Intel CEO slams Rambus

[BFG10K]

The excitement over Rambus memory is justified.

No it's a total rip-off.

Here we have a memory design that can be used in REALLY high speeds, multiple channels, and will soon deliver about 12Gb/s. Shouldn't we be excited about this?

Pfff. I'd like to see that. All I see at the moment is 800 MHz RIMMS getting demolished by PC100 SDRAM.

RAMBUS is quite possibly the biggest technology hoax ever implemented.

 

[NOX]

<< Serial protocol RAM is doomed to fail. >>




<< With the advent of DDRAM, the latency will be even further reduced (and with DDRII even further). >>




<< I understand that higher speed DDR is soon to be availible in video cards soon to be followed by QDR and possibly DDRII versions of the highest end video cards. IMHO these are the memory standards of the future. RAMBUS will be a great joke told over technical discussions for years to come while SDRAM and its variants will continue to evolve. >>




<< DDR-II is serial, like RAMBUS. >>


So I guess DDR-II will fail, because it is like RAMBUS?

Damn, I need to stop visiting this forum, so much conflicting stores.

 

[NOX]

<< No it's a total rip-off. >>


It was a rip off (check PW)! Whats a rip off is a $500 dollor video card!

 

[OneEng]

If DDRII is serial, then it too will be doomed to fail. The latency of main memory will become a MUCH MUCH more important factor as processor speeds increase making the cache miss penality a growing percentage of CPU idle time (while it waits for the memory controller to return with its information).

I was not aware that DDRII was intended to be serial protocol RAM. There is really no way around it. Serial protocol RAM = high latency.

On a server application, this would make more sence. In a server environment, large volumes of information are being pumped from the mass storage devices out through the network connection. Large single accesses of memory is where serial protocol RAM really shines.

The physical issues with creating a much higher clocked serial RAM with a more narrow data path will still keep the price of any chip produced using this technology higher than an equivelent SDRAM technology part. The higher clock frequency will also translate into higher power requirements and more heat production.

The proprietary standard issue is a completely different discussion. Companies like RAMBUS that purchase patents and hunt for a worthy company to sue are simply bottom feeders. There should be a law inacted that prevents the existance of companies like RAMBUS. I simply can not express the extent of my distain for this kind of activity in the technology market. I am anxiously awaiting RAMBUS's demise. Watching their stock fall into the basement will give me great personal pleasure and re-assert my faith in the free market and those who support it.

 

[NOX]

<< Companies like RAMBUS that purchase patents and hunt for a worthy company to sue are simply bottom feeders. There should be a law inacted that prevents the existance of companies like RAMBUS. I simply can not express the extent of my distain for this kind of activity in the technology market. >>



I agree, especially if they had nothing to do with the sole development of SDR/DDR. I would like to see Rambus the company eat sh!t and die, but I believe that their technology as something to offer. Though I could be really wrong in believing so, I just don?t see any current problem with RIMMS latency issues. Maybe for large servers this is something to get fuming about, but in smaller servers, home workstation/PC?s, I just don?t see it being that big of an issue.

 

[Sohcan]

DDR-II is not a serial protocol, it's a packeted parallel protocol (Let me repeat, it's not serial). In fact, Enchanced DDR-II, a version with SRAM column buffers, will reduce latency to levels below those of SDRAM and DDR SDRAM. The white paper documents conservatively estimate that the decrease in latency will improve performance by up to 22%.

Look on page three of this DRAM white paper. The latency ranges for various types of DRAM are:

PC133 SDRAM: 22.5 - 66.7 ns
DDR SDRAM: 18.8 - 64 ns
DDR2 SDRAM: 17.5 - 42.6 ns
RDRAM: 35 - 80 ns

 

[PCResources]

<< Pfff. I'd like to see that. All I see at the moment is 800 MHz RIMMS getting demolished by PC100 SDRAM. >>



Well, you started this, so don't be offended.

So you checked out a piece of memory, did you, and you found it to be slow?

Well well, you really have no clue, the Alpha 21364 will implement everything i mentioned.

I know all of you have read the bad articles about RBDRAM, and i know all of you hate the company Rambus, but when you look into their memory design a little further, you will see that adding more memory channels is an easy task, with Sdram, DDR or whatever, it is not a hard task it is an IPOSSIBLE task. This makes Rambus a design of the future. Please come back when you have another memory design that works as well, delivering above 12Gb/s, thank you.

Patrick Palm

Am speaking for PC Resources

 

[PCResources]

<< But this proves it is certainly doable. It was reviewed @ Tom's a while ago. It would be really interesting if ALI was to apply this to their coming DDR chipsets. 4200mbps, low-latency memory interface, would be terrific. >>



This memory interface was not implemented between the high speed CPU-MEM inteconnects, wonder why??

A i have explained before, the pin count would be too high, the signal sensitivity would actually require the memory to be &quot;on chip&quot;.

Please check your facts.

Patrick Palm

Am speaking for PC Resources

 

[PCResources]

<< QDR is double pumped DDR. DDR-II is serial, like RAMBUS. The memory manufacturers are fighting over which one they want to use after DDR. Does that answer your questions, PC? >>



Let's get this straight once and for all, DDR-II is NOT SERIAL!!!!!!!!!!

I do not know from what source you got that piece of info, but i would suggest that you never trust that source again.

Check your facts.

Patrick Palm

Am speaking for PC Resources

 

[PCResources]

Okey, i'm back with some new info about &quot;qdr&quot;.

This design will not be made by any of the leading RAM makers.

The idea was a good one, but to implement the new form of circuits in real life would be to costly, the manufacturing process would have to change, the PCB layers would have to increase and the heat produced would be considerable.

So i guess we are down to DDR, DDR-II and Rambus now, Rambus memory can deliver 12GB/s if implemented correctly with multiple channels, and latency can be reduced by moving the interface on-chip. So what is your pick, the really fast one or the fast ones?

Patrick Palm

Am speaking for PC Resources

 

[urbantechie]

This is a good read! Keep it up guys! YOu guys are talking about stuff I don't know of! Whats Serial?

 

[SPAnDAU]

When you are using ram in multiple channels, don't you need a stick of ram occupying(sp?) each channel, or is that something done after the data is retrieved from the memory?