Introduction

In February of this year, Intel released their first 90nm processor, code name Prescott. With a new architecture that promised everything from better energy efficiency to cooler temperatures to improved performance, the Prescott was arguably a failure. Hindered by extreme temperatures and disappointing performance, the Prescott was blasted by enthusiasts and reviewers alike for both its temperature issues (See the Prescott Survival Kit) and its performance as measured against the CPU it was designed to replace, the P4 Northwood.

 

Intel's "failure" to win hearts with the Prescott and AMD's notable absence from the 90nm CPU field has led many to question whether 90nm CPUs were worth the issues they raised.  A little more than six months later, shortly after releasing the socket 939 Newcastle, AMD has quietly released the 90nm Winchesters in 3000+, 3200+, and 3500+ for the socket 939. Noticeably absent from this list are the new 4000+ and FX-55 (The good news here is that AMD says their FX series of processors will make the jump to 90nm sometime during the first half of 2005).

 

 

 

Did AMD wait to see how Intel handled the thermal issues with the Prescott, or did they wait because they were having issues of their own moving from 130 to 90nm? After talking to a source at AMD, I was told that neither was the case. Instead it was simply a case of AMD choosing to work at their own pace on the 90nm processor, and only releasing it when they were satisfied with the end product (I wish all companies took this approach!!! Are you listening motherboard makers?).  Next was the rumor that the new 90nm processors would include changes to improve performance, most notably to the on die memory controller.  There was also talk that the < ?xml:namespace prefix = st1 ns = "urn:schemas-microsoft-com:office:smarttags" />Winchester would possibly include SSE3 commands. Again, this wasn't the case. Although I was told there were some "minor changes", a tune-up of sorts, there is no noticeable performance difference in the CPUs. This may come as disappointing news to some, but whether that really is the case, or whether AMD is letting the product speak for itself is something we'll touch on later in this article.

 

AMD has said that the major difference between the Winchester and Newcastle CPUs is that their 90nm CPU actually runs much cooler and consumes far less energy than its 130nm brethren. If the CPU does in fact run cooler and consumes less energy, does that make it a better overclock candidate than the 130nm chip? Again, we'll see later.

Features and Specifications

Athlon  64 3500+ (90nm)
Freq/Cache:                          2.2GHz / 512KB-L2
Process Technology:              90nm Silicon on Insulator (SOI)
Approximate Die Size:           84mm squared
Nominal Voltage:                   1.40 V
Max Thermal Design Power:   67 W
Max Icc (processor current):   45.8 A

Athlon  64 3500+ (130nm)
Freq/Cache:                          2.2GHz / 512KB-L2
Process Technology:              130nm Silicon on Insulator (SOI)
Approximate Die Size:           144mm squared
Nominal Voltage:                   1.50 V
Max Thermal Design Power:    89 W
Max Icc (processor current):   57.4 A

The difference between 90nm and 130nm:

Probably the most important aspect of moving to a 90nm process is cost, plain and simple. I've seen estimates where moving to a 90nm process allows AMD to generate 72% more processors per wafer. For a company like AMD, who doesn't have the resources of an Intel, this is huge. The benefit of lower manufacturing costs for AMD also benefits us, the consumers, simply by virtue of lower costs and more processors.

Are there other benefits of moving to a smaller process? Can AMD make the changes that Intel couldn't, and can AMD make 90nm processors a success?

What is obviously different is the die size. The 130nm die measures 144mm squared, while the 90nm die measures 84mm squared, a reduction of over 40%. The positive here is that by shrinking the die, you also shorten the distance electrons must travel; another benefit is that less voltage is also required to move them. The negative of shrinking the die is thermal leakage. When shrinking the die, the space between the traces shrink, leading to current leakage, current leakage means more heat, and more heat means a hotter running processor.

Another issue to consider is thermal density. The 90nm Winchester has the same number of transistors as the 130nm Newcastle, which increases its thermal density. By comparison, the 90nm P4 3.2 Prescott has 125 million transistors, while the 90nm AMD 3500+ has 68.5 million transistors. Although the die size is the same, the Winchester should run cooler because its fewer transistors mean the thermal density of the AMD 90nm die is less than Intel's 90nm die, requiring less current to power it. The 130nm Newcastle requires 89W, the 90nm Prescott requires 103W, while the 90nm Winchester requires only 67W!

The main issue to this point with the 90nm technology is high temperatures, but I believe the combination of fewer transistors (less thermal density) combined with a smaller die (less voltage) will actually lead to a much cooler CPU. A much cooler CPU tends to be a better performing CPU. Let's see how the new 90nm processor fairs against its 130nm counterpart, both thermally and in performance.

Testing

All testing was done using following system in a controlled environment:

*All tests were run at stock speeds and again at 240 FSB. For performance and stress testing, the following programs were used: Sisoft Sandra's CPU Arithmetic, CPU Multimedia, Memory Bandwidth, and Cache and Memory Test. 3D Mark 2005, Doom 3, Far Cry. Folding @ Home was used for 100% CPU usage to gain load temperatures.

All temperatures were recorded using ASUS PC Probe V2.23.01 in a 25C (+/- 1F) room. (Although no temperature probe is 100% accurate, they are good for reference when testing on the same motherboard, in the same environment.)

First, I needed to let the thermal paste settle to get an accurate temperature. To do this I let the Arctic Silver 5 cure for 10 days of 24/7 operation. During this time I played constantly with the system, trying to see if there was any difference at all between the 130 and 90nm processors. I ran all benchmarks with the multiplier set at 11 and the FSB at 200, then again at the best stable FSB I could maintain on my system, which ended up being 240 FSB.

Nothing too spectacular here, maybe a 1-2% increase in performance overall. Next I ran Doom 3, Far Cry, and 3DMark05, although these are graphics intensive benchmarks, I wanted to see if there was any performance difference using the exact same system with the different processor.

Benchmark 90nm@200FSB 90nm@240FSB 130nm@200FSB 130nm@240FSB
Far Cry (FPS) 72.07 83.00 68.19 80.65
Doom 3 (FPS) 88.04 96.7 85 87
3DMark05 3695 3752 3680 3741

Again the 90nm Winchester slightly outperforms the Newcastle almost across the board.

So, here at last we look at what makes the Winchester different from both its 130nm cousin as well as Intel's 90nm Prescott: the temperatures.

130nm @1.5V 200FSB Idle-37C  Load-45C
90nm @1.4V 200FSB Idle- 32C   Load- 40C
130nm @1.55V 240FSB Idle- 40C    Load- 52C
90nm @1.44V 240FSB Idle- 33C  Load- 42C

Although the performance of the 90nm Winchester is roughly the same as the Newcastle's, the temperatures are significantly lower across the board. While I saw a 5C difference at stock speeds, there was a 10C difference at 240FSB on the 3500+, that's 42C@ 2.64GHz, on air! In my opinion nothing short of amazing considering the issues with 90nm technology to date.

Rich's Thoughts

AMD Athlon 64

I didn't mean for this to seem like an anti-Intel article; Intel was the first to make the leap to 90nm technology, and being the first opens you up to both criticism and praise.

The issue with the Prescott has always been thermal, and to a lesser extent its lack of performance at rated speeds. I also feel that the technology involved was a little before its time as very few programs are actually optimized to use the extra pipelines and SSE3 technology.

I have personally owned three Prescott CPUs, a 2.8 and a 3.2GHz in the socket 478 and an unlocked testing sample in the socket T which I run at 3.6 GHz (This will be the subject of a future article); both socket 478 CPUs overclocked extremely well in my opinion. I ran the 2.8E 24/7 at 3.5GHz (until its untimely demise at my hands while trying to remove the IHS), I have also had the 3.2E clocked over 4GHz and run it at 3.71 GHz 24/7 without issue. In my opinion Prescott CPUs overclock extremely well and handle multitasking better than any AMD CPU I have ever used.

The only issue that I had, as did most everyone else who purchased a Prescott, was cooling. With the stock Intel HSF both 478 Prescotts idled into the 50C range at rated speeds, eliminating any chance of overclocking without extreme cooling. Once on water though, both CPUs scaled very well, the problem with this obviously is cost. Extreme cooling costs money, sometimes a lot of it. From $50 for a top end air-cooled solution to well over $200 for a good water cooled system, the Prescotts were not very cost effective, and in no way justified their price to performance trade off.

So far I have to also say that Intel's second generation 90nm CPU shows no signs of managing its thermal issues any better, even with an improved stock HSF. The CPU I have right now is clocked at 3.6GHz (multi-set to 18 and FSB at 200), and it idles at 53C!

I have been using a 130nm 3800+ Newcastle for about 4 months now, and while it is a very good processor, it doesn't have near the multitasking ability of the Prescott. However, it simply flies during gaming.  My issue with the Newcastle was the same as the Prescott, but to a far lesser extent. My 3800+ idled at 37-38C and easily broke 50C while under heavy load, limiting what it could achieve while air cooled.

After having the 3500+ Winchester for almost two weeks I am very impressed with this CPU.  It incorporates all the positives of the Newcastle it replaces, with a huge drop in operating temperatures. While some people will complain that there is virtually no performance boost over the 130nm technology, the optimists amongst us will look at an AMD processor that was 10C cooler at 2.64GHz, running without a hiccup. My system had some limiting factors that I need to iron out, but I believe this CPU has even more headroom for overclocking.

Those of you that already have an AMD 130nm CPU will see little if any benefit from upgrading to the Winchester, but those of you who are looking for a very good processor to power your system until we see dual core some time next year need to look no further than AMD's Winchester series.