Tech Update: AMD core changes on the horizon
by Kristopher Kubicki on April 16, 2004 2:29 PM EST- Posted in
- CPUs
All original Athlon 64s were based off the Clawhammer core; 130nm SOI with 1MB L2 cache. The recent Athlon 64 3000+ was also based on the ClawHammer core, but with half the L2 cache disabled. These Clawhammer (also known as Model 4) Athlons will undergo significant change sometime in the next few weeks. In a nutshell, Clawhammer cores will all be replaced by Newcastle cores (130nm SOI, 512KB L2 cache). The future Newcastle based Athlon 64s are also known as "Model C" Athlons.
| Socket-754 Roadmap Update for 2004 | ||||
|---|---|---|---|---|
| Core | Clock Speed | Cache Size | Release Date | |
| AMD Athlon 64 3400+ | Newcastle |
2.4GHz |
512KB |
Q2 '04 |
| AMD Athlon 64 3400+ | ClawHammer |
2.2GHz |
1MB |
Already Available
|
| AMD Athlon 64 3200+ | Newcastle |
2.2GHz |
512KB |
Q2 '04 |
| AMD Athlon 64 3200+ | ClawHammer
|
2.0GHz |
1MB |
Already Available |
| AMD Athlon 64 3000+ | Newcastle |
2.0GHz |
512KB |
Q2 '04
|
| AMD Athlon 64 3000+ | ClawHammer
(Half L2 Disabled) |
2.0GHz |
512KB |
Already Available |
Even though the new 3200+ and 3400+ will benifit from the 200MHz upgrade on the Newcastle core, the 50% smaller cache size has many negative implications; content creation and encoding may suffer. We did not find any mention of the Athlon 64 2800+ transitioning to the new Newcastle core, or confirmation as to whether or not the upcoming 3700+ will utilize the Newcastle core. Since the newest 3400+ receives an extra 200MHz boost in clockspeed, it is almost certain the 3700+ will debut with a 2.6GHz or higher core if it based off the 512KB L2 Newcastle core.
Here is a quick update on what the new SKU changes will look like.
| Socket-754 SKU Updates for 2004 | ||||
|---|---|---|---|---|
| Old SKU | New SKU | |||
| AMD Athlon 64 3400+ | ADA3400AEP5AR |
ADA3400AEP4AX |
||
| AMD Athlon 64 3200+ | ADA3200AEP5AR |
ADA3200AEP4AX |
||
| AMD Athlon 64 3000+ | ADA3000AEP4AR |
ADA3000AEP4AX |
||
This may affect some motherboards since they may not be able to detect the new Model C Athlon 64s without a BIOS update. The CPU core ID will change from F4A to FC0 with the new processors.
Facebook
Twitter
RSS
18 Comments
View All Comments
Nemesis77 - Monday, April 26, 2004 - link
"The FX51 is dual channel but in those recent AT tests it was virtually the same as the a64 in performance even in bandwidth tests like encoding"You need to keep in mind that Opteron/FX uses registered DIMM's, and those carry a latency-penalty. Socket939 gets the benefit of increased memory-bandwidth without the drawback of increased latency (like Opteron/FX does).
Pumpkinierre - Saturday, April 17, 2004 - link
Yeah the performance difference between the dual channel and the single channel as well as the P4 is a mystery. The FX51 is dual channel but in those recent AT tests it was virtually the same as the a64 in performance even in bandwidth tests like encoding(http://www.anandtech.com/cpu/showdoc.html?i=2002&a... FX51 creams the P4 in synthetic bandwidth tests (sandra, sciencemark2) which is strange because the P4 is quad pumped. But that dominance doesnt seem to translate to the media streaming apps tests where the P4 still has the edge. The FX53 is competitive so cpu clockspeed might be more important than bus width which makes me think that the bandwidth tests are as much latency dependent as bandwidth dependent. Based on the AT encoding test, the single channel a64 is almost as good and is probably lower in latency and stabler. So there must be a bottleneck here, or dual channel come into its own with cpu speed increase (the FX53 points to this but a scaling test is needed), or AMD's twin channel is a big con and you dont need it.Boozish - Saturday, April 17, 2004 - link
making for very very little performance differenceBrickster - Saturday, April 17, 2004 - link
To #14...Dual Channel memory support to all A64's also for 939.
Boozish - Saturday, April 17, 2004 - link
Now that the Newcastles and 939-socket cpus are almost the same, i dont see the point of going from 754 to 939. The only thing that looks good about the 939 is the pci-express mobos.AsiLuc - Saturday, April 17, 2004 - link
I didn't know this.But, is it correct to assume that the Athlon 64 fx will still feature 1 mb cache? Does fx go from 940->939 with dualchannel and 1 mb?
KristopherKubicki - Friday, April 16, 2004 - link
I hope i knew it, i mean, i wrote the last article too!the only real news here is the bump in clock speed.
Kristopher
Jeff7181 - Friday, April 16, 2004 - link
No offense to the author of the article... but... this is news??????? I thought we already knew this... I already knew this... did you?KristopherKubicki - Friday, April 16, 2004 - link
Trogdor: Thats IF you believe the PR ratings are really actual ratings and not just marketing labels -- :-X Not to be cynical or anything...Kristopher
TrogdorJW - Friday, April 16, 2004 - link
Don't get confused by the socket 939 CPUs, people. Here at least we are getting some good benefit from Newcastle. We lose a few % points in performance from the decreased L2 cache and got up anywhere from 8 to 10% in clock speed. (Though perhaps we don't benefit as much as AMD - they just trimmed core size by about 30 million transistors, it looks like... almost 25% fewer transistors! They cut costs by at least 20%, price stays about the same. Don't see any Fanboyz trumpeting that, do you?)What bothers me is the 939 CPUs that will have the same characteristics as Newcastle, but get higher ratings due to the dual-channel RAM capability. I believe the 2.2 GHz 939 CPU is going to be a 3500+ and the 2.4 GHz 939 will be a 3800+. Maybe we really will see a >10% increase by going to dual-channel and socket 939, but the cores will still be virtually identical, won't they? Or maybe there are extra HyperTransport connections on 939 CPUs that add cost and transistor space?
Oh, well. So we went from "Model 4" processors to "Model C" - what happened to models 5 through B? Hopefully they were just not as good as the new and improved Model C, right? Maybe overclocking might actually achieve greater than 10% bumps in performance now.