Date: August 28, 2006
Author(s): Rob Williams
It’s a great time to be building a dual core system, due to great competition between AMD and Intel. We are taking a look at the AM2 4600+ which proves to be a decent value due to recent price drops. Does it warrant a purchase, or should you be looking elsewhere?
In an age where computers and hardware are getting bigger and faster, there’s a recurring problem: The new part is hotter than the last. This is not always the case, but I have to laugh at my Intel 820 D, a 110W chip, that hits 70°C at 4.0GHz with water cooling. Needless to say, Intels are known for being hot to begin with, but that’s ridiculous! AMD on the other hand though, have never been as bad as that, but that doesn’t mean what’s there is perfect either. Dual core CPU’s in general will get far hotter than a single core, sometimes near twice as hot.
Heat is not the only reason Energy Efficient processors are beneficial though. Power consumption is on the rise, and anything to help ease the pain on the power bill is welcomed. Simply looking around any e-tailer today, you will see 1000W power supplies by numerous companies. This is the direction we are going, and anyway you look at it, 1000W is a lot of freaking power.
Both Intel and AMD have been doing incredible jobs at weaning the overall power requirements for their CPU’s down. Intel has even had 1.0GHz chips that only used 5W of power at 100% load. If that’s the future… then it’s going to make things much easier for the consumer… both for the wallet and the room temperature ;-)
Shortly after AM2 was launched in May, AMD released three different versions of the entry-level X2, the 3800+. The standard chip, like the other non-EEs, runs with 89W. Even that is not so bad when considering the 4600+ 939 version is rated for 110W. However, it gets even better due to the fact that there is also a 65W and 35W version of the 3800+. If only we could get the rest of the line down to 35W!
Recently, I wanted to purchase an AM2 machine primarily for memory benchmarking. The on-die memory controller is by far the best out there today; it can really push memory bandwidth records to the next level. When the time came to purchase, I decided on the EE for one reason, and one reason only. I had figured with a lower power/voltage requirement, the chip would likely have better chance of success during overclocking. I am not quite sure if this is true, but I will touch up more on this in the overclocking section.
Before going any further, lets take a quick comparison look at the 3800+ EE, 4200+ EE, 4600+, 4600+ EE and the 5000+ processors.
Besides clock speed differences, we can see the normal chips (5000, 4600) have a 89W power requirement, and the EE’s, with the exception of the 3800+, use 65W. Due to the different thermal design, the 4600+ EE has a higher max temp of 72°C. This is a good thing, as I have had this chip run that high on the stock cooler. The most notable thing here is that the 4600+ requires 1.20vcore, while the standard 4600+ requires 1.30vcore. If anyone out there has a standard 4600+ and has run it on 1.20vcore reliably, please let me know. I am curious as to whether the EEs actually have a greater advantage in this regard.
When talking about dual cores today, there’s a problem that lingers. I have people often ask me if they should pick up an X2, or a C2D. This is one of the hardest questions to answer right now, because C2D is cleaning up the benchmarking reports all around the web. Personally, if I were to go out any buy a Dual Core right now, I have no definitive answer on which I would buy. Here’s how I would sum up this buying predicament:
Core 2 Duo as it stands will cost you more money. The CPU’s start at $210, and a good motherboard is around the same. On the other hand, there are some great AM2 motherboards out there in the $150 range, and the 3800+ goes for the same. So it’s around $100 less if you choose to go with the budget solution for AMD, over the budget solution for Conroe.
While both are great performers, it’s evident that C2D is the better performer any way you look at it. The many reviews last month proved that fact. Even the ‘small’ E6400 which retails for around $300, went head to head with the $1,000 FX-62. So, what reason would you want to go with AMD after knowing this? The cost would be a good reason, or maybe even the superior memory controller that AM2s offer. Even though C2D’s have lower voltage requirements, I am unsure about which is cooler. If it’s anything like my previous netburst though, they can get rather hot.
If money is no object and you want only one computer, then the C2D is a great choice. For a budget minded consumer that wants dual core and doesn’t require a chip that’s top of its class, then smaller AMD X2’s is the choice to make… especially after recent price drops.
Ok, that rant is out of the way. Let’s take a look now at the overclocking ability of the 4600+ EE.
Upon the first boot into Windows, CPU-z reported 1.27vcore, which was higher than I expected. This is a 1.20v/1.25v chip, so I used AiBooster to lower it a bit. It turns out that the voltage fed to the BIOS is not likely to be the value in the end. I had to set the voltage through AiBooster to 1.175v in order to hit a 1.216v, according to both Ai and CPU-z.
From this point, I ran two instances of Prime95, and allowed them to run through 5 tests. If those proved successful, then I increased the HTT by 5 (5*12 = 60MHz) and let the tests continue. If I ran into errors, I increased the voltage by 0.125v and re-ran Prime95. I continued to increase the voltage and clocks like this until I found my max stable overclock.
One fun part about overclocking Dual Cores (sarcasm ahead) is the fact that whenever you “think” you have found a stable overclock, your computer laughs right at your face. In binary of course. Even if you have a great overclock and Prime95 has no issues, it’s probably because you only have one copy open. With both dual core CPU’s I have here, both exhibit the same issues. For example… If core 1 runs no problem at all, stressing both cores at once will likely cause the program running off core 2 to crash.
I usually would attribute this to being a weak core, however running a single instance on either core will run smoothly. It’s only when both cores are stressed at once, that the second will give way to errors. That being said, I only consider an overclock completely stable if both cores are stressed and don’t error out. Here are my results:
Stock voltage was fine for adding 132MHz to the frequency, but anything further required me to increase it. Once I hit 2.68GHz, I had to jump the voltage up quite a bit… all the way up to 1.41v. Still not too bad considering the clock. The max stable overclock I have found was 2.76GHz using 1.46v. This means, two instances of Prime95 ran for 8 hours without any issues. 2.85GHz is stable only on one core… the second instance of Prime95 would crash quick. I have tested the computer quite a bit using 2.85GHz though, and haven’t found an issue other than Prime95. I will use that setting in our testing. The room for testing currently hovers around 87°F, and I assume that there would be a greater chance of a stable overclock at 2.85GHz if it was more like 75°F.
Of course overclocking doesn’t matter much without a performance report, so let’s hop to it.
Sandra is a great tool to have because it includes many benchmarks that stress all of your components. The first two, Arithmetic and Multi-Media, rely heavily on your CPU and are designed to take advantage of more than one core. So, the higher the frequency, the better the score. That’s evident in the charts… the results scaled nicely with the overclocks. On the multi-media side of things, between stock and 2.76GHz we have a 14.5% performance increase. Not too hard to believe since it’s a 15% overclock though.
Everest, like Sandra, is another good tool to use in order to see how your CPU overclocks scale. However, the only thing I don’t like about the program is that each new version cannot be compared to the last, due to ever changing algorithms.
Once again, we are getting pretty congruent results here. ~14.5% performance increases for a 15% overclock. I am not paying attention to the top overclock as much as 2.76GHz for the sake of it being stable on only one core. However, if you manage to get a 4600+ or similar chip that clocks to 2.85GHz reliably you will know what to expect.
Sciencemark is yet another awesome benchmarking tool to have, because it’s very, very in-depth for those who want the nitty gritty of everything. All results are in seconds for time to compute. Yet again, we are seeing 14.5% – 15% increases, as one would hope. We didn’t manage to break 9s for a cipher attack, but it came quite close.
PC Mark and 3D Mark are my least favorite benchmarking programs to use, but they are decent in finding out how your overclocks scale with one another. Only problem is I find, that with each run of the program, even at the same clock speed, scores can fluctuate too much.
That aside, the results scale rather closely to what we have seen in the last few tests… between 13% – 15%. The only exception would be 3D Mark 01, with a 6% performance increase.
Cinebench is a good tool for those who plan on using their computer for multi-media production, because it renders a high resolution image in real time. Luckily enough, it supports multi-threading, so you can see just how much of a gain that dual core CPU gets you.
Yet another string of ~14.4% increases. Not bad for such a simple overclock with 1.45v!
Time to eat some Pi! Super Pi, for the uninitiated, calculates Pi to a set amount of digits past the decimal. So, for a 1 Million run, it would be 3.14359 + 999,995 more digits. Simple enough, but it does a great job to see how CPU overclocks scale. In this case, ram timings help a lot also. We are again seeing a steady 14% increase throughout all of the results, though we couldn’t break 30s 1M with our safe overclock. We did with our 2.85GHz overclock though, but again, that’s rather unstable on one core.
For some real world testing, I used LAME MP3 encoding and also DVD re-encoding. First, I ripped the first CD from DJ Tiestos latest album, In Search of Sunrise 5, as a single .WAV file, which weighed in at around 800MB. I then used both LAME and LAME MT (Multi-Threading) to convert it. The command I used was lame -b 320 *.wav *.mp3. Simple.
For DVD re-encoding, I took my Bad Religion – Live At The Palladium concert DVD and re-encoded it using Nero Recode. I took a real world scenario of backing up a DVD… I took the regular 8.7GB DVD and compressed it to 4.5GB so that it would fit on a normal DVD-R. The test only counts the time it takes to recode the movie, not the time it took to rip it first. Happily enough, both LAME and Nero keep track of the time for the process, so figuring out the final values was easy. All results in the graph are in seconds, and lower of course is better.
The results are pretty clear. We experienced a 14% increase in LAME, 13.5% in LAME MT and 10.5% in Nero. The stock performance was also nice, but a simple overclock easily adds 10% – 15% performance increase, which is welcomed if you are working on larger projects.
For a gaming test, I used one of my favorite games, Half-Life 2. I choose this game, because it’s very CPU dependant, and also quick to get into. The fact that I’ve played through it multiple times also helps. I ran the game with a 1024*768 resolution with default settings (0x AA, 0x AF), and used FRAPS to record the average and minimum frames per second. Each run is manually played through, no timedemos are used. Due to this, each run will always be different than the last, but this gives a real world result. I used d1_canals_07 as the level of choice.
If you don’t believe that HL2 relies heavily on the CPU, here is a graph to debunk any doubts. Our overclock gained us just under a 16% performance increase.
Overall, I am impressed by the results seen today, although all were expected. In some of the tests, such as Sandra and Lame MT, it’s clear how beneficial the extra core can be. Overall though, I can’t say I am particularly impressed by this specific model, overclocking aside. But, it all depends on how you judge things.
I never, ever buy a CPU and don’t plan to overclock it, but there are many out there who will indeed buy a $1,000 chip instead of overclocking a $300 one. That being said, as it stands, this 2.4GHz DC processor is great. The fact that it requires less voltage than a regular 4600+ makes it better suitable for many people. It should aide in a lower power bill and also heat. Of course in the Winter it could actually pay off heating your house with a dual core CPU instead of the heaters.
From an overclocker/performance perspective, there are better CPU’s to choose from. Energy Efficient aside, there are cheaper X2’s that should perform just as well. One friend has an AM2 4000+ that is stable overclocked to 3.0GHz with 1.45v. Clearly, he wound up with a better chip than I did! Though 4000+’s are hard to find now, the 4200+’s are not. Comparing the non-EE 4600+ to the 4200+ on a popular e-tailer proved $79 more expensive. Now, this is not a big deal, really. However, for the $79 you save, you should still be able to get some great clocks out of this thing, and could likely even beat the max stable I have acquired with this chip here.
If you are really money conscious, then it may pay off to take a look at the 3800+ X2 also, which retails for around $150US. In the end, there is a lot of choice, which is a good thing. The entire line of X2’s are well priced, but if you are to buy one now, I’d recommend either the 3800+ or 4200+. As for EE chips, I would recommend one only if you plan using at stock speeds. As we have found, the fact of EE didn’t seem to help in our overclocking results of the 4600+.
When it’s all said and done, I am awarding the 4600+ EE a 7 out of 10. If you are looking for a CPU strictly for stock speeds however, then it would warrant an 8 out of 10 :-)
If you have a comment you wish to make on this review, feel free to head on into our forums! There is no need to register in order to reply to such threads.
Copyright © 2005-2019 Techgage Networks Inc. - All Rights Reserved.