Nvidia not going to sign settlement agreement with Rambus
About time someone stood up to these leeches.
About time someone stood up to these leeches.
- Status
FrameBuffer
Banned
Agreed,. now if only NV would be forced to release CUDA/PhyX source code and all documentation so Intel/ATI/S3(Via) and whomever can freely use/develop Cuda/PhyX without any cost associated with licensing.
aaronspink
Veteran
Who are the leaches, the ones stealing the tech and not paying for it? And FYI, as soon as the ITC enforces the ruling, Nvidia with cave instantly or go bankrupt. The next step for Rambus is to permently injunct all imports for Nvidia, and bye bye nvidia.
I don't really agree, since if GDDR stalls, it'll probably be AMD's fault. They're the biggest contributer to the spec (and they chair the JEDEC committee for it) and they're responsible for its development by and large. That means it's in AMD's (the GPG/ATI) best interest to maintain their competitive advantage that comes from their involvement.
Note that AMD got to GDDR5 via 4, which they had a big hand in too. It wasn't anywhere near as conservative a transition as NV (not that I believe it was conservative for NV on their own terms either, since there's a lot of politics behind graphics DRAM traditionally).
So as we push to 6, I don't expect development or the technology to stall, and I expect AMD will keep their technological lead when implementing it in consumer designs. They'll have a bandwidth per pin advantage for some time, IMHO. Indeed, having that advantage drives their current products quite visibly.
Even if NV were forced to open it enough to let ATI or Intel or whoever write a driver and participate in CUDA's evolution, what's the advantage to ATI or Intel or whoever to do that now that we have a committee-driven alternative?
Actually I was thinking in terms of copper's limit itself. I do not know of any interconnect which can do >10gbps per pin on copper apart from rambus itself. USB3 was supposed to be optical, Light peak is optical, 10GbE is optical.
As you correctly pointed out, AMD's advantage in bandwidth per pin helps their products a lot. So may be just to preserve that edge, AMD is a more likely migration candidate.
Or just may be, there is a lot of juice still to be pumped from copper pins
Ah, I follow you now :smile: Maybe you're right about an 'early' migration away to new materials to keep that advantage. However, wouldn't that mean they'd have to take the industry with them, or do you think they'd create their own AMD-only memory technology for high-bandwidth discrete graphics?Actually I was thinking in terms of copper's limit itself. I do not know of any interconnect which can do >10gbps per pin on copper apart from rambus itself. USB3 was supposed to be optical, Light peak is optical, 10GbE is optical.
As you correctly pointed out, AMD's advantage in bandwidth per pin helps their products a lot. So may be just to preserve that edge, AMD is a more likely migration candidate.
Or just may be, there is a lot of juice still to be pumped from copper pins
Depends, if the rambus i/f is cheap enough for a generation or two (after counting the monopoly prices of XDR RAM, likely to be expensive though
) and alternatives are late enough, they just might go that way. Industry standard interfaces would undoubtedly be the preferable choice though.
1. You can run 10GBE over copper. Look at solarflare, they have a PHY that can run over 100 meters on cat7 with less than 10^(-12) BER and 45 meters for cat 5e. Remember that the memory interconnect is a much more controlled environment than an ethernet cable, inches instead of meters and a nice PCB.
2. Rambus has done a huge amount of work to push the industry forward on interconnects. They deserve recognition for all their work. The financial aspects are a little trickier. Certainly they haven't won a lot of friends with their approach, but frankly most large companies won't hesitate to screw over a small company at all. It's really clear that there was dirt on both sides (DRAM makers, Rambus), and that's about all that can be said - it's not black and white, it's very grey.
3. I have a hard time seeing the ITC ban NV products. That's only happened once or twice before that I can recall. I'm thinking of the BCOM/QCOM case and then there was one with a packaging company and someone else. But if ITC does ban NV, then they will settle in a hurry.
David
2. Rambus has done a huge amount of work to push the industry forward on interconnects. They deserve recognition for all their work. The financial aspects are a little trickier. Certainly they haven't won a lot of friends with their approach, but frankly most large companies won't hesitate to screw over a small company at all. It's really clear that there was dirt on both sides (DRAM makers, Rambus), and that's about all that can be said - it's not black and white, it's very grey.
3. I have a hard time seeing the ITC ban NV products. That's only happened once or twice before that I can recall. I'm thinking of the BCOM/QCOM case and then there was one with a packaging company and someone else. But if ITC does ban NV, then they will settle in a hurry.
David
I don't think it's cut and dried yet that there's any real stealing going on, but I'm pretty sure that at least one of the sides in this conflict has dynamite fishing with a broad range of patents as its main legal protection racket, err excuse me business model. Sure, many companies have accepted these robber barons as an unfortunate part of the cost of doing business, but I'm always glad when someone doesn't just roll over.
Sounds remarkably like wishful thinking to me.
Silent_Buddha
Legend
I believe there's already been a preliminary injunction on their products and products by companies making products based off of those. But since Nvidia is continuing to contest and appeal, there has been a stay put on the injunction.
Regards,
SB
aaronspink
Veteran
Nice PCB? PCB are absolute junk compared to cat5e and cat 6/7. Unless you are using an exotic PCB material, the losses through the PCB per inch are orders of magnitude higher than with cables.
The ITC rarely puts a ban on products because the companies almost always settle as soon as the ITC has given a ruling. Basically once you have an ITC ruling, you can pretty much get an import injunction at will.
aaronspink
Veteran
I know from a technical perspective that Rambus and its founders have pretty much pioneered the bullk of research on high speed signaling over FR4. Its amazing what a lot of money from a convicted law breaking cartel can due to a small company's reputation. As far as Rambus is concerned, they've won in enough courts now that anyone thing the patents don't have validity is a fool.
The whole point of going to the ITC is to get an import injunction to force through a settlement. Its why companies file with the ITC.
I stand corrected. I guess the saving grace is that the inches on a PCB are orders of magnitude shorter than the meters in a cable.
My understanding is that in 10GBE that only two of the wires are signaling and the other ones are mainly there for insulation/SI. Is that off the wall?
So the real question is how long with NV's appeal last, and whether Rambus can get a preliminary injunction based on the likelihood of victory...
I don't know much about ITC and how that works yet.
David
itsmydamnation
Veteran
i dont know about 10GE but 1000BASE-T uses all 8 pins in both directions at the same time and uses the fact it knows the pins are in colision to clean up the singnal . But 10GE base T has big latency and high power usage to the point that in datacentres the heat given off by it is a problem. AT 10GE SR/USR (ultra/short reach) optics are the best bet, cheap as chips as well compared to traditional networking optics.
FrameBuffer
Banned
Well being as pretty much ALL GDDR manufacturers (Samsung, Hynix, Qimonda etc and members of Jedec) have all settled with Rambus and the validity of the patents have stood, along with NV admitting their products "adhere to Jedec" standards (which have been found to use Rambus IP) .. NV has pretty much all but admitted to using tech that they have not licensed. It's not like this is a new thing either.. Rambus has been seeking a resolution with NV for the greater part of a decade and NV has refused, instead feeling they shouldn't have to pay for something (which everyone else does).
heres a repost of pretty much sums it up with links:
------------------------------------------------------------------------------------------------------------
I'm not up on all the details but wasn't it ATI in co-operation with Jedec that created GDDR3 in the 1st place ? Not sure if GDDR3 (and thus 3,4 and 5) all use Rambus IP, and if that was part of the licensing agreement with Rambus. The problem being that ATI and Jedec members (Those producing GDDR5, ir Hynix, Samsung, Qimoda) all pay licensing fees (royalties) to Rambus, and that NV touts itself as adhering to "Jedec Standards" which have been found to infringe on Rambus IP. So by NV's own admission they are in violation of the patents. Estimates of damages range from the low end (250M) up to 900M-1B (willful violation of patent - triple damages iirc).
Edit: Hynix/Rambus agreement
Samsung v Rambus - http://www.rambus.com/us/news/press_...10/100119.html seems be centered around rambus claims against samsung
http://www.reuters.com/article/idCNN...0100129?rpc=44
Those five-year worldwide license would cover SDR, DDR, DDR2, DDR3, GDDR3 and GDDR4 controllers, with initial rates ranging from 1.5 percent to 2.65 percent
For an easy to follow table of Rambus legal exploits see: http://www.investorvillage.com/smbd....8364387&pt=msg
-------------------------------------------------------------------------
aaronspink
Veteran
Yep
The 10GBase-T standard actually uses all 4 pairs each running at 125 MBaud with multi-level signalling.
Yes, ATI created GDDR3 with Jedec
rendezvous
Regular
We will know more about the signs of strain om Wednesday when NVIDIA holds their fourth quarter conference call.
Wednesday, February 17, 2010 2:00 p.m. PT
http://phx.corporate-ir.net/phoenix.zhtml?c=116466&p=irol-EventDetails&EventId=2723396
Wednesday, February 17, 2010 2:00 p.m. PT
http://phx.corporate-ir.net/phoenix.zhtml?c=116466&p=irol-EventDetails&EventId=2723396
chavvdarrr
Veteran
huh?
I still remember years ago when Rambus were damned to file for patents while discussing DDR creation in Jedec?
Being juridically right is not as morally right.
- Status
Similar threads
