<133 x 2 (for double data rate) = 266>
Dan, for the last time, you say "260% increase" over PC100, not "260% of the performance" of PC100. There is a difference. A "260% increase" means that DDR's bandwidth is 360% times PC100, which is absolutely false. A "100% increase" means a doubling of bandwidth, a "200% increase" means a tripling of bandwidth, and you're saying that DDR more than triples the bandwidth of PC100. DDR SDRAM provides a 166% bandwidth increase over PC100, not the 260% increase that you have said time and time again. Please, please, get this straight.
<Thats from NEC's white paper on VC DRAM - they fully document their test platform (on an intel processor!). On the Intel memtest they achieved a 171% of the performance of PC100. What is the corresponding figure for Rambus?>
Dan, you're picking and choosing your numbers to make your argument look good. Will you report the whole picture, not just one data point? The other three benchmarks that test VC133 vs. PC100 shows 26% to 31% improvement. Did you forget to tell us about that? Or how about the fact that none of those benchmarks really tell me anything about real-world performance? They're all synthetic benchmarks; where are the Winstone scores?
<Sure it's a best case for VC133, but VC266 will be available in the near term and will yield additional improvement.>
Tell me again, where is the industry support for VC133 or VC266? I've seen the NEC links before; all you're doing is repeating your own arguments, as if repetition makes them any more valid. Who else is supporting VC133? Where is the chipset support? Why is the anti-Rambus bandwagon focused on DDR SDRAM and ignoring VC133 or VC266? Why is AMD saying Athlon will move towards DDR SDRAM? How come AMD didn't say anything about VC133 or VC266?
<The great majority of complete motherboards, with power components, cardslots, 64 bit memory traces, chipset, serial, parallel, usb ports, cpu socket, etc. sell for $65 to $120 - at retail, not cost. Will reducing the number of memory traces (while requiring that they support 400MHZ instead of 133MHZ) reduce cost by $100 to $150?>
First of all, you are right. Those motherboards being sold for $65 are priced below cost because they are obsolete motherboards that are accumulating in inventory. Those low prices have little to do with cost, and more to do with customer demand or lack thereof.
And second, when you manufacture millions of motherboards per quarter, not to mention millions of chipsets, every little bit counts. Say for instance that the lower pin counts save you $10 per motherboard. Those savings really add up when you multiply that by a volume of a million motherboards per quarter. And that can make the difference between a company who prospers and a company who swims in the red ink. (But judging from your posts on the AMD thread, I guess you wouldn't know the difference.)
<300/600 has the full streaming data benefit and 150% of the burst performance compared to PC100 - if that wasn't enough to overcome rambus's drawbacks for PCs what difference will the jump to 400/800 make when the rest of the memory market has been making such significant strides?>
Dan, please quit with the false inductive reasoning. Once again, you are applying a double-standard. PC133 has shown to have no benefit in real-world performance over PC100. If I used your same method of induction, then I would say, "If PC133's increased bandwidth wasn't enough to overcome PC100's drawbacks then what difference will the jump to DDR SDRAM make?"
As for DRDRAM 300/600, I don't know exactly what its limitations are. If it is true that it provides equivalent performance to PC100, then I'll guess that the reasons are:
1) The benchmarks that were run don't take a lot of advantage of the bandwidth and the streaming
2) The latency is slightly longer
3) The memory controller design hasn't matured yet. DRDRAM needs a sophisticated controller in order to take full advantage of its capabilities. Sure, someone can put together a half-baked Rambus controller, but its performance is going to stink.
<Why have we never, ever seen any benchmarks for a rambus equipped PC? Jeez, how hard can it be to at least set up a behind the curtain ringer system for an outfit as big as Intel?>
Dan, just wait. I'm sure at the Intel Developer's Forum later this month, there will be official benchmarks for DRDRAM.
<Tenchusatsu, I've posted the issue of 300/600's failure periodically, and am either ignored, or get drivel back.>
This reply to you has taken more time than I would have liked. Not only do I have to give you technical explanations, but I also have to correct your faulty logic and address your same old FUD. I think the reason why people ignore you is that they consider you to be close-minded and not worth the effort. I'm beginning to see why.
Tenchusatsu |
