(Cliff note: An excellent discussion of EDO RAM technology, copied from some Usenet postings. I'm sure this is obsolete by now.).
There's been a lot of queries regarding EDO DRAM since the Triton motherboards support it.
EDO DRAM is "Extended Data Out" DRAM. Compared to normal page mode DRAM (EDO DRAM also supports "page mode" accesses, or accesses where only the column address changes), EDO DRAM offers higher bandwidth.
This comes about because normal DRAM can not be relied upon to keep the data on it's output pins after CAS has been deasserted. EDO DRAM guarantees that data will remain on the pins after CAS deassertion for some number of nanoseconds. This allows the memory controller to overlap latching data with CAS precharge.
Graphically (in ASCII, anyway), this looks like:
Normal page-mode DRAM RAS----\__________________________________________/--------- CAS-------\______/---\______/---\______/---\______/--------- Data-------XXXX==XXXXXXXXX==XXXXXXXXX==XXXXXXXXX==---------- EDO DRAM with same timing RAS----\__________________________________________/--------- CAS-------\______/---\______/---\______/---\______/--------- Data-------XXXX====XXXXXXX====XXXXXXX====XXXXXXX====-------- EDO DRAM with optimized timing RAS----\__________________________________/---- CAS-------\____/---\____/---\____/---\____/---- Data-------XXXX==XXXXXXX==XXXXXXX==XXXXXXX==---
Note that time time for CAS to data valid (=) is the same as the other scenarios. And the CAS precharge is the same. The time that CAS is asserted is greater with normal page-mode DRAM because it must remain low until the data is latched. The cycle its latched is the same for both EDO and normal DRAM, but because latches have non-0 (usually) data hold times and normal DRAM has a 0 data hold time, we must leave CAS asserted for some time (usually a half clock to a full clock). With EDO DRAM, we can rely on the DRAM to have a hold time greater than our latch's, which means we can simultaneously latch the data AND deassert CAS, saving that 1/2 cycle or more. Hence, greater bandwidth.
This is completely independent from whether the L2 cache is present or not or what type of memory it uses. So the highest performance system will use BOTH a fast cache AND EDO DRAM.
I'm currently looking for a motherboard that supports both EDO DRAM AND synchronous burst SRAM cache. I'd appreciate people posting benchmarks comparing the ASUS Triton based card with various config- urations of cache and DRAM. Surely ASUS themselves has this data.
[ Very nice description of the means by which EDO RAM achieves improved performance -- omitted for brevity. Also, I'll emphasize again, EDO RAM != EDRAM -- a different "improved" memory type, which combines integral cache with DRAM in a single package. ]
This is completely independent from whether the L2 cache is present or not or what type of memory it uses. So the highest performance system will use BOTH a fast cache AND EDO DRAM.
Exactly so! The new Triton chipset can be used with EDO RAM and with synchronous burst-mode SRAM L2 cache. In low-end systems or portables where power consumption is an issue, the cache can be omitted. With EDO RAM as main memory, the performance loss will not be too severe relative to today's standard SRAM plus DRAM architecture. Alternatively, with synchronous cache plus EDO RAM, a very high-performance memory subsystem can be configured. The ultimate << Speed-freak's delight >> memory subsystem, using memory devices that are presently available, would be something like 6-8 nsec burst-mode SRAM cache with EDRAM in main memory, interleaved (of course!) to squeeze out that last bit of performance. Such an architecture would require adding 72-pin SIMMS four at a time in the main memory array. While the memory devices are in production, I'm not sure all of the "glue logic" necessary to build a PC is yet available. Triton, which supports the synchronous cache plus EDO RAM configuration, is available.
I'm currently looking for a motherboard ...
As you note, the new ASUS Triton-chipset Pentium motherboard, which supports synchronous burst-mode cache (256K standard, available now; 512K supported) and EDO RAM, is starting to reach end users.
One vendor on the net has some performance data with the four possible combinations of cache type and main memory type. Finger netex@tdl.com for details. [I have no association with this vendor.] About 40% gain in overall system performance for the "maximum" configuration (Synchronous SRAM plus EDO RAM) relative to the "Plato Generation" boards.
Now, if this board can stand the hacking of the "BF" pin to obtain 133 MHz core operation in a P5-100 .......... An interesting prospect indeed!
Not sure what the spec's say, but it's my understanding that the sync is not as fast as the piplined, however the pipelined is not available just yet.
Triton with EDO or with Burst SRAM + DRAMs
Q1: Does an EDO-DRAM base system not require a SRAM cache?
No, it does not. The Asus TP4 will run with EDO Ram w/ burst cache (a max of 256 kB), EDO Ram without burst cahe, DRAM w/ burst, and Dram without burst. You can also use normal, sync ram cache (a max of 512 kB), but it's NOT compatible with the Burst ram cache - DO NOT use them together! Cache performance is roughly comparable to the NP4 boards, although all disciplines generally a few % quicker. DRAM performance is almost doubled, from 19 MB/s on Neptune (Asus TP4) to 39 MB/s on Triton using Burst/EDO... So expect *much* increased preformance with Burst/EDO ram, around 50 - 70% and *slightly* increased performance with Burst/DRAM, around 30 - 50%. Numbers with Syncram cache a lower, expect between 10 and 50 % increase w/ EDO, and 10 and 30 % with DRAM. Numbers from Asus.
My understanding was that the intel motherboard (aladin) supported a level 2 cache with either (or both) DRAM or EDO RAM. At least this was what was stated in PC Magazine (UK edition). They tested two systems using this board, one from Viglen (using 16MB EDO ram and SRAM cache) and one from Panrix (8MB of EDO ram + 8MB DRAM, pipelined SRAM) according to the editorial they had these machines for a month and they're the fastest they have tested. The panrix machine was about 10% faster on there DOS system benchmark.
The intel Aladin motherboard does look rather "sexy", with an onboard enhanced IDE controller (up to mode 4, apparently will suport mode 5 through drivers under windows 96, er 95) also has onboard floppy/parallel/ serial ports (with all those E TLA :-) ) and an onboard soundblaster compatible sound "card". 4 PCI slots and 3 ISA, one shared (from memory, i may have remembered the exact number wrong!) Also has full plug and play for ISA and PCI. Panrix normally sells motherboards as well as complete systems, but i have no idea what availability is like. Both panrix and viglen quoted prices for there systems.
The Triton chip set used in the aladin motherboard is supposed to offer twice the memory bandwith of the previous intel chipset, which should make these boards very nice.
Any info on the Intel manufactured board (or ones by others such as Asus) greatly appreciated. As soon as i hear you can run DOS, OS/2 and Linux in a stable fashion on these, I'm getting one :-) Of course availability is bound to be a problem!