Rambus is up to their old tricks again

[ST4RCUTTER]

Am I incorrect in my assumption that patents must be 'filed' and documented? Otherwise they are not patents. If they were filed and correctly patented, then Rambus would have 0% obligation to notify other memory manufacturers of the infringments. Patent review would fall squarly on the shoulders of those who whould be desinging DDR without consulting existing memory architecture patents...

Dullard has it right. RAMBUS joined the JEDEC and like all other companies that did, they were required to disclose any current or pending patents so that they would not be incorporated into industry standards. RAMBUS failed to do this, and documents disclosed at more than one trial indicate that this was intentionally done to defraud the other memory manufacturers and pave the way for "profit through litigation."

 

[Diable]

<< If you have a patent, and are asked if you have a patent then legally you must answer yes. Rambus answered 'no' - causing other companies to invest time and money in this technology thinking it was an open standard. Rambus had the patent the whole time and did nothing to stop them from infringing on it. That is where Rambus went wrong (legally). >>



If thats what Rambus did then their wrong but the other memory makers should have checked with the US Patent Office too. I can't see the legal departments of these multi billion dollar companies just taking Rambus' word that they didn't have the patents. If thats the case we should dis them for their stupidity just as we dis Rambus for their litigiousness.

 

[kgraeme]

<< If thats what Rambus did then their wrong but the other memory makers should have checked with the US Patent Office too. I can't see the legal departments of these multi billion dollar companies just taking Rambus' word that they didn't have the patents. If thats the case we should dis them for their stupidity just as we dis Rambus for their litigiousness. >>



Rambus did have the patents. They took ideas from the JEDEC meetings that were designed to create an industry standard, went home and patented them, didn't tell JEDEC members, and then proceeded to sue JEDEC members for using the industry standard memory tech that they had agreed on.

As a member of JEDEC, Rambus was required to inform other JEDEC members of outstanding patents in order to explicitly avoid having lawsuits. However, Rambus saw the opportunity to litigate their way to success.

 

[kgraeme]

<< So your saying that the screwy ideals of some dried up old men stop you from using it, even though they have an all around competitive product in terms of price AND performance? >>



Absolutely. The means don't justify the ends.

 

[BD231]

<< So your saying that the screwy ideals of some dried up old men stop you from using it, even though they have an all around competitive product in terms of price AND performance? >>



Thats right. What's your point?

Seems retarted to limit yourself like that....

 

[ViRGE]

<<

<< So your saying that the screwy ideals of some dried up old men stop you from using it, even though they have an all around competitive product in terms of price AND performance? >>



Absolutely. The means don't justify the ends. >>


Agreed. As geeks, we'll love any company that provides us with good parts and is good to us as people(Nvidia). Along the same lines, we'll even love companies that fail to give us good parts, but still is good to us(Matrox, ATI). However, the converse doesn't work; if you give us good parts but treat us(or our industry in this case) poorly, we'll be glad to return a favor.:disgust: Unlike Rambus, we have hearts and brains far larger than our Athlon dies; something that Rambus can't say the same of.

 

[Diable]

Kgraeme, your still over looking the fact that the other JEDEC member companies didn't double check if Rambus was telling the truth or not. Would you hire someone solely on their word that they worked for company X or would you check to see if they worked for company X before you gave them the job? IMO the other JEDEC member are being punished for not going there jobs.

 

[ST4RCUTTER]

...your still over looking the fact that the other JEDEC member companies didn't double check if Rambus was telling the truth or not. Would you hire someone solely on their word that they worked for company X or would you check to see if they worked for company X before you gave them the job? IMO the other JEDEC member are being punished for not going there jobs.

How can the JEDEC "double check" pending patents that have not been issued? RAMBUS did not have to join the JEDEC but they did. They signed the agreements that all the other memory manufacturers did, and they failed to comply. Not only that, internal RAMBUS documents show that RAMBUS used the meetings to formulate a strategy whereby they could guide the open-standard process towards their patents...some may call that successful business warfare, IMO and to many others this should result in punative action in civil court.

And in at least one case it has...

 

[cmdrdredd]

why don't I want to install in pairs...here's why. say I want 512Mb of RAm I have to buy 2 sticks of RDRAM. That means I'm paying more than 1 stick of 512Mb DDR. So That's why.

 

[microAmp]

I long for the day when their stock goes under $1.00/share. I'm sick of them and all their lawsuits.

 

[kgraeme]

<< some may call that successful business warfare, IMO and to many others this should result in punative action in civil court.

And in at least one case it has... >>



Don't forget that they were/are sued by their own stockholders.

 

[SexyK]

<< why don't I want to install in pairs...here's why. say I want 512Mb of RAm I have to buy 2 sticks of RDRAM. That means I'm paying more than 1 stick of 512Mb DDR. So That's why. >>



Pricewatch:

2x256MB PC800 - $122
1x512 PC2700 DDR - $122

Kramer

 

[MWWInc]

<< Would you hire someone solely on their word that they worked for company X or would you check to see if they worked for company X before you gave them the job? IMO the other JEDEC member are being punished for not going there jobs. >>



They didn't rely on RAMBUS' word, they relied on the legal agreement RAMBUS signed not to steal their ideas. This is common business practice, as much less would get accomplished without information sharing, trust, and a legal contract to hold those that break that trust accountable.

On a side note, my issue with having to buy RDRAM in pairs would be more of an upgradibility nuisance than a cost problem. For example, if I had 2x128MB and wanted to upgrade to 512MB of RAM, I'd have to get 2x256MB sticks and sell off my previous 2 sticks.

 

[ElFenix]

rambus broke a contract. why can't you understand that?!?

 

[dullard]

<< I want 512Mb of RAm I have to buy 2 sticks of RDRAM. That means I'm paying more than 1 stick of 512Mb DDR. >>


Since you only need to buy two cheaper 256 MB sticks, the overall price is nearly identical as SexyK pointed out. So you are flat out wrong in this example. Note: SexyK used PC2700 prices, buying PC2400 would save you a measly $1, and buying PC3200 would cost you $68 MORE for DDR.



<< On a side note, my issue with having to buy RDRAM in pairs would be more of an upgradibility nuisance than a cost problem. For example, if I had 2x128MB and wanted to upgrade to 512MB of RAM, I'd have to get 2x256MB sticks and sell off my previous 2 sticks. >>


Virtually every RDRAM motherboard has 4 or more slots. You will keep your 2x128MB in the computer and buy an additional 2x128 sticks for a total of 4x128=512. So you too are wrong (in this example).

Anyone care to come up with a better arguement?

See here for the speed difference between DDR and RDRAM. The upcoming PC1066 can easily best the speed of any DDR available and ones not yet available.

 

[JellyBaby]

I'm definitely not a Rambus fan but I wonder if the anti-Rambus vibe is overboard. Microsoft has committed far greater atrosities in its history yet people still flock to their products. I'd never put RDRAM in my machine unless I'm forced to but given the business practices of other companies, Rambus is a young innocent.

 

[MWWInc]

<< Virtually every RDRAM motherboard has 4 or more slots. You will keep your 2x128MB in the computer and buy an additional 2x128 sticks for a total of 4x128=512. So you too are wrong (in this example). >>


Populating all 4 banks while OCing does not a happy computer make. And most people here don't fault RAMBUS' technology, just the fact that they stole the ideas it's based on. As far as the article you referenced, interesting read about future chipsets and memory technology, but like all things tech, we'll have to wait and see. Intel has a lot of sway with memory makers now, whereas AMD's procs are more FSB than memory limited.

 

[MrDudeMan]

its kinda funny....ram is so fast now that it doesnt even make a difference

OMG MY RAM CAN DO .1mbps MORE THAN YOURS IN A BENCHMARK!

have you ever tried using it in a game? as long as your computer doesnt blow goat ass, its all the same when you have a decent graphics card

that is...unless some of you feel the need to play a game on 1600x1200 which has no practical use

 

[MWWInc]

No practical use except to make the game look better...

But your point is good in that reviewers do tend to use really low graphics settings when testing a proc/RAM combo, and when you jack up the settings to what you normally play it, your video card will be the limiting factor. (Although some of the very newest games like Comanche 4 *are* proc limited even with a Ti4600.)

 

[Buddha Bart]

dexvx:
Latency is a big matter on RAM. Because rambus is in serial, the more ram you have the overall latency of it increases. Thats why its NOT good to be used in server applications that demand a huge amount of ram.

zemus:
This Fud is still floating around, my god.

come on people, latecny does not magically increase because you ask your ram for more data/second.

zemus: the latency of RDRAM does in fact increase as you add greater amounts. Because the bus is serial there are timing issues involved in getting a signal from one end of the bus, to the end, and back again. For this reason, when first booted, each chip on the bus calculates its offset from the slowest chip, and automatically adds that latency to its operation (aka waits that long), so that everything stays in sync. There is a limit of 3 RIMM's to a rambus channel, and a max number of rambus modlues of 32. However since most RIMMs have 8 modules, you would be hard pressed to break 24. In addition I've never seen an implimentation that had more than 2 RIMM banks per chanel.
Simply put, rambus does not scale well to the gobs of ram that larger servers need. And when it tries to, the latency (performance) starts to get quite miserable.

here, link-a-tollah
http://www.realworldtech.com/page.cfm?AID=RWT110799000000
http://www.realworldtech.com/page.cfm?AID=RWT112299000000
http://www.aceshardware.com/read.jsp?id=5000172
http://www.aceshardware.com/read.jsp?id=5000183

bart

 

[kgraeme]

<<
Virtually every RDRAM motherboard has 4 or more slots. You will keep your 2x128MB in the computer and buy an additional 2x128 sticks for a total of 4x128=512. So you too are wrong (in this example).

Anyone care to come up with a better arguement?
>>



Over the years I've worked on systems that required pairs and systems that required singles. I can tell you in almost every case with the system with pairs that the customer will fill all the slots with the smaller memory because it's cheaper, then when they have to upgrade they end up having to toss out ram which really pisses them off. And remember, that Rambus memory used to be expensive back when the customer originally bought the system. And it's not just Rambus, all memory costs more when it's new. Here's a breakdown of how the customer sees it:

Rambus system

Starting system purchased in 2000: 128MB (which seemed really cool like a ton of memory to people used to 32.)

pair1: 64+64 $600 (remember, this is back when Rambus was new)
pair2: none

Upgrade: 256MB (which seemed like it would be more than enough)

pair1: 64+64
pair2: 64+64 $200 (getting cheaper)

Upgrade: 384 (Installed WinXP and OfficeXP and wanted more speed out of their 1GHz machine)

pair1: remove 64+64, add 128+128 $150 (the customer weeps at the original $600 cost for those two sticks of 64.)
pair2: 64+64

Total cost of memory over time just to reach 384MB: $950

DDR system

Starting system: 128MB (which seemed really cool like a ton of memory to people used to 32.)

slot1: 128 $130 (remember, this is back when DDR was new)
slot2: none
slot3: none

Upgrade: 256MB (which seemed like it would be more than enough)

slot1: 128
slot2: 128 $150 (DDR actually went up in price)
slot3: none

Upgrade: 384 (Installed WinXP and OfficeXP and wanted more speed out of their 1GHz machine)

slot1: 128
slot2: 128
slot3: 128 $43

Total cost of memory over time just to reach 384MB: $323

Yes, some of the cost difference had to do with the insane Rambus prices for a while, but that is what really happened. And my DDR system had 3 slots, and some people might only have 2. But when you only have to buy one, the cost for a larger stick isn't as bad as when they have to be bought in pairs.

If a person were building a system today and starting at the , the costs would certainly look different. But as I said, in my experience of working with customers and computers, people will always buy the smallest amount at each interval and end up paying more in the end and with pairs that always works against the customer more than single sticks.

 

[zemus]

---

zemus: the latency of RDRAM does in fact increase as you add greater amounts. Because the bus is serial there are timing issues involved in getting a signal from one end of the bus, to the end, and back again. For this reason, when first booted, each chip on the bus calculates its offset from the slowest chip, and automatically adds that latency to its operation (aka waits that long), so that everything stays in sync. There is a limit of 3 RIMM's to a rambus channel, and a max number of rambus modlues of 32. However since most RIMMs have 8 modules, you would be hard pressed to break 24. In addition I've never seen an implimentation that had more than 2 RIMM banks per

---

Oh I knew that one, I was refering to mis-information on latency dynamically increasing due to heavy usage ( High bandwidth )

 

[Miramonti]

<< RDRAM is good, but it must be repeated: RAMBUS IS SUCK. >>

 

[Buddha Bart]

, I was refering to mis-information on latency dynamically increasing due to heavy usage

actually, thats true as well.
Both RDRAM and SDRAM can only keep so many (more for rdram) bank's open at once. During heavy memory useage when a bank that is not currently open needs to be so, the open bank has to be written out, and the newly open bank has to be 'precharged'.
This is known as the PC (precharge) latency, and can be seen as the third number when you see ram latency denoted (2-2-2).

The more heavily saturated a bus, the more often that occurs, therefor the worse its latency gets.
In some aspects this is actually less of a problem for rdram, as it can have more banks open at once. But there are also other issues like what state the chip of the newly requested bank is in (active/standby/nap/sleep).

bart (i'm actually supposed to be writing a paper on this stuff right now... instead i post in forums.... so wish my professor would just read this stuff)

 

[Zakule]

<< That does not make a whole lot of sense. The reason RDRAM achieves it's bandwidth is because of the 16-bit bus. I'd rather it stay that way. As for the latency, that is bull. PC1066 RDRAM Latency on 850 is on par with PC2100 on 845-D. >>



Not true. I'll explain in laymen's terms!

The latency issue is what ruins RDRAM for current multimedia platforms. It has excellent overall bandwidth as demonstrated by countless benchmarks, but ultimately latency is more important then bandwidth for multimedia performance. That is the bottom line. When rendering 3D graphics it's much more important for your memory to have a quicker transaction time then to be able to stream x GB/s. It may not always be that way, but it is that way now. The majority of frame rendering takes place on the video card where the textures are stored in video memory. The only data the video card needs from CPU/memory is the parameters on how it is to paint the textures. This is a relatively small amount of data compared with the total bandwidth offered by either RDRAM or DDR. This fact is what makes technologys like AGP4x a joke since applications rarely if ever use the AGP function (in its entirety) and to this day don't fully utilize the bandwidth of AGP2x.

To reiterate, the high bandwidth offered by RDRAM doesn't really matter because the latency on transaction times is too high. Again, it may not always be this way, but as long as game designers continue to write their games to work with video card centric frame rendering vs. CPU to Memory to Video rendering (which is just way too slow at present) then latency is going to play a much bigger part then excessive bandwidth that's not even being used. The longer that the CPU has to wait for a memory transaction to finish the longer the video card has to wait for the parameters it needs to lay the textures and this means reduced performance. Being able to stream 4 GB/s through memory when each request is several clock cycles behind means squat when the video card needs less than 200 MB/s on average. Getting the data to the video card ASAP after it's requested is what's important.