Rambus XDR in desktop graphics?

[Dave B(TotalVR)]

Just coz they tried to steal all the DDR

 

[Dave B(TotalVR)]

DRDRAM is actually not good at all for memory access granularity; it has a minimum burst length granularity of 8 transfers (=256 bits for a 32-bit wide interface), whereas e.g. DDR1 has a minimum of only 2. I suspect that XDR is even worse.

Hmm, but how big are the memory pages in rambus? and XDr fopr that matter.

You might require fewer I/O pins for an RDRAM interface than a DDR interface of similar bandwidth, but you need a great deal more shielding and power/ground pins, so the pin count/packaging cost benefit isn't really that big at all.

Does a single 32 bit rambus channel require the same more or less grounding pins than 4 64bit busses?

 

[KimB]

Hmm, but how big are the memory pages in rambus? and XDr fopr that matter.

I seem to remember the memory pages in Rambus being ridiculously small (a few bytes...but it's been a long time).

I think you need to explain yourself a bit better than that.
Have they committed some unforgivable sin?

They sued a number of memory manufacturers to receive license fees for SDRAM and DDR SDRAM (more than they charged for RDRAM, by the way), which are industry-recognized open standards developed by an open collaboration under JEDEC. Many manufacturers settled and paid the fees. I do not know how many still pay Rambus for the DDR that they make. Rambus is no longer a member of JEDEC due to this.

Some of the larger companies took Rambus to court over these issues, and Rambus has lost consistently.

So no, I don't care how good any of their products are. Their executives should be castrated, all employees fired, and the company ground to dust.

 

[arjan de lumens]

Does a single 32 bit rambus channel require the same more or less grounding pins than 4 64bit busses?

Somewhat less I guess, but such a comparison makes very little sense unless you actually expect the RDRAM to supply 8 times the bandwidth per pin as the fastest available GDDR3s. (1.1 Gbps/pin * 8 = 8.8 Gbps/pin, which is far, far beyond what XDR has been shown to be capable of)

 

[Xenus]

A typical 32-bit XIO cell takes approx. 130 ASIC package pins according to their website not sure how 4x64-bit GDDR3 takes though.

 

[vrecan]

Hmm, but how big are the memory pages in rambus? and XDr fopr that matter.

I seem to remember the memory pages in Rambus being ridiculously small (a few bytes...but it's been a long time).

I think you need to explain yourself a bit better than that.
Have they committed some unforgivable sin?

They sued a number of memory manufacturers to receive license fees for SDRAM and DDR SDRAM (more than they charged for RDRAM, by the way), which are industry-recognized open standards developed by an open collaboration under JEDEC. Many manufacturers settled and paid the fees. I do not know how many still pay Rambus for the DDR that they make. Rambus is no longer a member of JEDEC due to this.

Some of the larger companies took Rambus to court over these issues, and Rambus has lost consistently.

So no, I don't care how good any of their products are. Their executives should be castrated, all employees fired, and the company ground to dust.

I couldn't agree more!

 

[KimB]

A typical 32-bit XIO cell takes approx. 130 ASIC package pins according to their website not sure how 4x64-bit GDDR3 takes though.

144

The GDDR3 chips are also 32-bit each. 8 of them make up a 256-bit memory bus.

The problem with using Rambus instead of GDDR is simply that Rambus is made to be a serial design in many chips, but you really want a parallel design on a graphics card for maximal performance.

 

[nelg]

So no, I don't care how good any of their products are. Their executives should be castrated, all employees fired, and the company ground to dust.

Was there also not talk of how employees of Rambus attended JEDEC meetings to poach ideas and patent them?

 

[KimB]

Was there also not talk of how employees of Rambus attended JEDEC meetings to poach ideas and patent them?

I don't think so. I read that the main concern was that Rambus did not disclose their patents to JEDEC. They then pushed technologies for which they had patents.

Thus, in essence, they told people, "Hey, look at this! This is so cool, you should use it!" Then, a few years later, they mobilized a team of lawyers and said, "You used this technology without permission. This is technology developed and patented by us, and we demand royalties."

 

[Xenus]

I pretty sure that it is parallel but damnb it's hard to read through all of that why don't they just state such thing upfront.

 

[aaronspink]

I seem to remember the memory pages in Rambus being ridiculously small (a few bytes...but it's been a long time).

If by few bytes you mean a few thousand bytes.

Also the minimum granularity for DRDRAM is 16 Bytes.

They sued a number of memory manufacturers to receive license fees for SDRAM and DDR SDRAM (more than they charged for RDRAM, by the way), which are industry-recognized open standards developed by an open collaboration under JEDEC. Many manufacturers settled and paid the fees. I do not know how many still pay Rambus for the DDR that they make. Rambus is no longer a member of JEDEC due to this.

A little more background is in order... These companies knew of the Rambus technology under NDA before these companies put said technology into the JEDEC standard. The reality is that Rambus is as likely the victim as the dram companies.

Some of the larger companies took Rambus to court over these issues, and Rambus has lost consistently.

This is actually incorrect.

So no, I don't care how good any of their products are. Their executives should be castrated, all employees fired, and the company ground to dust.

hmm, maybe you should spend a little more of that vitriol looking at the dram companies instead of focusing it on rambus.

Aaron Spink
speaking for myself inc.

 

[aaronspink]

Does a single 32 bit rambus channel require the same more or less grounding pins than 4 64bit busses?

Somewhat less I guess, but such a comparison makes very little sense unless you actually expect the RDRAM to supply 8 times the bandwidth per pin as the fastest available GDDR3s. (1.1 Gbps/pin * 8 = 8.8 Gbps/pin, which is far, far beyond what XDR has been shown to be capable of)

In the real world it works out to about 4 DRDRAM channels per DDR DRAM channel, at least in the server space.


Aaron Spink
speaking for myself inc.

 

[Xenus]

Toshiba has a 76 page pdf on XDR for anyone who is brave enough to try to decipher it.

 

[aaronspink]

A typical 32-bit XIO cell takes approx. 130 ASIC package pins according to their website not sure how 4x64-bit GDDR3 takes though.

144

Um, I fail to see how the number of balls on a dram chip package has any relation to the number of pins required at an interface for a 4x64b GDDR3 interface.

The data bus alone will require 256b of data pins. 3 pins for clocks. Then per bus: CS, RAS, CAS, WE, BA0, BA1, A0:A11, RDS0:RDS7, WDS0:WDS7. 34 Control pins per bus * 4 = 136 pins in control. 2 P/G pins for every 8 data pins -> 64 power ground pins (minimum).

Total pins: 256 + 136 + 64 -> 456 pins.

The problem with using Rambus instead of GDDR is simply that Rambus is made to be a serial design in many chips, but you really want a parallel design on a graphics card for maximal performance.

This is incorrect. And simple not true in relation to XDR.

Aaron Spink
speaking for myself inc.

 

[{Sniping}Waste]

I seem to remember the memory pages in Rambus being ridiculously small (a few bytes...but it's been a long time).

If by few bytes you mean a few thousand bytes.

Also the minimum granularity for DRDRAM is 16 Bytes.

They sued a number of memory manufacturers to receive license fees for SDRAM and DDR SDRAM (more than they charged for RDRAM, by the way), which are industry-recognized open standards developed by an open collaboration under JEDEC. Many manufacturers settled and paid the fees. I do not know how many still pay Rambus for the DDR that they make. Rambus is no longer a member of JEDEC due to this.

A little more background is in order... These companies knew of the Rambus technology under NDA before these companies put said technology into the JEDEC standard. The reality is that Rambus is as likely the victim as the dram companies.

Some of the larger companies took Rambus to court over these issues, and Rambus has lost consistently.

This is actually incorrect.

So no, I don't care how good any of their products are. Their executives should be castrated, all employees fired, and the company ground to dust.

hmm, maybe you should spend a little more of that vitriol looking at the dram companies instead of focusing it on rambus.

Aaron Spink
speaking for myself inc.

You need to do some reserch.
Rambus just stole the tech that the others companys shared to JEDEC and not sign to join JEDEC at the last minute. They patented the tech that was shared in JEDEC. I say its thift in my book. There are 2 companys left that are still fighting rambus and its Hynix and Micron. For Micron, The US Goverment will not let Rambus sue Micron for patents still and see Rambus is in the wrong but in other countrys, Rambus was able to sue other companys. The last big case was Infinion and if I remember right, Infinion won that battle.

JEDEC learned a lesson from this mess. Companys need to sign first before getting invoved in JEDEC because theres always someone that will stabe you in the back.

 

[Xenus]

Oh I forgot the link

http://www.toshiba.com/taec/components/Datasheet/59YM916BKG_041215.pdf

There you go.

 

[aaronspink]

You need to do some reserch.
Rambus just stole the tech that the others companys shared to JEDEC and not sign to join JEDEC at the last minute. They patented the tech that was shared in JEDEC. I say its thift in my book.

Nice memory company PR. MAYBE YOU SHOULD DO SOME RESEARCH. Learn a little bit of the background. Some of the things that went on behind the scenes. The fact that all the memory companies were under NDA with Rambus and encouraged rambus to join JEDEC. That there was a decent amount of conspiricy going on among the memory makers. Or the fact that Rambus never voted on anything they are suing over.

There are 2 companys left that are still fighting rambus and its Hynix and Micron. For Micron, The US Goverment will not let Rambus sue Micron for patents still and see Rambus is in the wrong but in other countrys, Rambus was able to sue other companys. The last big case was Infinion and if I remember right, Infinion won that battle.

Infinion won the initial court case which was overturned on appeal.

The US government doesn't care about Micron/Rambus and patents. Micron at one time abused their government conections to get the FTC to investigate rambus but rambus overcame that issue which was orthogonal to the patent suits.

Aaron Spink
speaking for myself inc.

 

[Guden Oden]

That there was a decent amount of conspiricy going on among the memory makers. Or the fact that Rambus never voted on anything they are suing over.

I recall reading some years ago now that rambus consistently amended their patents and backdated the amendments to the original filing date after certain specifics of DDR had been discussed in JEDEC meetings, thus patenting these specifics. Of course not disclosing what they were doing either...

 

[karlotta]

That there was a decent amount of conspiricy going on among the memory makers. Or the fact that Rambus never voted on anything they are suing over.

I recall reading some years ago now that rambus consistently amended their patents and backdated the amendments to the original filing date after certain specifics of DDR had been discussed in JEDEC meetings, thus patenting these specifics. Of course not disclosing what they were doing either...

Your both right.