What do AMD need to do to stay ahead? Intel creep closer with Northwood.... In my opinion, this is what AMD must do with thoroughbred to keep ahead: 1) Increase FSB. 333 would be a good start, along with DDR333/PC2700 memory. Maybe even then go to a 400MHz FSB (as AMD said in their original K7 releases, the FSB was scalable to 400MHz). 2) More cache. Intel have done it with northwood and Tualatin. Thoroughbred needs to be at least 512KB L2 cache, and maybe make 1MB versions available also (perhaps MPs?). Also, increase the cache width from the archaic 64 bits in the present Athlons to 128 or 256 (as in coppermine, Tualatin, Willamette and Northwood). Maintain the 16-way associativity with the increased width and size, and performance will increase massively. 3) Better thermal protection and measurement. Follow intel's lead. 4) Heatspreader. Again, Intel have done this, to prevent core cracking, etc. 5) Produce clearer guidelines about the 4 mounting holes around the socket, so that the HSFs will fit more motherboards (eg PAL8045 doesn't fit on many), and about the diameter (PAL8045 comes with washers for both sizes of holes used in motherboards). 6) SSE2. Intel are forcing this on people, in order that the P4 will actually perform reasonably. We've seen the difference SSE made to some benchmarks (and mirrored in real-world apps), and so AMD have to take this on in order to not be left behind by P4 optimised code.
I think with the push AMD are making into the dual market, they should probably bring the MP price more inline with the XP.
Yeah, same with the heat protection. It royally sucks if your ZIF tabs broke off due to the heatsink and your chip fries along with it.
im suprised that AMD havent picked up on the heat-plate idea yet with the athlons/durons. why they used them on the K6s and then scrapped them i dont know! they might need some improvement as the athlons/durons obv make more heat than the K6s used to, so perhaps theyre inventing some really good way of implemeting it, so i still hold hope for AMD
The K6-2s and so on were not flip-chips (about which I posted an explanation on these forums some time ago...http://forums.bit-tech.net/showthread.php?s=&threadid=2745). Adding a heatspreader is not quite as simple as just glueing on a bit of metal. However, give them time, and I am sure AMD will do it.
I agree with everything IS said, however id like to hmm "edit" a few points 1) Def 333/400 fsb perhaps both with the option of either like 200/266? 2) Northwood only got at max a 9% increase in performace from doubling its cache with a large increase in cost (relative to performance) cause getting Sram to run at 2gig+ must require some specialist stuff. Perhaps a bit more, but not much to keep costs down. Should have the option of 256 or 512 cache. 3) MUCH better thermal protection!!! like p4 standard!! 4) Heatspreader that's copper! Im not sure how the P4 spreader is attached to the core but i hope its with loads of some sort of thermal epoxy not the odd spoldge of white goo. 5) Hmm.. that's up to manuf. of heatsinks + mobos as well as AMD to sort this out though. 6) SSE2 would be good, but its really not fully utilised yet. I agree that they should implement it into their new chip to keep up and for the future cause they waited WAAAAAAAAAAAAAAAY too long to stick full SSE1 extensions into the XP.
The cache issue is interesting.... It really depends what you're doing. For my photo-editing work (which some of you are familiar with), more cache would translate into enormous performance gains. There's no difficulty in making SRAM work at 2GHz+...it just takes up space on the die. This is why the Duron is cheaper, because it has a smaller die with less cache than the Athlon. Heatspreaders aren't simple, because you need to conduct heat from the semiconductor, through the gate oxide and/or metal contacts (interconnects) and up into your heat spreader. You need to fill the space with something very thermally conductive, but completely electrically insulating. You then have to be careful not to get capacitance from this material, and the metal heat-spreader....not easy.
i didnt know bout the sram speed thingy or the complicated answer to the heat spreader just you learn something new every day huh?! They should def. give an opition about the cache though, whether u wanna pay more/less depending on what u do - like the old Xeons that used to come in 512k/1M/2M versions.
Cache does help quite significantly for some applications....my old P3 500 with 512k half speed L2 cache was a good SETI cruncher, would do a unit in about 13 hours which was good considering its clockspeed.....an 800 Duron takes about that long I agree about the thermal protection thing most of all though.....they need to sort that out!
Indeed they should give the option of the cache. We call it the Duron. I think, with the 128K L1 Cache, they should make Durons 128 or 256 (as opposed to 64) and Athlon 512/1MB.
1M!! whow id like to see that. However i think they should keep the budget 64k durons, but 64 is sod all compared to 256/512k. Maybe they should stick duron @ 256 and like u say - Athlon @512/1M?
On a .13micron process, a 256K CPU (ie the present AthlonXP) could be produced for less than the current duron. The factor in CPU costs is yield mainly. The P4, with it's 200mm^2+ die is expensive. The Duron with it's sub 100mm^2 die is cheap. The reason is that the size of the die dictates how many CPUs can be made on a single wafer. The more that are made, the more that are likely to work, for the same production cost. Thus, each single CPU is cheaper.