Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Rambus (RMBS) - Eagle or Penguin -- Ignore unavailable to you. Want to Upgrade?

To: Dan3 who wrote (58884)	10/26/2000 9:14:36 AM
From: sylvester80	Read Replies (2) \| Respond to of 93625

Which IMO, it means that DDR is slower than PC133 SDRAM as the i815 has shown to be 10+% faster than KT133. DDR is slow expensive junk. Mobos are supposed to be even more expensive than P4 mobos. $216. Ouch! Expensive mobos, expensive incompatible that may or may not work sticks, and pathetic performance. Nobody wants this junk (but a few idiot bashers). No name brand OEM volume pusher will touch this junk. DDR is D.O.A. and dead dead dead. And dead slow to boot.

To: Dan3 who wrote (58884)	10/26/2000 9:38:20 AM
From: SBHX	Respond to of 93625

If you use today's apps represented by the winbench and bapcos of the world, latency to memory is still the slowest link in the chain. This is why if you look closely, a 200MHz (actually 100MHz, but dual data rate) DDR having higher latency than 133MHz SDR will actually perform slower in tests with less optimal reuse of data (the so called locality of reference problem). If you take the 32byte cache lines, the real kicker is that if only part of the 32byte is used for any computation, the biggest component in your measurements of clocks into memory will still be latency. (This is known as the granularity loss problem.) Hence, the benchmarks have to switch to tests of brute force sequential streaming types of applications where all read memory is used to make up for any increase in memory latency. Future apps apparently have this need. (This is known as the if you build it, they will come problem). You can actually see a real world application of this in the latest nvidia graphics chips where in the high resolution 32bpp bruteforce tests a DDR board can actually run close to 2x of a SDR board if you make the test heavily geared to huge triangles with lots of textures. SbH