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.

Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?

To: Joe NYC who wrote (101291)	3/31/2000 12:07:00 PM
From: Scumbria	Read Replies (1) \| Respond to of 1573028

Joe, the question is what is the improvement in the hit ratio in ging to 192 L1 + L2? Doesn't the CPU have to (on L1 miss) lookup the memory in L2 before going to main memory? The way a victim cache works is- it allocates a new line when one is cast out from the L1. This guarantees (at least temporary) exclusivity from the L1. A victim L2 cache behaves as a slower extension of the L1. As such, the L1/L2 hit rate can be approximated from the sum of L1+L2 cache sizes. This makes a victim L2 a very effective solution. The only reason that victim L2's have not been used in the past, is that the L2 was off chip, and it had no visibility into the L1. Victim caches can only be used in integrated L2 designs. Answering your other question, there is no reason why you have to delay DRAM access pending an L2 lookup. (Although you may want to.) Scumbria

To: Joe NYC who wrote (101291)	3/31/2000 1:07:00 PM
From: Bill Jackson	Read Replies (1) \| Respond to of 1573028

Jozef, Having it full speed makes the difference. Half or 3/5 on die is better than off die but not as good as full speed on die. For the on die cache misses it must go down the cache hierarchy and eventually getting the data from main memory as the slowest but surest source(ignore hard drive cached data..~10000 times slower. So an optimization takes place. You put as much cache as close to the CPU and run it at the fullest speeds, making your cost, space and yield tradeoffs. The lower order cache is not usually as fast as so much time is wasted going to get the data off chip that full speed off chip cache is a luxury that few wish to pay for, unless you are in server space? Having a wristwatch gives you the fastest time, watch stopped...go to the clock radio, power out, pickup a transistor radio and tune in the time....nuclear war and all radios are off the air....fly to London and look at Big Ben(the hard drive). It is faster to look at the intermediary caches...even if you get a miss....than skipping them, unless you know they are a miss. Bill