View Full Version : 90 nm Athlon 64 with SSE3 ???
Star_Hunter
03-02-04, 08:39 PM
AMD Senior Vice President and Chief Financial Officer Robert J. Rivet has presented at the Morgan Stanley Semiconductor and Systems Conference today and here are some information he has shared regarding the upcoming 90nm Athlon 64 :
* AMD is right on target to ship out 1M units of Opteron and Athlon 64 in this quarter
* AMD's strategy is to establish in server first (high end client first), then mobile then desktop
* AMD to start fab conversion for 90nm process in 5 weeks
* First 90nm production silicon in April, volume production in July and availability around September-October targeted at Athlon 64 family
* 90nm Athlon 64 (Winchester) will be around 98-102mm2 in die size, about half the size of a 130nm Athlon 64 (193mm2) and smaller than Intel Prescott (112mm2)
Some of the architectural improvements such as SSE3 and DDR-2 might just be present in the 90nm Athlon 64.
Source: AMD
http://www.vr-zone.com/?i=554
If its true then things look even better for AMD (even though SSE3 does offer many Hyper-threading improvements it still does add some things the athlon 64 could gain with its use)
(this was in another forum but wrong one so the old one was deleted and I then posted this one )
Star Buddy, you realise you are posting specs on future hardware ? Its not set in concrete (as they say...it might just be present) so it goes in this section. Or any post about non-released hardware. Thank you
MUYA
Bring on SSE3, I'm ready. I think. :D :p
waiting for a athlon64 4000+ or higher meself :D
Sounds good, except for DDR-2. I hope they don't even consider it until DDR-II 667, even then i think DDR500 would be a better option.
woz_mann
03-07-04, 03:25 AM
DDR500 is too higher latency
From my understanding (I am probably worng) to main advantage of DDR2 is that it runs cooler and faster. If my thinking is correct maintain current VDIMM's and decrease the CAS rating will give it a lower latency.
but I am probably wrong!!
:surrender
Yes, wrong you are;)
DDR-2 has double the internal latency than DDR has. Add to this the extra cycles CAS (and tRAS ect timings) have and the latency of DDR-2 should be more than twice that of even DDR550.
So IMO, for just about any system DDR-2 is not a viable option until at least DDR-2 667. And even then i only think intel systems will benefit most. Of course this probably wont be the case as 2 is always a more attractive number than 1:mad:
DDR-2? Whats that? Where is my DDR-3? I would adopt XDR before i accept the poorly high latency DDR-2 is offering. Even if the internal latency wasnt twice as high, the CAS ect timings are very high even in comparison to DDR500+.
http://www.corsairmicro.com/corsair/xms2.html
DDR-2 Corsair CM2X512-4300 4-4-4-12(!)
DDR Corsair TWINX1024-4400 3-4-4-8
Remember DDR-2 also has effectively 2x the latency added on to what is shows by the timings.
So lower clock speed, with over twice the latency. Or less than half the latency with a higher clock speed :confused: Decisions? nah, there are none.
Of course as for which will be implemented into future systems.. well one is approved by JEDEC and the other isn't. Looks like we will be force fed DDR-2.
woz_mann
03-07-04, 10:21 AM
see i told you I prob be worng :p
Geforce4ti4200
03-07-04, 10:43 AM
oh so ddr2 will have double the latency ontop of its timings? then I dont see ddr2 being any good till 300fsb minimum and only if itll do like 3-3-3-8 timings at worst. im hoping bh4 ram comes out, this will kill ddr2 for the next few years anyway :D
GamblerFEXonlin
03-11-04, 08:00 AM
And what about socket 939, a A64 with two memory controllers is 6GB/s mem bandwidth!
http://www.anandtech.com/cpu/showdoc.html?i=1884&p=10
super-fast program switching in windows and game loading. the low latency already makes FPS games feel smoother i hear, lots of geometry and AI and sound stuff in memory for the CPU to do stuff with?
woz_mann
03-11-04, 12:22 PM
I have actually read that latency did not make as much difference with the FX chips as it did in previous chips......cant remember where I read it tho
bkswaney
03-11-04, 06:46 PM
But even DDR2-500 will offer higher memory bandwidth than any current DDR1.
Geforce4ti4200
03-11-04, 07:05 PM
Originally posted by bkswaney
But even DDR2-500 will offer higher memory bandwidth than any current DDR1.
yes but the latencies negate the extra bandwith anyway. Its not until perhaps ddr600 that ddr2 ram will become useful over todays pc3200 and pc3500 ram. bh5 and bh6 ram is the best now, some can do as high as ddr500 speeds at low latencies with mods.
caramon81
03-12-04, 12:01 AM
http://discuss.futuremark.com/forum/showthreaded.pl?Cat=&Board=techmobocpu&Number=3555315&Search=true&Forum=All_Forums&Words=dude101&Match=Username&Username=&Searchpage=0&Limit=25&Old=allposts&Main=3555315
This guy goes to that college, and posted on what he heard. He also had another post on the day the VP of AMD gave a speach,m but i can't find it.
Ok, found it
dude101
(Professional)
03/10/04 07:53 PM
Re: AMD CEO giving speech in an hour!!!!!
Ok guys I am back......
First of all I would like to make a correction in my previos statement. It wasnt the CEO who came but it was the Vice President. This was actually better because he was a design engineer before he was promoted. Now before you ask I didnt win the 64 but I did get a T-shirt.
Points of Interest
-AMD plans to use the x86 instruction set for all forms of computing... from cell phones too super high end servers. (a little far fetched imo)
-AMD has 90 nm cpus ready to go
-AMD also has hyperthreading ready to go
-It costs about $4 to manufacture a high end chip (not including the i/o)
-Windows XP-64 should be out by the second half of this year
-Tomorrow is officially "AMD" day and there will be speaches from R and D engineers all day which will have more details on whats to come in the future
-Derk Meyer really hates Intel
I'll keep you guys updated on tomorrows speeches
http://discuss.futuremark.com/forum/showthreaded.pl?Cat=&Board=techmobocpu&Number=3551799&Search=true&Forum=All_Forums&Words=dude101&Match=Username&Username=&Searchpage=2&Limit=25&Old=allposts&Main=3551139
Originally posted by caramon81
http://discuss.futuremark.com/forum/showthreaded.pl?Cat=&Board=techmobocpu&Number=3555315&Search=true&Forum=All_Forums&Words=dude101&Match=Username&Username=&Searchpage=0&Limit=25&Old=allposts&Main=3555315
This guy goes to that college, and posted on what he heard. He also had another post on the day the VP of AMD gave a speach,m but i can't find it.
Ok, found it
http://discuss.futuremark.com/forum/showthreaded.pl?Cat=&Board=techmobocpu&Number=3551799&Search=true&Forum=All_Forums&Words=dude101&Match=Username&Username=&Searchpage=2&Limit=25&Old=allposts&Main=3551139
Interesting about Hyperthreading.
Their venture into 90nm was already known with speculations/confirmations they are going to start runs in april and mass production 1 quater later or sumat.
$4 bucks only for a die??? hmmm
Ninja Prime
03-12-04, 12:41 AM
Umm I don't know where he's getting $4 a die, I recall Tom's Hardware having actual numbers saying it's closer to $40.
Originally posted by Ninja Prime
Umm I don't know where he's getting $4 a die, I recall Tom's Hardware having actual numbers saying it's closer to $40.
I was thinking that quoted number of $4 was a bit odd. Well I would belive the figure u quoted as more accurate :D
caramon81
03-12-04, 09:04 AM
Originally posted by Ninja Prime
Umm I don't know where he's getting $4 a die, I recall Tom's Hardware having actual numbers saying it's closer to $40.
He posted that it was $4 for the die, not including the i/o. so it probably is $40 for the die and i/o. not sure, you know how people forget/recall incorrectly after they are told.
Ninja Prime
03-12-04, 10:56 AM
I found it. (http://www20.tomshardware.com/cpu/20030923/athlon_64-11.html)
Yikes. AMD must not be making too much off of Athlon64's.
I can tell you this AMD is NOT even considering Hyper Threading what so ever! They said it was a choice 64bit or HT and they picked 64bit. They won't bother with HT as they are moving onto dual cores anyway.
vBulletin® v3.7.1, Copyright ©2000-2010, Jelsoft Enterprises Ltd.