SI
SI
discoversearch

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: Jdaasoc who wrote (27721)8/26/1999 11:03:00 AM
From: MulhollandDrive  Read Replies (2) | Respond to of 93625
 
Wow, look at the puny volume.....

bp



To: Jdaasoc who wrote (27721)8/26/1999 11:05:00 PM
From: Dan3  Read Replies (2) | Respond to of 93625
 
Re: 2 RDRAM channels, then 4 RDRAM channels

But that's just it, you don't need 2 or 4 channels, the constraint isn't bandwidth, it's latency. Modern systems have caches of at least 32K, most have a multiple of that. 128K cache is becoming almost standard. When cache is that large, the CPU is in cache close to 99% of the time. A cache miss results in a huge penalty at high clock speeds as the system is stalled while it waits for instructions or data, but then it is back running in its cache again - it doesn't need anything more for many cycles, plenty of time for even PC100 bandwidth, much less DDR 266.

Most big data moves involve either network or disk i/o, and FPM RAM is fast enough for that! Video cards need to move data around quickly, and rambus or DDR DRAM are useful in such an application, but that's about it.

Look at the results of business application testing at speeds up to 800MHZ at:
tomshardware.com
The performance isn't tapering off.

Now look at video gaming applications:
tomshardware.com
A slight flattening of the curve is visible. If you look at the other tests in the series, you'll see that the video cards are generally the limiting factor. If more bandwidth is necessary, DDR DRAM can more than double the bandwidth of existing DRAM, while using all of the same fab and testing equipment, and with a much smaller increase in die size than rambus.

If there were any basis to the notion that rambus memory bandwidth provides benefits any where near to its cost, it would have shown here.

Dan