Building A Budget QuadFX System
The AMD QuadFX platform hasn't made too much noise since it was launched a couple months ago, but the reasons are obvious to all. It's expensive to build, consumes energy like a muscle car and there is only one motherboard to pick from for those thinking of building a QuadFX system. When the QuadFX processors came out on November 29, 2006 they were expensive and not a great value for the level of performance they produced. Now over a month later and after rumors of slow sales on QuadFX systems, one would think that prices would go down on the parts, but they have actually gone up! Looking at an online retailer like Newegg it's obvious that each pair of QuadFX processors has gone up ~$100 across the board.
|Processor||Original Price||Current Price||Price Increase|
AMD Athlon 64 FX-74
AMD Athlon 64 FX-72
AMD Athlon 64 FX-70
When we posted our original article we were told by AMD that the QuadFX series of processors would be paired together and that it wouldn't be possible to buy them individually. What really happened is the opposite of this. Retailers are selling them individually as they never shipped together in a single box. When pricing these Socket F processors remember to multiply the price by two! Regardless of the price increase, we got our hands on a pair of FX-70 processors to see how they would perform. Legit Reviews has already done an in depth article on the FX-74 processors the day they came out so be sure to read that review before going any further if you haven't already.
The pair of FX-70 processors look no different than the FX-74 processors and to be honest there is no difference at all between the parts other than what tests they passed before they were labeled. The FX-74 was able to run at a higher frequency at a lower voltage and was binned the flagship processor at the factory. The processors that didn't make the cut were again tested at lower speeds and if the passed that is what they get named. Since all AMD FX series processors are factory unlocked with a little luck, or more voltage, we should be able to get our FX-70 processors running at FX-74 clock speeds in a matter of seconds, but before we talk about that another minor QuadFX update that needs to be brought to light.
Since our original article published a new BIOS has come out that will enable the Asus L1N64-SLI WS motherboard to support 4-pin CPU heatsinks like the Ajigo MF091-097 that AMD just recently sent out. These Ajigos heatsinks replace the pre-production heatsinks that came with our “4x4” system in its original configuration. The new Ajigo heat sink (pictured on the right) is a bit smaller than the original ones that we were sent and feature a 4-pin fan versus the 3-pin fan design. After we updated the BIOS and install the new heatsinks, the machine runs much quieter as the fans now properly modulate their speed based on thermal feedback from the CPU to the motherboard. The result is a system with an acceptable noise level for what we consider a workstation-class machine. No significant difference in temperature was noted between the two sets of heatsinks.
Overclocking The QuadFX System
With the latest BIOS (0124) installed on the ASUS L1N64-SLI WS motherboard and the new 4-pin reference heat sinks with thermal pads installed on the test system we started overclocking to see how high we could get the system with no modifications at all to any of the components.
The starting point with all defaults loaded in the BIOS was 2.6GHz at 1.3V according to CPU-Z 1.38. Let's see if we can make our FX-70 an FX-74 by adjusting the multiplier up to 15 and keeping the HyperTransport bus speed the same.
We quickly found out that using a multiplier of 14 required no changes at all, but to hit 15x we had to increase the voltage on the board to the highest setting possible in the BIOS (1.35V). In a matter of minutes we had overclocked our FX-70 to FX-74 frequencies.
Knowing that the processor could then be overclocked higher by increased the bus speed we were able to get another 10MHz with a little more tinkering in the BIOS. Speaking of the BIOS, let's take a look at the settings that were changed to reach this 654MHz overclock.
All the settings that were changed were found in the 'advanced' menu under the system frequency/voltage sub menu. The first thing to do is change AI Overclocking and CPU FID/VID Control from Auto to Manual. Once these have been changed to manual all the settings we need will pop up. Notice the CPU voltage needs to be changed in two spots on this board. Under CPU FID/VID Control the CPU Voltage goes up to 1.350V, which wasn't enough to hit more than 3GHz on our test system. To reach above 200MHz on the FSB frequency we have to go down to the general voltage section and max out the CPU 1/2 VCORE Over-Voltage section. By adding 200mV to each processor we were able to reach 210MHz x 15 with some stability and 207MHz x 15 with full system stability. This is also the menu where you can mess with all the other voltages on the board. We didn't have to change any other voltages, but since we were running 4GB of CL4 memory we bumped up the voltage on the modules to 2.0V, which is not much more the suggested memory voltage.
Interesting Fact: Many motherboard vendors frequently mis-label the main clock signal on the motherboard as the “FSB” when in-fact there is not “front side bus.” It’s just the main 200MHz signal that other signals (like the CPU) are driven off of. So if you are wondering why we are talking about the FSB when AMD doesn't have one this is why. Since the pictures show FSB we say that, since we don't want to confuse our readers any more than they already are.
For Those That Like To Drop The Multiplier
By dropping the multiplier we were able to reach 263MHz x 11, which was rock solid and got the system at 2.9GHz. When comparing the benchmark data from both overclocks it seems that the overall clock frequency matters most and the system running 207MHz x 15 meant better benchmark numbers than 263MHz x 11. Now that we which overclocking method works the best for our system let's take a look at performance numbers.
Remember that our overclocking numbers will vary from yours as every platform is different. We used the stock heatsinks and left the thermal pads on them, so more than likely you'll see better numbers than ours with improved cooling.
The Test System
All testing was done on a fresh install of Windows XP Professional build 2600 with Service Pack 2 and DirectX 9.0c. All benchmarks were completed on the desktop with no other software programs running. All of the modules were run in dual channel mode! The memory on the Intel and AMD AM2 test platforms was run with Corsair PC2-6400C3 memory running 4GB at 3-4-3-9 2T timings at 800MHz. The Intel quad-core system was run at 3-3-3-9 2T thanks to improved memory performance on Bad Axe 2. The memory on the AMD DDR1 motherboard was running 4GB at 400MHz with 2-3-2-6 1T timings. The memory timings on both of the Quad FX system was set to 4-4-4-12 2T. Please reference our original QuadFX article for specific details on the other test systems.
|AMD QaudFX FX-70 Test Platform|
AMD Athlon 64 FX-70 Open Box
ASUS L1N64-SLI WS
4 x Corsair PC2-8500C5 Dominators
2 x XFX GeForce 8800 GTX SLI
2 x Western Digital 500MB
PC Power & Cooling 1KW
Thermaltake Armor Extreme
POV-Ray 3.7 Beta 13aProcessor Performance on Pov-Ray 3.7 Beta 13a:
The Persistence of Vision Ray-Tracer was developed from DKBTrace 2.12 (written by David K. Buck and Aaron A. Collins) by a bunch of people (called the POV-Team) in their spare time. It is an high-quality, totally free tool for creating stunning three-dimensional graphics. It is available in official versions for Windows, Mac OS/Mac OS X and i86 Linux. The POV-Ray package includes detailed instructions on using the ray-tracer and creating scenes. Many stunning scenes are included with POV-Ray so you can start creating images immediately when you get the package. These scenes can be modified so you do not have to start from scratch. In addition to the pre-defined scenes, a large library of pre-defined shapes and materials is provided. You can include these shapes and materials in your own scenes by just including the library file name at the top of your scene file, and by using the shape or material name in your scene. Since this is free software feel free to download this version and try it out on your own.
The most significant change from the end-user point of view between versions 3.6 and 3.7 is the addition of SMP (symmetric multiprocessing) support, which in a nutshell allows the renderer to run on as many CPU's as you have installed on your computer. This will be particularly useful for those users who intend purchasing a dual-core CPU or who already have a two (or more) processor machine. On a two-CPU system the rendering speed in some scenes almost doubles. For our benchmarking we used version 3.7 as all of the processors we are testing today are dual-core.
Once rendering on the object we selected was completed, we took the score from dialog box, which indicates the average PPS for the benchmark. A higher PPS indicates faster system performance.
The pixel rate counter (PPS) in POV-Ray is based off of the number of pixels rendered in the current frame divided by the total amount of time spent on the whole animation. This gives the effect of dividing the true pixels per second by the current frame number. With POV-Ray 3.7 Beta 13a we are able to look at a recent SMP benchmark to judge the differences between AMD and Intel dual-core processors. The benchmark shows that the Intel Core 2 Extreme Processor X6800 is over three times faster than the previous generation Intel Extreme Edition processor. The old Pressler core is no match for Conroe in this benchmark! In a benchmark that AMD used to own it is now getting beat. The QuadFX FX-70 system that we are looking at today comes in near the bottom with a score that isn't too impressive.
Looking at the overall render score the winner is clear by a long shot. The Intel Core 2 Extreme processor QX6700 blows the other processors away and more than doubles the performance levels seen on the AMD FX-62 processor, but once again take note that the Quad FX system places second and third. When our system was overclocked from 2.6GHz to 3.1GHz we saw a nice increase in the score.
POV-Ray Real-Time Raytracing
Legit Reviews was e-mailed by one of the developers over at POV-Ray to see if LR could include real-time raytracing in our performance analysis of Kentsfield and Quad FX. After spending a bit of time to get the beta software to work correctly LR has some of the only real-time raytracing numbers available for Conroe, Kentsfield and Quad FX platforms.
E-Mail From POV-Ray -- I thought I might ping you about an experimental feature we've added to the POV-Ray SMP beta: real-time raytracing. It's mostly useful to folks who have multi-core systems and in fact is something that I've wanted to do for years but the hardware just wasn't there (at least not in the consumer price range). It works best on a kentsfield or later, but a core 2 duo should be sufficient if you don't mind sub-10fps frame rates.
If you want to try it out it please feel free to grab it from: http://www.povray.org/beta/rtr/
This experimental software by POV-Ray was a welcomed addition to our testing and was able to spread the work load across the four processors as seen above. The number that is used to gauge performance is shown in the status bar at the bottom of the main window as seen in the above image of with Intel Core 2 Extreme QX6700 taking care of the rendering.
The Intel QX6700 processor takes the lead at default clock speeds and when overclocked it shows without a doubt what CPU is the best on this benchmark. The AMD QuadFX FX-70 test system did well and came in at 10.95 Frames Per Second. When overclocked it was able to pass up the more expensive QuadFX FX-74 system thanks to the 7MHz boost on the front side bus during the overclock. Here we see where an overclocked to the max QuadFX system still can't touch the 2.66GHz Intel QX6700 at all default speeds.
RARLAB - WinRar v3.61 has a multithreaded version of the RAR compression algorithm, which improves the compression speed on computers with several CPU, dual core CPU and processors with hyperthreading technology. Multithreading is enabled by default, but you can disable it in "General" part of "Settings" dialog.
When we ran WinRAR on the Intel Core 2 Extreme Processor QX6700 we noticed that it used roughly 80-84% of the 2.66GHz quad-core processor, so it makes for a decent SMP benchmark. Before we take a look at the multithreaded testing results let's take a look at the single threaded performance chart.
When running WinRAR v3.61 on a number of processors it was noted that Core 2 processors took the lead once again. The AMD Quad FX system gets killed in the single CPU test as the memory issue with NUMA has an impact on the benchmarking numbers. When overclocked the few extra MHz on the memory really helped the performance numbers.
When SMP was enabled the performance greatly improved on all of the processors. The Intel Core 2 Extreme QX6700 processor easily took the lead thanks to the fact that it had four cores and passed up the X6800 that had the highest score in the single core test.
MAXON; CINEBENCH 9.5:
CINEBENCH is the free benchmarking tool for Windows and Mac OS based on the powerful 3D software CINEMA 4D. Consequently, the results of tests conducted using CINEBENCH 9.5 carry significant weight when analyzing a computer?s performance in everyday use. Especially a system?s CPU and the OpenGL capabilities of its graphics card are put through their paces (even multiprocessor systems with up to 16 dedicated CPUs or processor cores). During the testing procedure, all relevant data is ascertained with which the performance of different computers can subsequently be compared, regardless of operating system. Again, higher Frames/Second and lower rendering time in seconds equal better performance.
Here in Cinebench 9.5 we see the AMD QuadFX FX-70 system hold it's own and when overclocked it did very well.
Our SMP Cinebench results show Intel QX6700 and the overclocked AMD QuadFX FX-70 finish at the same exact time. Not bad as this benchmark loves all the cores that it can run.
3DMark06 includes an array of 3D graphics, CPU and 3D feature tests for overall performance measurement of current and future PC gaming systems. With this broader design approach, 3DMark06 has become the benchmark of choice for all PCs with top-of-the-line graphics hardware and CPUs. 3DMark06 is the first product from Futuremark using the AGEIA PhysX software physics library in two very complex, game-like threaded CPU tests conceived to measure properly performances of single processor, multi-core and multiple processor systems in next generation of games. In addition to using real-time physics, both CPU tests also employ multi-threaded artificial intelligence algorithms. By combining the results of the two CPU tests and four graphics tests, 3DMark06 enables users to get a 3DMark score which reflects the overall gaming performance of their PC.
With both processors at stock speeds we were happy to see both processors and platforms hitting overall 3D mark scores like the ones above. These GeForce 8800GTX's can take every CPU MHz you can give them and with the QuadFX platform overclocked we saw a nice gain.
In the CPU test the Intel QX6700 was able to best the QuadFX FX-74 platform by nearly 600 points, but when we overclocked the FX-70's the score was identical.
The AMD QuadFX FX-70 test system ran very well and overclocks better than expected for a system that is running two processors for a total of four cores. Sadly, even when overclocked the pair of FX-70 processors just didn't have enough to pass up the non-overclocked Intel Core 2 Quad QX6700 processor. It is now crystal clear that the dominance Opteron based platforms once had has been clearly uprooted by Intel's Core 2 microarchitecture and taken to the next level with their quad-core solution.
When it comes to building a QuadFX system there is only one motherboard available to use and that is the ASUS L1N64-SLI WS. The ASUS L1N64-SLI WS motherboard will set you back $419.99, which is pricey to say the least. AMD also recommends using the expensive PC Power & Cooling 1000W power supply (or equivalent) to ensure the system has enough power. With a pair of FX-70's and the ASUS L1N64-SLI WS motherboard one will find that it will set you back ~$1120 for both. Keep in mind there are two processors to cool and each processor has two banks of memory to fill. If one wants to improve cooling then two water blocks or heat sinks will be needed, which doubles the heating or cooling bill in many situations. In the same price range one could also pick up an Intel Core 2 Quad Q6600 processor at $879.99, which has numerous motherboards and chipsets that can be used from the $100 price range to the $400 price range. By overclocking the FX-70 processors we were able to improve the price versus performance 'value', which does help bridge the performance gap between the AMD QuadFX and Intel Quad-Core Kentsfield solutions.
The AMD QuadFX platform is a success in the sense that it works and has helped bridge the performance gap that was present before its launch. AMD has already annouced that the QuadFX platform will support their first quad-core processors that are code named K8L (greyhound)and planned for later in 2007. Greyhound is rumored to be the first quad-core AMD chip to use the HyperTransport 3 bus, and the memory controller will support both DDR2 and DDR3. It looks like K8L couldn't get here fast enough!
The Legit Bottom Line: On 32-bit Windows XP Pro the QuadFX platform hangs close, but just can't keep up with Intel's Quad-Core processors right now. Our $4,000+ QuadFX test system is a pricey system for something that offers dubious benefits.