Rambus Question

[PliotronX]

Serial benefits speed greatly and lowers the amounts of traces needed. The AMB is a hotfix solution. Also with DDR4 for example, you only have 1 DIMM per channel. Unless you add a switch.

About everything else than the memory is serialized now.

I understand ya, I'm all for competition and innovation but the way RAMBUS went about it rubbed me the wrong way. It seems to me, and I agree with the cartel sentiment, that it's an easy path to evolve the memory market when a brickwall is hit with parallel.

 

[imagoon]

They stole what? Rambus designed their original syncronous rdram circa 1990. Jedec designed nothing. And of course Rambus get royalties from all DDRx versions sold today as they are clearly just different implementations of their memory design. Jedec tried to stole all Rambus patents and implement them one by one for free but failed.

And drdram from late nineties is about same as we got now as DDR3. Similar data rates and address signals are routed similarly from mb to dimm itself unlike older sdram-variants where address channel stays at mb, clearly there was some real geniuses at Rambus and what they implement was what everybody have to do with higher frequencies. Remember that we got similar speed memory what we use now fifteen years ago thanks to Rambus but somehow we settled down to slower memory types for decades, are we stupid or is there some big money to make decisions for us......

Well for #1, the big players settled. That isn't "guilt." They simply didn't want to deal with the lawsuit happy Rambus. In May of 01 they were found guilty of fraud for claiming that they owned SDRAM and DDR and all infringement cases were dismissed against the memory manufactures.

Jan 05 they had their case against Infineon dismissed. Infineon decided it was cheaper to just settle than go through another multiyear case and appeals.

Jul 07, EU launched antitrust investigations against Rambus because they considered Rambus deceptive. They declared it "patent ambush" and it is still pending.

Nov 2011 Rambus lost its case against Micron and Hynix

Jan 2012, USPTO overturned one of their 3 key patents that were already used to win litigation previously.

FTC brought Anti-trust actions against Rambus that was thrown out and brought back on appeal.

Sound like your typical patent troll. However the EU and the FTC ended up setting their royalty limit to 1.5% for 5 years back in 2007 and was then set as low as .25 for SDRAM and .5% for DDR. The fact that the royalties are so tiny, generally indicates to me at least that they didn't really add much to it.

Also while RD-RAM's theoretical data rate was higher than DDR, it was crippled by the fact that it wasn't needed by the processor and the poor latency (45ns + additional time for each additional chip in the chain).

 

[ShintaiDK]

I understand ya, I'm all for competition and innovation but the way RAMBUS went about it rubbed me the wrong way. It seems to me, and I agree with the cartel sentiment, that it's an easy path to evolve the memory market when a brickwall is hit with parallel.

You might take a note that the DRAM manufactors got fined a billion $ plus several executives jailed. Then think about who stood in the way for innovation for self interrest and profit.

The DRAM chartel destroyed what would be RAMBUS primary business. No wonder the following ytears went as they did.

 

[PliotronX]

You might take a note that the DRAM manufactors got fined a billion $ plus several executives jailed. Then think about who stood in the way for innovation for self interrest and profit.

The DRAM chartel destroyed what would be RAMBUS primary business. No wonder the following ytears went as they did.

No doubt the technology has been stagnant no matter what shiny heatspreaders are placed onto them but Rambus had good ideas at one time and even won major contracts for the PS2 and N64 but their primary business appeared, from the outside, to be revenue through litigation. They also violated the nondiscriminatory clause when they joined the JEDEC. They could have played it out better and totally broken the cartel as you describe but ended up botching it.

 

[naukkis]

Also while RD-RAM's theoretical data rate was higher than DDR, it was crippled by the fact that it wasn't needed by the processor and the poor latency (45ns + additional time for each additional chip in the chain).

There was nothing wrong with rdram. Actually it's very well engineered for minimum latency and maximum bandwith efficiency. There ain't any unnecessarily latency involved, only that 1 or 2 ns related to little bit longer address and data path chain but that is only way to make such a high speed busses. DDR3 has exactly same address bus chaining type as drdram and have similarly longer latency for last chip of chain.

Actually drdram was well ahead latency wise all sdram variants, every memory chip has 32 independent pages to keep open so with eg 16 memory chips sdram has possibility keep 8 pages open versus 256 pages drdram. With open page hit memory latency is much lower than miss. At 90's there was no memory controller made which could really support what drdram was cabable of, for example intel I850 was cabable of keeping 8 pages open.....

And there was so much more engineered to rdram which was absent from Jedec-memories, for example memory controller and memory chips load capasitance was adaptive so you could mix and match different manufacturers memory modules without problems, something that really good memory standard need but what newer wasn't jedec priorities, threw away you old memory for compatibility sake instead......

 

[naukkis]

If they're so genius, why don't they come up with fresh ideas and designs instead of going around suing larger companies? That is not innovation.

Need more memory bandwith? For few years Rambus has got working solution up to 20 gbps, low power, could use more than just one dimm per channel etc. Why we don't use it? If you think it's about Rambus royalties think again. Instead we got soon big companies designed up to 4 gbps DDR4 memory which don't violate Rambus newer and better patented memory techologies... But big companies rule at memory manufacturing, consumers get what they want us to have.

 

[zir_blazer]

Need more memory bandwith? For few years Rambus has got working solution up to 20 gbps, low power, could use more than just one dimm per channel etc. Why we don't use it? If you think it's about Rambus royalties think again. Instead we got soon big companies designed up to 4 gbps DDR4 memory which don't violate Rambus newer and better patented memory techologies... But big companies rule at memory manufacturing, consumers get what they want us to have.

And is that RAM technology affordable enough to manufacture on mass-scale to supply the market? One of the reasons why RDRAM was unpopular was because it was hard as hell to manufacture, yields were extremely low.


BTW, the way you defend Rambus makes me feel like if you were a stock holder. Rambus was pretty much born as a patent troll company, and the way they behaved was rather funny: They quitted JEDEC, left two moles inside, and as JEDEC prepared the SDRAM standard, Rambus made an amend on one of their patents to be able to claim royalties from that.
http://www.simmtester.com/page/news/showpubnews.asp?where=5795186&num=53
http://www.geek.com/chips/rambus-spy-at-jedec-543079/
http://www.theregister.co.uk/2001/04/06/rambus_received_leaked_jedec_sdram/

 

[imagoon]

There was nothing wrong with rdram. Actually it's very well engineered for minimum latency and maximum bandwith efficiency. There ain't any unnecessarily latency involved, only that 1 or 2 ns related to little bit longer address and data path chain but that is only way to make such a high speed busses. DDR3 has exactly same address bus chaining type as drdram and have similarly longer latency for last chip of chain.

Actually drdram was well ahead latency wise all sdram variants, every memory chip has 32 independent pages to keep open so with eg 16 memory chips sdram has possibility keep 8 pages open versus 256 pages drdram. With open page hit memory latency is much lower than miss. At 90's there was no memory controller made which could really support what drdram was cabable of, for example intel I850 was cabable of keeping 8 pages open.....

And there was so much more engineered to rdram which was absent from Jedec-memories, for example memory controller and memory chips load capasitance was adaptive so you could mix and match different manufacturers memory modules without problems, something that really good memory standard need but what newer wasn't jedec priorities, threw away you old memory for compatibility sake instead......

Out of all of that, RDRAM still had the poorest latency per read request. Even with the highest 20ns chips, you needed to chain through all of the chips in the machine. With 8 chips per RDIMM + at least 2 RDIMMs the performance would suffer going through all the chips, more so as you added more chips and larger capacities. SDRAM did have worse latency but as the clocks (and virtual clocks) increased on DDR DDR2 and DDR3 the performance simply ran past RDRAM and was cheaper over all.

As you mention DDR3 uses the same addressing design, DDR3 may chain addressing but that is only part of the latency. DDR3 still has parallel data and strobe back to the controllers rather than streaming the data back through all the chips. RDRAM would have needed to ramp up the clock rates or start using multiple channels like DDR+ has been doing to even begin to touch.

This also ignores sequential vs random access. RDRAM can be incredibly fast sequentially because it has built in provisions to do something like: "Hey you know that address I used last time? Get me the next block." This isn't really common in RAM (especially now with VM's etc.) In most real world tests, DDR, even with its lowly 133mhz clock (double pumped to 266mhz" simple could return random data at a faster rate with a lower latency that RDRAM. There even was some good tests on Tom's at one point that showed a single 512MB DDR DIMM typically out performed 4 128MB RDIMMs by nearly 20% and was about equal to 2 x 256MB RDIMMs.

Real world, RDIMM were expansive and didn't live up to promises. Hence the market went else where.

 

[imagoon]

Need more memory bandwith? For few years Rambus has got working solution up to 20 gbps, low power, could use more than just one dimm per channel etc. Why we don't use it? If you think it's about Rambus royalties think again. Instead we got soon big companies designed up to 4 gbps DDR4 memory which don't violate Rambus newer and better patented memory techologies... But big companies rule at memory manufacturing, consumers get what they want us to have.

20gbps serial compared to a 128bit (channel) 4gbps? DDR4 is already benchmarking at 51gigabytes per second to a single DIMM without any dual, triple or quad channel tricks. It is also doing this using less power and outputting less heat than DD3 (46% last I have seen out of Samsung). This is just under double the raw performance of Rambus XDR designs with a significantly lower power envelope. XDR2 design might beat that but since no one has bothered to buy the rights for or fab that yet, I'll take what is on the market, in production vs a drawing on a Rambus table.