Hi John Walliker; John Walliker quote fest. And re: "... John Walliker claimed to be an engineer, but actually turned out to be just another software weanie."
Here you are implying that you're an engineer:
John Walliker, September 5-7,1999
I have designed many printed circuit boards myself and have regular contact with manufacturers. #reply-11173365
I too have worked with high speed digital interfaces, starting with digital television in 1974 and currently with digital signal processors and FPGAs. I have also worked on microwave systems. #reply-11164389 I design digital speech processors for cochlear implants. #reply-10582175
Your profile lists you as a designer of DSP hardware: "Design DSP hardware and software for cochlear implants and signal processing hearing aids." #reply-17007299 Doesn't sound much like software. But now you "do other things": #reply-17008043
Also see where you recently failed to deny that you claimed to be an engineer: #reply-16292247 And your degree turns out to actually be in physics: #reply-11114232
Here you are admitting that you're just another software weanie:
Bilow, February 14, 2002
Hi John Walliker; We may be old fogies, but at least we haven't been forced to write software for a living... #reply-15355046
John Walliker, February 15, 2002
Well, I have, but I try to stick to a fairly low level - mostly C and DSP assembler. So far I have avoided any Windoze programming, but I'm not sure how much longer I can hold out. John #reply-15355494
The reason this is significant is because not being a hardware designer led you to some simplifying assumptions about what hardware design is all about. For example, from one of your first posts on SI:
John Walliker, June 17, 1999
My work involves ultra-low power consumption wearable devices using the Texas Instruments TMS320C54x family (soon to operate at 1.2V). Programs and data reside in on-chip static RAM with occasional accesses to flash eprom, so I don't have any direct experience of DRDRAM. However, I have spent enough hours probing circuits with a high speed digital oscilloscope to recognise that transmission-line technology such as that used by Rambus is the only way of increasing data rates across printed circuit boards substantially beyond those currently available. This is the most important reason why I am optimistic about the long-term prospects of the company. I did recently ask Texas Instruments when they intended to support Rambus in the TMS320C6000 family, for which it seems perfectly suited, but the person just smiled and said nothing.#reply-10160854
(1) Spending "hours probing circuits with a high speed digital oscilloscope" is hardly design experience. What you recognized is that there is an easy way to get increased data rates on a PCB, and that Rambus was using it. Transmission line theory is kid's play. Every decent hardware designer has known about this for years. ECL used transmission lines and you didn't see the memory industry converting to ECL compatible inputs, did you? Engineering is about making complex tradeoffs, and you were simply unaware of the disadvantages of going the Rambus route.
Engineering design is not about getting by the elegant (easy) way. It's about making the cheapest product that meets the requirements of the customer. Design is about cheap, it is not about elegance. With simple SDRAM or DDR, the signals ring. We design engineers know this. We're not stupid. We have software that predicts how much it will ring. We take this into account. We adjust the timing accordingly. Since you're not in the industry, you didn't bother to make the calculations to see what the fundamental limits (caused by ringing) that were present with DDR. So you were not in a position to make estimates of how far DDR would be able to be pushed. It was simply out of your league.
What you saw was an easy way to solve a problem. What you didn't realize is that that was not the most efficient (i.e. cheapest) way to solve that problem. But man, you sure made out like you were the big time expert on high speed digital oscilloscopes.
(2) Almost three years have gone by, and TI still doesn't support RDRAM. (By the way, I'm still waiting for TI to put their first DDR SDRAM interface on a DSP, but it's not like I predicted it.) So you wonder why the TI guy just smiled at you back in 1999. At that time, I was saying that RDRAM was a bad choice, and that Intel wouldn't be able to drive it mainstream. TI engineers were saying the same thing. So when you talked to "TI", all they would do was just smile at you. What did you expect, for him to tell you that you were an idiot straight to your face? If you were a design engineer, you'd have known that RDRAM was one hell of a nasty piece of work to debug, and that TI would have had to be insane to support it for DSPs. Rambus never made the slightest inroad into any small volume segment, it was always a technology for large volume production only. TI probably recognized this, recognized that their DSP market had too much small volume production in it, saw that the memory makers were having problems with it, and pulled the plug on it, if they ever even started an RDRAM project. Since you didn't have a buddy relationship with a TI sales rep, you couldn't get the real story. There are plenty of sales reps that I've known for nearly two decades. I drink with them. We ogle the same women. We go on vacation together. I've never got them in trouble for something they've said to me, and I've always treated them as well as I can. I don't quote the stuff they tell me. I always tell them the straight truth. If I don't think the product I'm working on is going to be a big seller I don't shine them on. They don't quote the stuff I tell them. We know each other, and we trust each other. So they tell me things that they're simply not going to tell you.
You're an outsider here. It's like when you walk into a small country bar and all the patrons suddenly shut up. Why are you surprised that you don't know who's dating whom?
-- Carl
P.S. More John Walliker quotes:
John Walliker, July 20, 1999
Have you actually worked at the hardware level with high speed digital signals and seen how extensively they are messed up by reflections from discontinuities? I have. Rambus have the pragmatic way of implementing things. Pragmatic because it will work comfortably now and will scale to MUCH higher speeds later without fundamental redesign. This is no ivory tower position (although I do have a physics degree and a postgraduate computer science diploma). I was involved with digital television development in 1974 (as a pre-university trainee at the BBC Research Department) and have practical experience up to the microwave region. I am long RMBS. #reply-10581379
Similarly, Rambus is the memory interface which can readily be scaled to the higher performance requirements of the future. #reply-10586963
Okay. Here it is 2002, and DDR has increased from PC1600 to PC2100, and now to PC2700. Samsung has announced PC3200. Meanwhile, graphics has gone from DDR200 to DDR600, and Samsung has announced DDR800. And where is RDRAM? Still, after 30 months, stuck at PC800, and maybe going to go to PC1066 in the next few months.
John Walliker, July 31, 1999
The use of the word "just" glosses over a much harder problem [getting PC200 to work] than most people appreciate. ... It is also wrong to suggest that doubling the data rate will not increase the power consumption. In CMOS circuits power is dissipated as a result of charging and discharging circuit capacitances. If the rate of doing this is doubled then so is the power consumption (if the voltage stays the same). Another issue is that increasing the data rate of conventional RAM will considerably increase radio frequency emissions. Many aspects of the Rambus design minimise these. Yet another issue is that conventional clocking arrangements fail when the wavelength of the clock signal is of the same order as the dimensions of the conductors. Rambus overcomes this by using two clocks, one for each direction of signal travel. #reply-10752735
It's kind of odd that both you and PTNewell, both with Physics degrees, would both get hung up on the concept that DDR had fundamental physical problems that would limit it. You both said that there would be problems with PC1600, but now the industry is shipping PC2700 with no great problems. Does this have anything to do with the guy who showed that bumble bees were incapable of flight?
John Walliker, August 2, 1999
So although it has not been reached yet, DDR and derivatives have a fundamental speed limit getting close, while Rambus does not. (Its limits are more to do with imperfections in the implementation and at extremely high frequencies they will occur when the WIDTH of the printed circuit tracks start to be close to a wavelength and multiple propagation modes become possible. That is a LONG way away. #reply-10775021
Also, both you an PTNewell were convinced (by Rambus lies) that RDRAM used less power than DDR, and that it was therefore a natural for mobile applications. This was contrary to the actual facts, as is demonstrated by Rambus' inability to get design win #1 in the mobile market even after 30 months of RDRAM:
John Walliker, September 8-11, 1999
The DDR interface is much more complex and will have a much higher power consumption. There are many different control signals which must all be timed correctly. This will become a nightmare when further speed increases are attempted. Rambus will scale up to much higher speeds. The DDR data confirms that this technology does not support multiple bits active on the bus at one time, in contrast to Rambus which already does according to published data. Therefore DDR will hit a performance brick wall which does not exist for Rambus. #reply-11184979
Therefore, the system power consumption using Rambus is about 1.4 times LOWER than DDR even though the Rambus chips used in this example are of an older generation than the DDR ones. #reply-11220856 Also, from 2001, see #reply-16585668
You were totally clueless as to how memory is priced, and what the implications of Samsung having lots of DDR part numbers was:
John Walliker, April 9, 2000
The tables show that the DDR market will be much more fragmented than the Rambus market, with lots of different types and less economy of scale. No manufacturer wants to produce lots of different variants of a product if they can help it. #reply-1338419
You've continued to be hopeful that Rambus would become mainstream since then.
John Walliker, April 9, 2001
I wonder how long it will be before the major fpga vendors start shipping devices with Rambus support? Once that happens, one of the major objections against using Rambus in designs intended for small to medium production runs will disappear. #reply-15635942 also #reply-15638535 and from way back in June 18, 1999, #reply-10171402
And I mean really, how desperate can you get with this one:
John Walliker, quoting Intel, October 31, 2001
... Intel wants to see "the vast majority of the Pentium 4 systems ship with RDRAMs. #reply-16586462
Also, for those who are into ECC esoterica, there's this fun sequence: #reply-16011504 #reply-16011537 #reply-16015451 #reply-16021015 #reply-16041726 #reply-16291927 #reply-16291930 #reply-16291945 #reply-16292222 #reply-16292419 #reply-16292470 #reply-16294071 #reply-16294345 #reply-16294038 #reply-16294397 #reply-16294412 #reply-16294038 #reply-16294343 #reply-16294369 #reply-16294410 #reply-16294419 |
