I remember racing home to install my new 64mb set that would finally get me to 128mb I’d RAM.
@@Darsithis i used my dads PC so i Had128 mb and i remember upgrading it to 456 mb
40$ a megabyte!
phoenixSC music
1011 likes, 11 dislikes... What???
Ahhh, you meant for an Intel based machine because PC != Intel exclusively but does include them since PC means Personal Computer and that includes Mac, Amiga, C64, Atari, Sinclair, etc... (you get the idea) and, for instance, the Amiga used ZIP chips for its extra memory so this video needs to have some specificity to it instead of the generalization of PC.
Sorry, been enjoying the content of your videos for a while now but my irritation at the way you pronounce 'th's has got the better of me. If it's not the constantly-annoying "muvvaboard", it's the litany of "ve"s instead of "the"s in your videos that are slowly putting me off. It's like listening to Jonathan Ross. Love the videos otherwise! 😁