While all of this is true one of the biggest reasons why Rambus failed is that in the PC industry standards are not owned by any one company. Rambus had patents on all of it's hard ware. Meaning if you wanted to make Rambus compatible hard ware you'd have to pay them a licensing fee. Basically they were trying to corner the market on ram/memory.
@@prizmatik8696 You can make a computer without HDMI. You can go with display port which is a open standard and royalty/license free. Rambus also wanted HUGE licensing fees at the time. That coupled with the issues pointed out in the video made it a hard sell to manufactures. Rambus was also having issue making enough of it to go around leading to even higher prices.
This story is missing so much. The PS2 also used RD RAM, the high price of Rambus memory was (allegledly) at least in part due to suppliers conspiring to drive up the price and make DDR more popular. Rambus was on a lawsuit about that for years, but eventually lost.
The reason RDRAM, despite being tricky and expensive to manufacture, was cost effective for the N64 was because the console only had 4 MB of system RAM, which could be expanded to 8 MB with the Expansion Pak. When the Pentium 4 came to market, 512 MB was rapidly becoming standard for high end PCs, and 256 MB of RDRAM was much more expensive than 512 MB of SDRAM. Intel tried to alleviate that by including a stick of 128 MB RDRAM and a continuity module with the Pentium 4, but even this didn't help increase sales, and they ended up giving up this exclusivity by 2001. Also, the reason Intel chose RDRAM for the Pentium 4 was due to its many architectural... Umm... Characteristics. Lol. Well. It was much more bandwidth-dependent than previous architectures due to a much deeper pipeline and a very fast but very small L1 trace cache system, so RDRAM's bandwidth advantages would keep the caches and the pipeline well fed. In fact, P4 shows a very mediocre performance when paired with SDR SDRAM. This of course was reversed since the introduction of DDR SDRAM. Fun fact: Intel also supported RDRAM with some high end Pentium III chipsets. In fact, they were the only ones that supported a higher speed AGP bus version for a while, which of course made people really mad. Intel eventually also gave up here, and soon they released a high end chipset that supported traditional SDRAM, as well as this faster AGP slot. And, since P3 didn't have the same bandwidth hunger as its successor, the performance gains with RDRAM were minimal.
I had one of those Pentium 3 with 128mb RDRAM at my first workplace 15 years ago on an old Win NT computer. It even had support for 2 CPUs(only 1 installed). Sadly, we had to throw it away when it became obsolete because work.
@@jonsmith-z1v Yep, RDRAM mostly had high throughput for sustain serial reads IIRC, and for it's bit-width. And P4's could take advantage of that. However they also faired just as well with a ddr based MMC, better in some corner cases and better with rdram in a few more corner cases. Money changed hands or Intel would likely have only used RDRAM based P4's for specialty scenarios and big iron where they could put a lot of rdims together for a wider buss.
The RDRAM on the N64 use a 9bit bus to save on traces even more (the extra bit is meant for parity, but the N64 use it for anti aliasing). It is fast if you want to read just the next byte and the next byte etc.. you just go read-read-read-read and the memory has this internal address pointer thing that increase automatically. But if you need to do random memory accesses, you need to change the internal address, and that takes like 4 writes, so it's very very slow. And if you have the CPU and video chip fighting over the memory address, things gets very slow
N64 RAM is slow. AFIFAK N64 has ~0.5GB/s memory BW (same as Pentium 2 - could have 0.5GB/s with SDR-66 memory and 66Mhz FSB) Those 2 came 1996 (N64, in US and JP) and 1997 (N64 in EU and Pentium 2). Even worse, Pentium 2 has all that memory bandwidth for itself (without GPU) and it also has 256k-512k of L2 cache, 16k L1I, 16k L1D caches where N64 shares bandwidth with GPU, has no L2 cache and only has 16k of L1I, 8k of L1D caches.
@@volodumurkalunyak4651 for the time that's super fast, given the CPU runs at 93Mhz and the coprocessor at 62 However it rarely ran at 500MBs because of the adress changing hell
@@volodumurkalunyak4651 It was far worst than that. It 500MB/sec on paper/peak theoretical. In practice it could barely maintain above 100MB/sec. It has been fun seeing all these home brew hackers figure this out in far greater detail that what was available in the 90's. Recently they have shown there are undocumented registers that show when the CPU/GPU are active or not. Turns out he CPU is idle about 70-80% of the time and the GPU is idle about 60-70% of the time. This is all due to it waiting on the RAM. Could have really used that info back then! Most of the visual edge on the N64 was handled via additional core functionality and internal caches. Reality Engine and its functionality (Texture filtering, proper perspective calculation, actual understanding of depth via a Z-buffer!) was basically the only thing stopping the N64 dropping down to Ps1 performance.
@@dan_loup Is it just me that finds ironic that the console marketed as 64-bit had a 64-bit CPU cut down to have a 32-bit bus, and now I discover its memory had a 9-bit data bus? Lol.
I was calling around computer repair shops in town asking if they had any DDR4 RDIMMs in stock, so I could finish a job. One of them said "oh, no, they haven't made that for decades." I was flabbergasted, and argued with the guy until I found out he thought I was looking for Rambus memory. (FYI "RDIMM" is registered memory for some servers.)
I remember the DDR2 FBDIMMs for servers being dirt cheap. They didn't fit in consumer boards so they were absolutely worthless when they became professionally obsolete.
Exactly! RDIMMs or RDRAM were terms used to describe ECC Registered (buffered) memory mainly used in Intel Xeon based servers or 3D workstations. I assembled many Intel barebone systems based on that tech back in the days.
Im sorry but what’s wrong with protecting one’s intellectual property ? I mean I do think they should be forced to let others use it at a price but u shouldn’t be able to just rip off the IP
@@MathCuriousityYou don't know what a patent troll is if you're taking their side, or you are one. A patent troll is someone who files patents on a avalanche of ideas having never lifted a tool to make any a reality, then sitting back and waiting for someone to come up with a similar idea without realizing it and slamming them with bank breaking lawsuits. They do nothing but stunt the rate of technology by forcing you to consult an army of lawyers before doing anything, assuming you even can after the lawyers let you know if a patent troll is sitting on the idea.
You could not run one C-RIMM and one RD-RIMM together. The sticks needed to be installed in pairs, and the C-RIMMs needed to be installed in any unpopulated sockets.
still got a hand full of EDO, SIMM, and RAMBUS in my office lol Deal with a lot of manufacturing clients who have REALLY old machines. I even had a P3 and P4 Rambus boards as well (Not my daily drivers). Now i did see a bump in performance with it though.
Personally, i7 4770k - RTX 2060 12Gb 😅 But in a lot of games my GPU and CPU can't reach 100% of usage, so I am assuming that it's my old DDR3 that can't keep up. I will soon move to AM5.
when I was in highschool I had a p4 with 256mb RD ram. with an ASUS geforce 3 ti card. it was given to me by my friends neighbor who said it was out of an old alienware
Explanation of double data rate: The clockspeed here refers to the wavesignals waveform, so from rising edge to another rising edge. As far as instruction cycles or logical cycles go the line needs to of course be read twice per each cycle. So in that sense its not double rate. The advantage from using both the rising and falling edge to transfer information comes from how alternating current introduces design problems as the frequency rises. Impedance and phase shift are porpotional to the frequency, meaning it becomes the limiting factor. But if you use both the rising and falling edge to transmit data you still have to hit the same timings on low and high peak, meaning the second bit of information is ”free” as far as designing the bus lane goes. The devices still need to act upon the lane twice as many times as usual, else they couldn’t distinguish between rising + falling edge and no pulse. As far as I understood it they don’t get the second bit of information for ”free”. But that’s fine, because the long bus lanes and their signal intergrity is the limiting factor, not the devices at each end or their logical clock speeds.
That last line... bad guys who become less bad and finish... ... ... ... eventually! THAT was priceless! This def brought back some memories... thank you for this piece.
I had RDRAM in my P4, in fact, I still have it somewhere. I even managed to upgrade it to 512mb a few years after buying the PC with 256mb. It cost an arm and a leg. But the PC was actually quite fast and I was very happy with it for years.
I am a computer nerd, last night while in bed I recalled RAMBUS RAM & how we needed continuity module, wondered what happned to it, and here we are today....
I had a dell XPS desktop that had RAMBUS. It was so cool. My memory might be mixed up, but I think it was one of the few computers that operated the BTX case form factor which also didn't last long.
Oh yeah, I picked up a used 1st gen P4 motherboard with RDRAM in about 2001 and that served as the core of my daily driver for ...5-6 years? I remember selling a couple sticks of RAM out of it when I got something newer because the prices on ebay were quite remarkable...and I was smart enough to not lose those continuity modules :-)
I had a customer come in asking us to fix their computer that had a Pentium ii and SDRAM 😅🤣 Ended up convincing them to upgrade, took SOOOOOO many attempts to pull data off the hard drive, the drive was just dying 🤣🤣🤣
back when we all ran xp ram worked entirely different. It was like a cars rpm it only ever ran when you hit the gas and opened a program. like computers idled at like 0-5% ram usage.
My first computer build was going to be a pentium 4.... but the implementation of RDRAM with the P4 was very 'bursty'. Very high performance for short stints of time, but poor long-term performance for things like rendering. As I was getting into video editing at the time, this forced me onto the previous gen P3 chips which didn't benchmark as well on synthetic tests, but all reviewers agreed had much better performance for sustained workloads.
Most of my stuff is still DDR3. It's okay because I'm a retro sort of guy anyway. Yeah, I went on a bit of a used office-PC buying spree when third and fourth-gen Intel machines were the ones to get a while back. They still do their thing quite well. I remember when all the Rambus controversy was happening. I remember it was not well received on the tech sites I checked. Opposition to it seemed overwhelming from the user side. The nice thing from the user side is that AMD was there to go in their own direction so Intel and Rambus couldn't try to brute force this arrangement once it became clear that no one was happy. As I recall, I got a computer a little bit before, and shortly after all this, so I was able to sit the whole thing out and never dealt with Rambus stuff.
I went to a recycling center and found an old P4 motherboard with RDRAM, 2x 128MB modules and 2x dummy dimms. Did some brief tests and the results are quite interesting. Sadly the motherboard only supports the early P4's, so no option to use the newer P4 HTs.
Remember RDRAM? YOU BET! The N64 Expansion Pak did give you more Memory...which was 4MB. It did help improve graphics, but it was more for Donkey Kong 64 on memory leakage help.
It took _fast_ DDR3 to match or exceed the throughput of RD800. My RAMBus machines are my vavorites, easily. Pairing a GDDR3 AGP card with RAMBus 600 RAM kept my P.C. playing some of the newest and most intensive games on the market for _four years,_ and then only started a slow decline.
@@volodumurkalunyak4651 Um, no at the time. DDR2 800 didn't exist yet, not even in engineering concept, at that time. DDR3 was off the engineering development board and in production LONG before DDR2 got that fast. DDR2 overclocking eventually made DDR2 that fast commercially viable as a market, for us guys who couldn't afford a DDR3 motherboard! Whereas, RD800 had more bandwidth than the early, slow, DDR3 that _was_ on the market at the same time!
@@davidgoodnow269 single channel RD800 and single channel DDR2-400 are equivalent, dual channel RD1600 and dual channel DDR2-800 are also equivalents in memory bandwidth. RD800 comparable to DDR3 - no way. Good luck finding dual channel rdram boards.
I had a slot card Pentium 3 with RAMBUS. I remember overlocking to 500mhz on it and being completely blown away by its mind boggling power. Man those were the days
I love ram videos. Can you explain why ram signals can't be multiplexed or why a bus can't be optical where different interconnects can't use the same bus on a different frequency/phase rather than shutting off the bus while it's in use? It's incredible to me how fast Ethernet has gotten so damn fast over the years by doing this, but ram takes massive power and specific slot placement, channels available depending on sticks installed, bios profiles and all that. It seems crazy to me how slow ram is tbh. Also, why is ram built into the motherboard on high end boards? Solder 4gb of hbm next to the cpu and let sticks be added for when that mobo ram fills. Even if it only runs system and os, a level 5 cache measured in gigs would absolutely phenomenal even if only the os used it to help speed up the chipset io. Chipset hbm buffers would be amazing!
If you’re talking about straight up throughput, then yeah, ethernet is faster. But that’s where it ends. RAM is _orders of magnitude_ lower latency than Ethernet (a million times lower, to be exact). Ethernet latency is measured in milliseconds while RAM latency is measured in _nanoseconds_. Why is Ram situated as it is? Latency. At that scale of latency minute differences in the length of traces materially impact the latency of one bit vs the next. Traces need to be the same length to keep the latency consistent. CPUs/GPUs need to switch between instructions far, _far_ faster than Ethernet could ever even dream of being able to provide the requested data for. Sure, compared to RAM, Ethernet is analogous to a fire hose vs a squirt bottle. However, try changing the direction of a firehouse as fast as you can a squirt bottle. All that throughput is useless if it takes comparatively forever to retrieve it. Besides all that, Ethernet and RAM are two completely different things, they’re not interchangeable. As for HBM with extra cache, that’s somewhat the direction things are going in, but HBM alone sends complexity through the roof, adding an external cache to the mix just multiplies the complexity to be (currently) out of the realm of practicality for all but the most demanding use cases. There’s a reason those high end boards you mentioned are high end: it’s not due to the presence of it, but rather due to the complexity required to have it there in the first place (and have it work in a beneficial fashion).
I wonder if a continuity module would make sense with DDR4/DDR5 for all of these memory busses that have reflection issues if you only populate the nearest slot, or apparently in some systems on new high speed memory, any empty slots.
DDR3 user here. I felt that comment xD but yeah if it wasn't for the end of windows 10 support i still wouldn't have upgrade plans xD.. i7 4770k @ 4.5 Ghz since 2013 xD
The computer that i use in my job still uses ddr2, i wish i was lying (the processor was released in Q1 of 2009). We use windows 7, office 2013 and we can't even update chrome.
My first PC wad a custom made Dell PC with 800 MHz Pentium 3 and 128 MB of RDRAM. I wanted this expensive config (at the time) to play the newly released Unreal Tournament. Damn I feel old...
I remember telling a friend to invest in RamBus but his advisor told him no. But he put in a $1k anyway on my word. The next day Intel made its announcement, and the stock started soaring. He sold pretty quick and did make about three times his money. He called me, thanked me, and said he owned me lunch next time I was around. LOL.
Ah, the times when i had a fulltime job but still lived at home with mom and dad. So i could spend all my money on hardware instead of kids and mortgage :) Had a watercooled 2ghz P4 @ 2.4ghz and 768MB (256+256+128+128) of RDRAM. And 4 single platter 500gb HDD's in raid-0 for absolute peak performance. Made absolutely zero difference in gaming but damn did i have fun wasting a ton of money on that old PC.
I don’t think RDRAM would have been the right approach for modern RAM. We’re seeing a trend towards more cores (wider CPUs), because frequency is falling off, and my suspicion is that RDRAM would have likely not had a huge speed advantage over current high end RAM approaches, but having a lot less bandwidth due to the smaller bus.
RD-RAM was a huge bottleneck on the n64 as well. The high latency caused a lot of problems for the system, and were a major contributing factor of why the system was difficult to develop for.
Worst choice I ever made,jumped all the way from our first windows 95 pc (pentium 100MHz) to a new Dell PC, Intel Pentium 4 CPU, RD RAM and Windows ME, the horror! Mercifully died after less than 5 years by which time laptops were practical for university and I think we were on xp by then
Didn't Rambus showcases industry’s first DDR5 MRDIMMs & RDIMMs memory modules speeds Up To 12,800 MT/s just a few days ago this week ? In 2004, Rambus sued the quartet of Samsung, Infineon, Hynix, and Micron for conspiring to low price fixing DDR RAMs that causes Rambus expensive solution to lose sales. In 2005 Infineon settled with Rambus for $150 million while in 2010 Samsung settled with Rambus for $900 million. Hynix and Micron won against Rambus in 2011.
Joke's on you, I already knew about RDRAM one thing I remember that was different between SDRAM and RDRAM was the heat spreading, SD ram would use each chip separately, filling one after the other, while RDRAM would use the whole bank filling them all evenly, which generated less overall heat per chip than using 1 chip after the other. If that's wrong, I would love for someone to educate me :)
While all of this is true one of the biggest reasons why Rambus failed is that in the PC industry standards are not owned by any one company. Rambus had patents on all of it's hard ware. Meaning if you wanted to make Rambus compatible hard ware you'd have to pay them a licensing fee. Basically they were trying to corner the market on ram/memory.
BINGO!
So basically dvd?
so hdmi?
@@prizmatik8696 You can make a computer without HDMI. You can go with display port which is a open standard and royalty/license free. Rambus also wanted HUGE licensing fees at the time. That coupled with the issues pointed out in the video made it a hard sell to manufactures. Rambus was also having issue making enough of it to go around leading to even higher prices.
While having an inferior product...
You said “ Nintendo “ , this video is demonetize already and it’s flag for taking down
💯🤣
Too bad TH-cam doesn't demonetize him for shilling that dodgy VPN.
Flagged*
(Your comment has been removed by DMCA)
Not only that, they showed Nintendo's I.P. with the expansion pack and jumper pack. The ninjas are already on the way to the studio.
This story is missing so much. The PS2 also used RD RAM, the high price of Rambus memory was (allegledly) at least in part due to suppliers conspiring to drive up the price and make DDR more popular. Rambus was on a lawsuit about that for years, but eventually lost.
The reason RDRAM, despite being tricky and expensive to manufacture, was cost effective for the N64 was because the console only had 4 MB of system RAM, which could be expanded to 8 MB with the Expansion Pak. When the Pentium 4 came to market, 512 MB was rapidly becoming standard for high end PCs, and 256 MB of RDRAM was much more expensive than 512 MB of SDRAM. Intel tried to alleviate that by including a stick of 128 MB RDRAM and a continuity module with the Pentium 4, but even this didn't help increase sales, and they ended up giving up this exclusivity by 2001.
Also, the reason Intel chose RDRAM for the Pentium 4 was due to its many architectural... Umm... Characteristics. Lol. Well. It was much more bandwidth-dependent than previous architectures due to a much deeper pipeline and a very fast but very small L1 trace cache system, so RDRAM's bandwidth advantages would keep the caches and the pipeline well fed. In fact, P4 shows a very mediocre performance when paired with SDR SDRAM. This of course was reversed since the introduction of DDR SDRAM.
Fun fact: Intel also supported RDRAM with some high end Pentium III chipsets. In fact, they were the only ones that supported a higher speed AGP bus version for a while, which of course made people really mad. Intel eventually also gave up here, and soon they released a high end chipset that supported traditional SDRAM, as well as this faster AGP slot. And, since P3 didn't have the same bandwidth hunger as its successor, the performance gains with RDRAM were minimal.
I had one of those Pentium 3 with 128mb RDRAM at my first workplace 15 years ago on an old Win NT computer. It even had support for 2 CPUs(only 1 installed). Sadly, we had to throw it away when it became obsolete because work.
Also there was money between the two companies that provided a purely non-technical motive.
@@jonsmith-z1v Yep, RDRAM mostly had high throughput for sustain serial reads IIRC, and for it's bit-width. And P4's could take advantage of that. However they also faired just as well with a ddr based MMC, better in some corner cases and better with rdram in a few more corner cases. Money changed hands or Intel would likely have only used RDRAM based P4's for specialty scenarios and big iron where they could put a lot of rdims together for a wider buss.
@@kaseyboles30 Uuhh... I think you meant P4s in this last comment? PIIIs never had a DDR MCC in their chipsets, as far as I remember...?
@@kaseyboles30 About your first comment, yes, totally.
The RDRAM on the N64 use a 9bit bus to save on traces even more (the extra bit is meant for parity, but the N64 use it for anti aliasing).
It is fast if you want to read just the next byte and the next byte etc.. you just go read-read-read-read and the memory has this internal address pointer thing that increase automatically.
But if you need to do random memory accesses, you need to change the internal address, and that takes like 4 writes, so it's very very slow.
And if you have the CPU and video chip fighting over the memory address, things gets very slow
N64 RAM is slow. AFIFAK N64 has ~0.5GB/s memory BW (same as Pentium 2 - could have 0.5GB/s with SDR-66 memory and 66Mhz FSB)
Those 2 came 1996 (N64, in US and JP) and 1997 (N64 in EU and Pentium 2).
Even worse, Pentium 2 has all that memory bandwidth for itself (without GPU) and it also has 256k-512k of L2 cache, 16k L1I, 16k L1D caches where N64 shares bandwidth with GPU, has no L2 cache and only has 16k of L1I, 8k of L1D caches.
@@volodumurkalunyak4651 for the time that's super fast, given the CPU runs at 93Mhz and the coprocessor at 62
However it rarely ran at 500MBs because of the adress changing hell
@@volodumurkalunyak4651 that's really fast for a 96Mhz machine, assuming you can reach that peak (you can't)
@@volodumurkalunyak4651 It was far worst than that. It 500MB/sec on paper/peak theoretical. In practice it could barely maintain above 100MB/sec.
It has been fun seeing all these home brew hackers figure this out in far greater detail that what was available in the 90's. Recently they have shown there are undocumented registers that show when the CPU/GPU are active or not. Turns out he CPU is idle about 70-80% of the time and the GPU is idle about 60-70% of the time. This is all due to it waiting on the RAM. Could have really used that info back then!
Most of the visual edge on the N64 was handled via additional core functionality and internal caches. Reality Engine and its functionality (Texture filtering, proper perspective calculation, actual understanding of depth via a Z-buffer!) was basically the only thing stopping the N64 dropping down to Ps1 performance.
@@dan_loup Is it just me that finds ironic that the console marketed as 64-bit had a 64-bit CPU cut down to have a 32-bit bus, and now I discover its memory had a 9-bit data bus? Lol.
I was calling around computer repair shops in town asking if they had any DDR4 RDIMMs in stock, so I could finish a job. One of them said "oh, no, they haven't made that for decades." I was flabbergasted, and argued with the guy until I found out he thought I was looking for Rambus memory. (FYI "RDIMM" is registered memory for some servers.)
I remember the DDR2 FBDIMMs for servers being dirt cheap. They didn't fit in consumer boards so they were absolutely worthless when they became professionally obsolete.
Exactly! RDIMMs or RDRAM were terms used to describe ECC Registered (buffered) memory mainly used in Intel Xeon based servers or 3D workstations. I assembled many Intel barebone systems based on that tech back in the days.
And rambus is now a patent troll
And have been for a long time
Im sorry but what’s wrong with protecting one’s intellectual property ? I mean I do think they should be forced to let others use it at a price but u shouldn’t be able to just rip off the IP
I remember Linus saying that in the Xbox Alpha I and II dev kit video.
@@MathCuriousityYou don't know what a patent troll is if you're taking their side, or you are one. A patent troll is someone who files patents on a avalanche of ideas having never lifted a tool to make any a reality, then sitting back and waiting for someone to come up with a similar idea without realizing it and slamming them with bank breaking lawsuits.
They do nothing but stunt the rate of technology by forcing you to consult an army of lawyers before doing anything, assuming you even can after the lawyers let you know if a patent troll is sitting on the idea.
@@MathCuriousityBecause it isnt their IP. They use loopholes to claim it is.
Oh. The good ole optiplex. They were ahead of their time back then.
GX200, still got it working here with 512mb RDRAM and Windows 2000
@@WackoX1337 I used one for deltaforce some twenty years ago and it worked fantastic!
Still got my trusty GX620. 😆
You could not run one C-RIMM and one RD-RIMM together. The sticks needed to be installed in pairs, and the C-RIMMs needed to be installed in any unpopulated sockets.
1:41 The N64's main bottleneck was actually the memory bandwidth, which got even worse due to the memory being shared between CPU and GPU (RDP)
still got a hand full of EDO, SIMM, and RAMBUS in my office lol Deal with a lot of manufacturing clients who have REALLY old machines. I even had a P3 and P4 Rambus boards as well (Not my daily drivers). Now i did see a bump in performance with it though.
Intel also made an RDRAM to SDRAM translator chip, used it at least on one Intel MB, which was later recalled.
DDR3 crew (with RTX GPU) 😂😂😂
Hmm.. Something like a Skylake/Kaby Lake on a DDR3 mobo - Turing pairing is definitely conceivable.
Personally, i7 4770k - RTX 2060 12Gb 😅
But in a lot of games my GPU and CPU can't reach 100% of usage, so I am assuming that it's my old DDR3 that can't keep up. I will soon move to AM5.
DDR2 if even longer
DDR if even longer longer
@@richieqs7789 DRAM if even longer longer longer
@@mozzjones6943 RDRAM if you are cool 😎
@@valtarijunkkala edo ram
RDR2
The fact manufacturers gave up on Red Dead RAMdemption is wild
ram bus goes vroom vroom
_a goomba gets yeeted offscreen_ "BEEP BEEP, MOTHER[TH-cam.]R!"
finally someone who knows!
I was around during the days of Rambus RIMMs. They sucked. I still have like 8 blanks in a box somewhere.
2:33 skip ad
COMMENT BOOSTER !!!!!!
This has to stay at top
Unfathomably based
when I was in highschool I had a p4 with 256mb RD ram. with an ASUS geforce 3 ti card. it was given to me by my friends neighbor who said it was out of an old alienware
Explanation of double data rate:
The clockspeed here refers to the wavesignals waveform, so from rising edge to another rising edge. As far as instruction cycles or logical cycles go the line needs to of course be read twice per each cycle. So in that sense its not double rate.
The advantage from using both the rising and falling edge to transfer information comes from how alternating current introduces design problems as the frequency rises. Impedance and phase shift are porpotional to the frequency, meaning it becomes the limiting factor. But if you use both the rising and falling edge to transmit data you still have to hit the same timings on low and high peak, meaning the second bit of information is ”free” as far as designing the bus lane goes. The devices still need to act upon the lane twice as many times as usual, else they couldn’t distinguish between rising + falling edge and no pulse. As far as I understood it they don’t get the second bit of information for ”free”. But that’s fine, because the long bus lanes and their signal intergrity is the limiting factor, not the devices at each end or their logical clock speeds.
Why did the name ram after dance dance revolution anyway?
That last line... bad guys who become less bad and finish... ... ... ... eventually! THAT was priceless! This def brought back some memories... thank you for this piece.
Ram bus lol more like ram bus
Rombus
Ram bus more like slow short bus.
I said what I said.
vrooom screech crash???
Into the building
I had RDRAM in my P4, in fact, I still have it somewhere. I even managed to upgrade it to 512mb a few years after buying the PC with 256mb. It cost an arm and a leg. But the PC was actually quite fast and I was very happy with it for years.
Can you download more RDRAM?
No, you can't and you need termination cards too
Yes but it's much more cost effective to download more CPU cores
You can try, but Rambus will probably sue you.
@@helamanmenendez2948 I just download a complete preconfigured machine. much easier. plug and play
I am a computer nerd, last night while in bed I recalled RAMBUS RAM & how we needed continuity module, wondered what happned to it, and here we are today....
I had a dell XPS desktop that had RAMBUS. It was so cool. My memory might be mixed up, but I think it was one of the few computers that operated the BTX case form factor which also didn't last long.
We need a video from you guys on ReRAM next. Interesting work going on from companies like 4DS and Weebit Nano at the moment.
2:12 my keyboard JUST died on me and i see this in the next video, maybe its a sign
Oh yeah, I picked up a used 1st gen P4 motherboard with RDRAM in about 2001 and that served as the core of my daily driver for ...5-6 years? I remember selling a couple sticks of RAM out of it when I got something newer because the prices on ebay were quite remarkable...and I was smart enough to not lose those continuity modules :-)
I had a customer come in asking us to fix their computer that had a Pentium ii and SDRAM 😅🤣 Ended up convincing them to upgrade, took SOOOOOO many attempts to pull data off the hard drive, the drive was just dying 🤣🤣🤣
back when we all ran xp ram worked entirely different. It was like a cars rpm it only ever ran when you hit the gas and opened a program. like computers idled at like 0-5% ram usage.
Thank you for the info...! I couldn't understand the whole video but I found it informative...!
0:17 Kaze Emanuar: RAMBUS goes vroom vroom
My first computer build was going to be a pentium 4.... but the implementation of RDRAM with the P4 was very 'bursty'. Very high performance for short stints of time, but poor long-term performance for things like rendering. As I was getting into video editing at the time, this forced me onto the previous gen P3 chips which didn't benchmark as well on synthetic tests, but all reviewers agreed had much better performance for sustained workloads.
I love to hear about retro tech and stuff like that
Most of my stuff is still DDR3. It's okay because I'm a retro sort of guy anyway. Yeah, I went on a bit of a used office-PC buying spree when third and fourth-gen Intel machines were the ones to get a while back. They still do their thing quite well. I remember when all the Rambus controversy was happening. I remember it was not well received on the tech sites I checked. Opposition to it seemed overwhelming from the user side. The nice thing from the user side is that AMD was there to go in their own direction so Intel and Rambus couldn't try to brute force this arrangement once it became clear that no one was happy. As I recall, I got a computer a little bit before, and shortly after all this, so I was able to sit the whole thing out and never dealt with Rambus stuff.
I used to have a Pentium 4 1.70 GHz with 256 MB RDRAM.
The memory was very fast for it's time. But the processor was quite slow and quite expensive.
I went to a recycling center and found an old P4 motherboard with RDRAM, 2x 128MB modules and 2x dummy dimms. Did some brief tests and the results are quite interesting. Sadly the motherboard only supports the early P4's, so no option to use the newer P4 HTs.
I still have an old 15+ year old server at work with RDRAM. 5 or so years ago I tripled its ram for like $150.
Remember RDRAM? YOU BET!
The N64 Expansion Pak did give you more Memory...which was 4MB. It did help improve graphics, but it was more for Donkey Kong 64 on memory leakage help.
Urban legend. It actually utilises the memory
Kaze Emanuar and his mario 64 mod uses Expansion pak, its pretty nuts.
It took _fast_ DDR3 to match or exceed the throughput of RD800. My RAMBus machines are my vavorites, easily. Pairing a GDDR3 AGP card with RAMBus 600 RAM kept my P.C. playing some of the newest and most intensive games on the market for _four years,_ and then only started a slow decline.
@@davidgoodnow269 it takes fast (dual channel of course) DDR2 to exceed bandwidth that dual channel PC1600 RDRAM ever had.
@@volodumurkalunyak4651 Um, no at the time. DDR2 800 didn't exist yet, not even in engineering concept, at that time. DDR3 was off the engineering development board and in production LONG before DDR2 got that fast. DDR2 overclocking eventually made DDR2 that fast commercially viable as a market, for us guys who couldn't afford a DDR3 motherboard!
Whereas, RD800 had more bandwidth than the early, slow, DDR3 that _was_ on the market at the same time!
@@davidgoodnow269 single channel RD800 and single channel DDR2-400 are equivalent, dual channel RD1600 and dual channel DDR2-800 are also equivalents in memory bandwidth.
RD800 comparable to DDR3 - no way.
Good luck finding dual channel rdram boards.
I had a slot card Pentium 3 with RAMBUS. I remember overlocking to 500mhz on it and being completely blown away by its mind boggling power.
Man those were the days
Nice
I love ram videos. Can you explain why ram signals can't be multiplexed or why a bus can't be optical where different interconnects can't use the same bus on a different frequency/phase rather than shutting off the bus while it's in use?
It's incredible to me how fast Ethernet has gotten so damn fast over the years by doing this, but ram takes massive power and specific slot placement, channels available depending on sticks installed, bios profiles and all that. It seems crazy to me how slow ram is tbh.
Also, why is ram built into the motherboard on high end boards? Solder 4gb of hbm next to the cpu and let sticks be added for when that mobo ram fills. Even if it only runs system and os, a level 5 cache measured in gigs would absolutely phenomenal even if only the os used it to help speed up the chipset io. Chipset hbm buffers would be amazing!
If you’re talking about straight up throughput, then yeah, ethernet is faster. But that’s where it ends. RAM is _orders of magnitude_ lower latency than Ethernet (a million times lower, to be exact). Ethernet latency is measured in milliseconds while RAM latency is measured in _nanoseconds_. Why is Ram situated as it is? Latency. At that scale of latency minute differences in the length of traces materially impact the latency of one bit vs the next. Traces need to be the same length to keep the latency consistent.
CPUs/GPUs need to switch between instructions far, _far_ faster than Ethernet could ever even dream of being able to provide the requested data for. Sure, compared to RAM, Ethernet is analogous to a fire hose vs a squirt bottle. However, try changing the direction of a firehouse as fast as you can a squirt bottle. All that throughput is useless if it takes comparatively forever to retrieve it.
Besides all that, Ethernet and RAM are two completely different things, they’re not interchangeable.
As for HBM with extra cache, that’s somewhat the direction things are going in, but HBM alone sends complexity through the roof, adding an external cache to the mix just multiplies the complexity to be (currently) out of the realm of practicality for all but the most demanding use cases. There’s a reason those high end boards you mentioned are high end: it’s not due to the presence of it, but rather due to the complexity required to have it there in the first place (and have it work in a beneficial fashion).
I wonder if a continuity module would make sense with DDR4/DDR5 for all of these memory busses that have reflection issues if you only populate the nearest slot, or apparently in some systems on new high speed memory, any empty slots.
DDR3 user here. I felt that comment xD but yeah if it wasn't for the end of windows 10 support i still wouldn't have upgrade plans xD.. i7 4770k @ 4.5 Ghz since 2013 xD
I once had to fix a PC which used Rambus in 2000. Can't remember what the issue was, but it was releated to Rambus.
00:06 'ddr3 if you haven' t gotten around upgrading your pc for a really long time'
😢 Yeah i know, i know 😢
The computer that i use in my job still uses ddr2, i wish i was lying (the processor was released in Q1 of 2009). We use windows 7, office 2013 and we can't even update chrome.
My first PC wad a custom made Dell PC with 800 MHz Pentium 3 and 128 MB of RDRAM. I wanted this expensive config (at the time) to play the newly released Unreal Tournament. Damn I feel old...
We still have our old Dell 8100 around here somewhere.
My Pentium4 Socket423 system uses RDRAM and it works great!
I remember it as REALLY REALLY FAST when it comes to loading games
i still use DDR3
Fancy, I'm still rock SRAM. (The S stands for SIMM for the kids who don't know)
@@GeorgeWashingtonLaserMusket I keep a ziploc bag full of them plus a couple sticks of DDR1. Not sure what I can do with them tho
Sell them at an antique auction
Yes me too! :)
@@unknown_to_everyone but... Why?
PS3 used Rambus XDR RAM as well.
You could explain Stt-mram , it's like DRAM but non volatile.
I remember telling a friend to invest in RamBus but his advisor told him no. But he put in a $1k anyway on my word. The next day Intel made its announcement, and the stock started soaring. He sold pretty quick and did make about three times his money. He called me, thanked me, and said he owned me lunch next time I was around. LOL.
Back in the day I met one person who used RD RAM. The high cost for the system was ridiculous and value must have decreased within days.
I had that ram in my high school desktop alongside a gforce card for smooth unreal tournament and cs 1.0
I remember RDRAM. Stuff got hot mate!
Got two of those pretty blue sticks of memory in my collection...😊
I remember just having DRAM. No DDR, no RD, not even SD.
I learned NOT to shred evidence while in a legal dispute with some of the largest companies in the world. WOW! The more you know.
Interesting history lesson. Pity about the audio background "earth" hum, very annoying and not used the LMG having this problem.
Ah, the times when i had a fulltime job but still lived at home with mom and dad. So i could spend all my money on hardware instead of kids and mortgage :)
Had a watercooled 2ghz P4 @ 2.4ghz and 768MB (256+256+128+128) of RDRAM. And 4 single platter 500gb HDD's in raid-0 for absolute peak performance.
Made absolutely zero difference in gaming but damn did i have fun wasting a ton of money on that old PC.
I still have my 768megs of RDRam around here somewhere.
Let's not forget the PS3, which has a the RDRAM successor and managed to get such graphics in freaking 256 MB only
Also, the first time I looked at it, it was $800!! BACK THEN. That was real money.
I remember when the first Pentium 4 systems with very expensive Rambus RAM came out and were not that fast for the money.
My friend when rambus realeased costs 3 times more than normal ddr,and works only with intel,this was one of the bigest monopoly scandals of intel.
There were p3 systems with rambus as well. I have one kicking around..
I don’t think RDRAM would have been the right approach for modern RAM. We’re seeing a trend towards more cores (wider CPUs), because frequency is falling off, and my suspicion is that RDRAM would have likely not had a huge speed advantage over current high end RAM approaches, but having a lot less bandwidth due to the smaller bus.
RD-RAM was a huge bottleneck on the n64 as well. The high latency caused a lot of problems for the system, and were a major contributing factor of why the system was difficult to develop for.
Worst choice I ever made,jumped all the way from our first windows 95 pc (pentium 100MHz) to a new Dell PC, Intel Pentium 4 CPU, RD RAM and Windows ME, the horror! Mercifully died after less than 5 years by which time laptops were practical for university and I think we were on xp by then
Rambus sounds like something you see on that black and yellow site
The latency of RDRAM also screwed N64, so much so it was often faster to pull data directly from cartridges instead of putting it to ram
Better than Dance Dance Revolution!? Preposterous!
I have a working pc with ddr2 ram
I could hear the music in my head from the blue man group p4 ad
Who says nice guys can’t finish first
- LTT
5:26
Damn... we were so close to excellence, and instead we went for bloody goddamn Dance Dance Revolution memory. Sheeeesh!!
Damn, is crazy how different things could have been
Going purely off stock prices ignores that you also would have gotten dividends. Those are actually (typically) the main reasons to invest in stocks.
Didn't Rambus showcases industry’s first DDR5 MRDIMMs & RDIMMs memory modules speeds Up To 12,800 MT/s just a few days ago this week ?
In 2004, Rambus sued the quartet of Samsung, Infineon, Hynix, and Micron for conspiring to low price fixing DDR RAMs that causes Rambus expensive solution to lose sales.
In 2005 Infineon settled with Rambus for $150 million while in 2010 Samsung settled with Rambus for $900 million.
Hynix and Micron won against Rambus in 2011.
I used the jumper pak as a security system. I would take it to school with me to keep my brother from playing lol
Litigious Rambus and Litigious Nintendo made quite a team
Offline store arounds me still selling ddr3 ram like they are new stuff. . 😂.
I built a PC with RDRAM back in the days. Back then, I fought it was quite superior. But damn it was expensive.
i have an OLD DELL with C-RIMM memory! that was a weird design
I am amazed that references to Nintendo and lawsuits made it into the same video without being related ...
I remember Rambus back in the day. They were way too expensive. Nobody ever bought them who actually worked for a living.
I had a pentuim 4 with this type of ram. I jumped ship, it was too expensive and quickly was phased out.
That is so cool that you can just add more RAM to the N64. Would be awesome if you could upgrade modern consoles like that.
Fun Fact (or not because it is not) = I have the same exact Memory DIMMs in my Storage Room, 64 MB Version and 128 MB too
Man ddr3 seems like yesterday I still have a gaming PC running sdram
I'm stuck on DDR3.
Joke's on you, I already knew about RDRAM
one thing I remember that was different between SDRAM and RDRAM was the heat spreading, SD ram would use each chip separately, filling one after the other, while RDRAM would use the whole bank filling them all evenly, which generated less overall heat per chip than using 1 chip after the other.
If that's wrong, I would love for someone to educate me :)
Germans found out DDR is so great they have DDR7
PlayStation 2 also used RDRAM.
"ddr3 if you didnt upgrade your pc for a long time" meanwhile me casually rocking 8gb of ddr2 960mhz ram in my pc