Introduction:



As you probably already have noticed, Madshrimps did not publish an article regarding the Phenom II's performance in comparison to competitive products. There are two main reasons: on one hand, we only received our Phenom II sample a few days after the product launch, which means that most hardware reviewing websites already had an article regarding the Phenom II. Ours would just have been 'yet another' comparison and would only serve to confirm everything we already know. On the other hand, we had the possibility to test the widely hyped overclockability of the processor and check if the rumors about 6GHz were indeed true.

We were actually surprised by the excellent overclocking scaling of the Phenom II X4 940 on LN2! As you may or may not know, our retail (but sent by AMD) sample went beyond 6GHz quite easily and there was absolutely no sign of any coldbug. Unlike its predecessor, the Phenom II (also referred to as Ph-II) is indeed meant for overclocking ... well, extreme overclocking. In 24/7 configurations, it's still very difficult to reach 4GHz stable on normal air cooling, which doesn't match the current C2D overclocking capabilities.

These were our findings of our performance scaling article on AMD’s Phenom I:

"If you're planning to spend a couple of minutes on overclocking, you should go for the raw cpu megahertz overclocking ... which is fairly obvious. Everything else gives a nice boost, but that's it; no real extreme increases in performance. Note that it's important how you interpret the graphs: this is a performance scaling investigation, not an overclocking scaling investigation. Please understand that the given increases are expressed in comparison to a very low base frequency, not at stock speeds. For instance, the performance scaling of the HT Link frequency may seem very impressive in the 3D benchmark, but if you look more closely, you'll notice that 1GHz is almost enough to maximize performance."
(~ AMD Phenom In-Depth Performance Scaling Analysis)

We already had a peek at the overclocking performance scaling of the Phenom I. In the article, you were able to find out that the processor frequency had the biggest effect in performance, which is pretty obvious, and that overclocking the IMC frequency is actually quite interesting to get the absolute most out of your machine. Since we got a lot of positive feedback about the first article, we did the exact same for the Phenom II. In this write-up, you will find out what's the impact of each overclocking variable performance-wise.

Test setup

We prepared the following test system to be used for all performance results mentioned on the next pages:

Madshrimps' AMD Test Setup '>
CPU	AMD Phenom II X4 940
Cooling	Noctua NH-U12P
Mainboard	DFI Lanparty DK 790GX-M2RS
Memory	2 * 1GB G.skill F2-8500CL5S
Other	Sapphire 4870X2 Antec 1000W PSU Western Digital 320Gb SATA HDD Windows XP SP3

Methodology

The following benchmarks were used:

Lavalys Everest: Memory latency

SuperPi 1M

Wprime 32M

3DMark2001SE: Nature

HT Link frequency:



Before we start analyzing the obtained results, let's find out what the HT Link is designed for.

In Intel-terms, the HT Link frequency is quite similar to the FSB: taking care of the communication between processor and Northbridge, which features for instance the PCIe lanes, sata ports and audio support. The faster the link between processor and Northbridge, the more data can be transferred between both and, in theory, the higher the performance will be. Without going to much into detail, it's necessary to know that the current HyperTransport link is clocked at a frequency of 1800MHz, which is 1000MHz higher than the first edition of the HT link used on the s754/s939 Athlon64 platforms. Note that although the original HT Link frequency was set at 800MHz later revisions of the s939 motherboards had a stock HT Link frequency of 1GHz.

Test settings:

Core frequency: 3000MHz

HT frequency: 200MHz

Northbridge frequency: 1800MHz

Memory frequency: 400MHz (1:2 divider)

Memory timings: 4-4-4-10 2T Ganged

Variable:

HT Link frequency (200 - 1800MHz*)


(Click for bigger version)

The HT Link frequency is not what what we'd call interesting in terms of overclocking for performance. The difference in performance between very low and high frequencies is in most benchmarks non-existing, although we see a noticeable performance increase in the 3D benchmark. Note that the effect of overclocking the HT link frequency depends on what type of video card you're using. IGP's and high-end video cards (for instance dual gpu cards) will benefit more from HT Link overclocking, because they rely on high pci-e transfer rates. Low- and middle-end cards, which are less powerful and have a lower transfer rate, will benefit a lot less.

*: Because of the NB frequency of 1.8GHz, we were not able to test HT link performance scaling beyond 1.8GHz. The HT Link has to be equal or lower than the NB frequency at all times due to the design.

Memory frequency performance scaling:

These days there are two different types of common desktop platforms: the ones using a cpu with Integrated Memory Controller, such as AMD's Athlon64/Phenom and Intel's Core I7, and the ones with the memory controller inside the Northbridge. For the non-tech minded people under us, this may not seem to be that important -memory is still memory- but if you have a closer look at it, it actually does matter. Using an IMC has many advantages, the most important one being higher bandwidth and lower latencies, which will be translated into higher performance. And that's not so strange: the data lanes between memory and cpu, which serve as communication ports, are much shorter than those in an non-IMC platform. An example:

AMD Phenom <=> DDR2

Intel C2D <=> P45 <=> DDR2

In the first example, the latency is only bottlenecked by the length of the lane between cpu and memory, whereas in the second example, the latency is bottlenecked by both lanes between cpu and Northbridge and Northbridge and memory.

Test settings:

Core frequency: 3000MHz

HTT frequency: 200MHz

HT Link frequency: 1800MHz

Northbridge frequency: 1800MHz

Memory timings: 4-4-4-10 2T Ganged(*)

Variable:

Memory frequency: 200, 266, 333, 400 and 533MHz


(Click for bigger version)

The effect of an increase memory frequency is only noticeable in memory intensive benchmarks, or better put: memory benchmarks. In real-life applications or benchmarks for non-memory hardware you will see no significant increase in performance.

(*):Due to the design of the K10 processor, it's impossible to use memory timings below 4-5-5-X using the 533 divider. We used 4-5-5-16 in the comparison.

IMC frequency performance scaling:



As I already mentioned on the previous pages, the AMD processors have their memory controller build inside the processor to produce higher bandwidth and lower latencies. The memory controller has its own frequency, not linked to either CPU frequency or even memory frequency, but linked to the HTT frequency, which is the base clock frequency from which all other frequencies are derived. Using a multiplier, you can increase or decrease the frequency of the memory controller to obtain more performance.

Test settings:

Core frequency: 3000MHz

HTT frequency: 200MHz

HT Link frequency: 1000MHz

Memory frequency: 400MHz (1:2 divider)

Memory timings: 4-4-4-10 2T Ganged

Variable:

Northbridge frequency: 1000-3000MHz


(Click for bigger version)

The biggest increase in performance can be found in the memory benchmark that measures the memory write bandwidth. In addition to the obvious increase in performance of the memory, we notice that the 3DMark01 benchmark also scales quite nicely: 20% up in FPS going from 1GHz to 3GHz.

For the people who are interested in tweaking their configuration for optimal SuperPi 1M performance, we can tell you that the IMC has a great effect on the so-called efficiency of your system. Underneath you can find a 1M result with an efficiency of 675xx, which is at the moment one of the most efficieny 1M runs on record.


(Clickable)

In overclocking circles, the efficiency is often mentioned when discussing the overclocker's skill to tweak his system. The 1M efficiency is calculated as follows:

Efficiency = Time (sec) x Frequency of processor (MHz).

The lower this value, the better as you need less cpu power to end the benchmark at a given time!

CPU Frequency performance scaling



Last but not least, the overall CPU frequency performance scaling chart. Ever since Intel released its Core 2 Duo processor series which showed OC scaling over 50% even with stock cooling, we re-evaluated our OC results with AMD, on average a 20% OC is far from bad, but not at all impressive compared to the Intel counterpart. The main problem though is the low overclockability when you want more than just 24/7 stable settings, when you're aiming for absolute high-end performance and need to squeeze every last megahertz out of your system ... something that can be very tricky on AMD systems.

Why? Well, most processors won't hit 3,5GHz on air cooling and even with exotic cooling solutions you won't hit far over 4Ghz if you don't have a so-called gem chip, basically a very good overclocker. Partially because of the coldbug problems, which has been every overclocker's nightmare for many, many years ... well, actually since AMD switched to the IMC with the Athlon 64.

Test settings:

HTT frequency: 200MHz

HT Link frequency: 1800MHz

Northbridge frequency: 1800MHz

Memory frequency: 400MHz (1:2 divider)

Memory timings: 4-4-4-10 2T Ganged

Variable:

Core frequency: 800-4000MHz


(Click for bigger version)

Of course, we expect the cpu frequency to make the biggest difference in all benchmarks, except maybe Everest as that's a real memory test. It's good to see that the Phenom II is actually still scaling well beyond stock frequencies; if we're correct, it should even be scaling pretty well beyond 4GHz.

HTT Frequency performance scaling

The HTT frequency is the base frequency of which all other frequencies (processor, ht link, memory, northbridge) are derived. Standard, the HTT frequency is set at 200MHz, which is more than the base clock of the i7, which is set at 133MHz. However, in contrary to what you'd expect from frequencies in computing hardware, more does not equal better by definition. The base frequency is not directly related to the performance of your processor, it's what determines the clock speed of your components. To explain with a small example: 400x6 will always be slower than 200x20. The higher the base clock frequency can be clocked, the more room one has to tweak his configuration, though, so in that aspect more does indeed equal better.

Unlike with the first Phenom series, the Phenom II is again pretty decent in terms of HTT overclocking. I had no issues with booting around 300MHz, but 330MHz was the limit. Other overclockers have been able to reach over 400MHz, though, so I reckon we haven't seen the end of this yet.

Test settings:

Core frequency: 3600MHz

HT Link frequency: 2400MHz

Northbridge frequency: 2400MHz

Memory frequency: 400MHz (1:2 divider)

Memory timings: 4-4-4-10 2T Ganged

Variable:

HTT frequency: 200 - 240 - 300MHz


(Click for bigger version)

What we expected also came true: almost no gain going from 200MHz to 300MHz, which is a 50% increase in clockspeed. That increase gives you at most 3,5% performance increase and that only in one benchmark, the rest is still far below that number.

Gathering all information

Okay, let's sum up. If you want more performance, this list will give you a guideline for what to focus on first:

• CPU frequency
  
• IMC frequency
  
• Memory frequency
  
• HT Link frequency
  
• HTT frequency

In other words, you will see the biggest performance gain from overclocking the processor frequency, which is actually pretty straight-forward. In contrary to this list, I would suggest a different approach in the actual overclocking attempt: finding limits is much easier if you know the size of the playing field. In more hardware-related terms, it's important to know the limits of the HTT frequency and, preferably, the memory frequency as well. Only when you know the maximum HTT frequency, you are able to squeeze those last few MHz's as the higher the HTT, the lower the multiplier will be. The lower the multiplier, the smaller the steps you'll be able to take when overclocking (using the HTT frequency method).

Conclusive thoughts

The people who often visit this website will probably already have noticed that I've been playing with this Phenom II technology for quite some time now. Actually, even before I started conducting the tests for this article, I have been pushing the limits of the technology and, as I announced in my Phenom I performance scaling article, it's indeed possible to hit over 6GHz with the Phenom II processor, be it on LN2 though. However, the actual absolute frequency has very little significance, what's really important is that we can finally play with an AMD again. In addition, the relative frequency (obtained frequency weighed against the stock frequencies) is again something that can be spoken positively about: Stock frequency is 3Ghz, but this time it can be pushed beyond 4GHz on air cooling, beyond 5GHz on single stage cooling and beyond 6GHz on LN2. The Phenom I only offered 2.6GHz stock, 3.2GHz air cooled, 3.5GHz single stage and ... no LN2.

I've heard rumors that AMD is still improving the production process of the Phenom II technology, which means that the AM3 products might even be able to overclock even better! We received our first DDR3 motherboard a few days ago, so we will have a look at the new processor series soon. In any case, there will be a Phenom II X4 955, which will be clocked at 3.2GHz at only 1.25v!

Any hardware enthusiast and overclocker should at least give the Phenom II a shot, certainly when you get chance to cool it down below zero. It's just so much fun, not having to watch the temperatures as there's absolutely no coldbug!

---

I hope you enjoyed reading this OC report, until next time, click the logo below to read up on our previous overclocking endeavors: