backtop


Print 28 comment(s) - last by Trisped.. on Jul 19 at 12:20 PM


An excerpt from the Rambus ruling in April
Rambus may still receive $133M USD opposed to the $307M originally agreed upon, but the company is far from done fighting lawsuits

Earlier this year Rambus won a staggering $307M USD civil award against Hynix claiming the company infringed on its intellectual property.  Specifically, Rambus claimed that Hynix used techniques for producing memory that were proprietary to Rambus.  On Friday morning, Judge Ronald Whyte of Northern District of California overturned this award (PDF).

The latest statement from the court claimed "For the reasons set forth below, plaintiffs' motion for a new trial on the issue of damages is GRANTED unless Rambus files notice with the court within thirty (30) days of this order accepting remittitur of the jury award to $133,584,129 for damages through December 31, 2005."

The focal point for this overturn is due to discrepancies of the actual amount of DRAM sales during the time period Hynix infringed on Rambus IP.  Rambus at this point may either accept the lesser award of $133.6M USD or begin a new trial against Hynix.

Despite the interconnections between Hynix and Rambus, this slew of IP lawsuits are not related to the price fixing sentencing from earlier this year where four Hynix executives were given jail time. Within the past year every major DRAM manufacturer has faced litigation from Rambus, with the exception of the company's primary manufacturing partner Samsung.  Rambus has also filed suits claiming that all of the major DRAM providers artificially manipulated pricing on DRAM to hurt RDRAM sales -- with fairly damning evidence on the company website (PDF) -- again with the exception of Samsung.  On Friday 34 states filed anti-trust suits against most  major DRAM providers claiming the companies colluded to artificially inflate memory pricing, which is illegal in the US. Once again, Samsung was absent from this volley of litigation as well.

More can be seen from the official Rambus Litigation Update page.  The company currently has twelve outstanding major cases to follow.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: I dont get the rambus hate to this day...
By Acanthus on 7/18/2006 9:39:19 AM , Rating: 2
And that quote is talking about I820, which didnt benefit from the extra bandwidth granted by RDRAM.

Your latency statement is false, PC1066 RDRAM has less latency than PC3200 CAS 2-2-2-5 DDR.

I875 on Northwoods showed significant gains in many applications for RDRAM, its prohibitive cost (blamed on Rambus royalties, not volumes, (which were actually only about $2.50 per 256MB memory module).

RDRAM was killed because DRAM makers wanted to preserve DDRs future.

Do you honestly believe that DDRII could touch 128bit 1333mhz rambus? (Remember that rambus memory latency gets LOWER as the clockspeed increases).


RE: I dont get the rambus hate to this day...
By masher2 (blog) on 7/18/2006 12:07:18 PM , Rating: 3
> "PC1066 RDRAM has less latency than PC3200 CAS 2-2-2-5 DDR."

PC-1066 wasn't released till 2002, by which time RDRAM was already essentially a dead platform. And, more to the point, PC4200 was already out as well.

Finally, I'd like to point out that RDRAM latency is dependent on the number of sticks installed. Even by 2002, it was impossible to find RDRAMs in sticks larger than 512MB, whereas DDR was available in sticks of 1GB and even 2GB. There were signalling and termination issues with the larger sizes chips.

Another factor was Intel being late to the table with dual-channel support, and I won't even mention the prodigous heat output on RDRAMs.

> "I875 on Northwoods showed significant gains in many applications for RDRAM"

Of course...which explains why I purchased several RDRAM-based platforms myself. There were also applications where it did worse than DDR. Overall, RDRAM was the net performance winner...but convincing anyone to buy it, given its size constraints, cost, limited platform support, and other problems, was next to impossible.

> "Remember that rambus memory latency gets LOWER as the clockspeed increases"

All memory latency is lowered by increased clock rates. This is nothing unique to RDRAM.

> "RDRAM was killed because DRAM makers wanted to preserve DDRs future..."

Total rubbish. The same "DRAM makers" also made RDRAM. And they made *more* profit per stick on it...licensing fees didn't affect their profits much. They had no incentive whatsoever to kill RDRAM.

I spent many a post arguing *for* Rambus in the 1999-2001 era. Fact is, I could rarely convince anyone it was worth buying. It wasn't just the extra cost of the chips...there were many, many other factors bruiting about.


By dgingeri on 7/18/2006 12:56:14 PM , Rating: 2
quote:
Finally, I'd like to point out that RDRAM latency is dependent on the number of sticks installed. Even by 2002, it was impossible to find RDRAMs in sticks larger than 512MB, whereas DDR was available in sticks of 1GB and even 2GB. There were signalling and termination issues with the larger sizes chips.


One thing to note here: RDRAM, and the new XDR, have increasing latency per device attached, not just per stick. It works somewhat like Token Ring, and in fact is based off that idea. Each device has a dedicated addressing processor to decide where to store the data. the memory controller addresses each device in sequence to access the data. The memory controller has no ability to pre-fetch or pre-address a particular device, making the latency unavoidable. On top of that, each device has a 32 cycle response time. This far outstrips SDRAM's latency, no matter if it's SDR, DDR, or DDR2. IT does transfer faster once the device is found and it is all transferred, but if the memory is fragmented or the retrieval is small, it introduces much higher latency. The advantage of RDRAM and XDR is in single devices and large transfers, as in the Sony PS2. For document work and PC games, it sucks.


RE: I dont get the rambus hate to this day...
By fsardis on 7/18/2006 7:28:28 PM , Rating: 1
quote:
All memory latency is lowered by increased clock rates. This is nothing unique to RDRAM.


wrong. increasing the ram speeds inevitably means more latency. check out DDR modules for proof and even better check out the latency differences between DDR3200 and DDR4000. another good example is DDR vs DDR2.
yet another example is overclocking the ram which definitely means increasing the latencies to achieve higher speed.
so no, increasing the clocks doesnt lower the latency in any way.


RE: I dont get the rambus hate to this day...
By masher2 (blog) on 7/18/2006 8:05:41 PM , Rating: 2
> "wrong. increasing the ram speeds inevitably means more latency"

Err, no. It doesn't work like this at all. Latency is based on clock cycles...and the length of a cycle is the inverse of the operating frequency. Increase the clock rate on a memory module, and the latency decreases. In an exact, perfectly linear manner. Twice the clock rate = half the latency.

What you're thinking of is a different issue entirely. Quite often, you cannot increase the clock rate without relaxing the timings. And that may wind up increasing the overall latency more than what the higher clock yields. Result...more latency.

DDR2 has a higher latency than DDR. But its not due to the higher clock rate, its in despite of it. It has significantly higher timings, a prefetch buffer and additional mux/demux logic...all of which adds latency.

The OP was implying that the normal relationship between clock rate and latency was something special to RDRAMs. It is not...its a basic element of all memory technologies...though the latency is, of course, affected by many other factors besides clock rate.


RE: I dont get the rambus hate to this day...
By emboss on 7/18/2006 9:46:08 PM , Rating: 2
Just one minor clarification about an otherwise excellent post ... IIRC, latency (except for CAS) is actually specified in ns in the SPD data (which makes sense if you think about how RAM works). RAM controllers work at particular clock rate, which requires changing this ns time into clock cycles. Due to this quantization alone, increasing the clock speed (assuming the timings are strictly adhered to) may require an increase in latency (measured in ns).


By masher2 (blog) on 7/19/2006 12:17:13 AM , Rating: 2
> "Due to this quantization alone, increasing the clock speed (assuming the timings are strictly adhered to) may require an increase in latency (measured in ns)."

Exactly so....an excellent point which I was lax to not have mentioned.




By Acanthus on 7/19/2006 3:09:19 AM , Rating: 2
No, what i was saying was that with RDRAM, the timings were static, so the latency was significantly reduced as the clockspeed went up.


"I'm an Internet expert too. It's all right to wire the industrial zone only, but there are many problems if other regions of the North are wired." -- North Korean Supreme Commander Kim Jong-il














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki