AMD's Athlon Stepping Improvements
Table of contents
- 1. Step By Step: AMD's Athlon 64 X2 Progress Analyzed
- 2. What's A Stepping?
- 3. Athlon 64 X2 5000 Steppings
- 4. 90 nm Windsor F3: ADO5000IAA5/6CZ
- 5. 65 nm Brisbane G1: ADO5000IAA5DD
- 6. 65 nm Brisbane G2: ADO5000IAA5DO
- 7. Test System

The world is holding its breath as AMD readies the launch of its redone Phenom processors. As most tech-savvy users know, AMD has gotten off to a really bad start with the Phenom 9000 processors and their Opteron counterparts for the professional market. The quad core called Agena was supposed to get AMD back in the fast lane, after the firm was unable to offer a device that could keep up with Intel’s Core 2 Duo. Unfortunately, the new product was not only many months late, but a nasty bug forced AMD to suspend shipments until a new revision of the micro chip would finally be available. Such a revision is referred to as a stepping, and it is imminent for the Phenom family. But what is the potential impact of such a new stepping? We looked at four different steppings of the Athlon 64 X2 5000 to look at the improvements.
AMD already said that people should not expect major performance improvements when the new B3 stepping hits the market, as it is mainly a bug-fix release. The Phenom design consists of the typical 64+64 kB data and instruction L1 cache and 512 kB L2 cache per core plus a shared 2 MB L3 cache that is dynamically allocated to one or all cores. None of this will be changed and the instruction set and power management remain unchanged as well. This also applies for the manufacturing process; although AMD is on its way to producing 45-nm silicon later this year, Phenom remains tied to the 65-nm DSL SOI process (dual stress liner, silicon on insulator). Still, we believe that at least some clock speed improvements will be possible, as AMD has already handed out a few samples of a 2.4 GHz Phenom processor, which hasn’t been commercially available yet.
AMD’s progress is nicely summarized on its corporate Website at Desktop CPU Selector. Here you can select a processor family, model number, frequency, cache size, socket and other parameters to see if there are models specifically matching your requirements. We used this tool to find a processor model based on the dual-core Athlon 64 X2, which has been available long enough to have undergone multiple revisions. The processor of our choice was an Athlon 64 X2 5000, which we obtained in two 90-nm stepping versions and two more revisions based on the 65-nm process. Let’s find out about the impact of the individual steppings.
Latest Motherboards News
- 25/05 – Thunderbolt Will Become Key Motherboard Spec in 2H 2012
- 22/05 – Does Your AMD FX Platform BSOD with Steam? Read This.
- 04/05 – 54 Million Enthusiast and Performance PC Gamers Globally
- 01/05 – Intel's Next Unit of Computing Rivals Raspberry Pi in Size
- 10/04 – Intel P67 Express Chipset Begins Product Discontinuance Cycle
Latest Motherboards reviews
- 30/04 – Six Sub-£160 Z77 Motherboards, Benchmarked And Reviewed
- 23/04 – Video Teaser: Intel's DZ77GA-70K Motherboard
- 17/04 – Overclocking: Asus Rampage IV Extreme Versus EVGA X79 FTW
- 08/04 – Intel’s Z77 Express And Lucidlogix MVP: New Features...
- 27/02 – Six £165-£215 LGA 2011 Motherboards, Reviewed
you could of chucked in a core2duo so that we could see the difference between intel and the new stepping.
You are reading the bottom of a coffee cup.
You say PCMark05 results show a significant difference and believe it to explain something to you. However, just beneath the PCMark05 result you show the results of another synthetic benchmark - SiSoft Sandra - and you ignore it. Can you explain why the G2 sticks out from all the others?
You then use the single F3 stepping to claim that the 65nm process makes no difference in power consumption compared to the 90nm process, and again ignore a detail. Is the F3 not using much more power than the other steppings and was it not manufactured in a 90nm process?
Last but not least, you are not giving out any information about the variances of the results, which however would have made them conclusive. When some results are inconclusive because they are all the same it does not mean that it makes difference in another benchmark automatically conclusive. To show that results are conclusive you need to look at the variances, too, when you see that the results are close together.
In short, you have made a large effort for saying nothing conclusive. If you are sure about the difference between the steppings (i.e. with the increased L2 latency) you should be able to make it much more visible since you know what to look for. If it turns out to be impossible then you should search for the reason why it is impossible instead of ignoring important details.