Summary: After months of waiting, AMD's quad-processing solution previously codenamed 4x4 is finally here! 4x4 involves much more than just a new CPU however, as the new Quad FX Platform also encompasses a new chipset and motherboard. See how the new processors perform with NVIDIA's GeForce 8800 GTX GPU in today's article!
Over the following months, AMD revealed more details on their 4x4 platform, including the fact that 4x4 CPUs would be sold under the high-end FX brand, and that 4x4 CPUs would be sold in pairs, with kits starting “well under $1,000”. AMD also committed to supporting cheaper unbuffered memory and tweakable motherboards that offered a range of HyperTransport speeds and other BIOS options for CPU and memory overclocking. AMD also concocted a new nickname for their 4x4 technology: the quadfather.
Today marks the official introduction of AMD’s quad-processing technology and it is indeed quite a performer under the right situations. Before we get into that though, let’s first discuss why AMD felt now was the time for four processing cores…
Why four cores?
With so few games taking advantage of dual-core processsors, much less four processors, many of you have questioned why AMD’s been targeting 4x4 at hardware enthusiasts and the hardcore gaming crowd.
Besides gaming, another usage scenario for 4-cores is what AMD describes as “megatasking”. Megatasking takes multitasking to the next level, as it involves running multiple CPU-intensive tasks at once. An example would be encoding an HD video (or two) while also watching an HD video, or MP3 encoding while also touching up a batch of photos in Photoshop. For those of you who are into MMOs, you could load two instances of the game at once and trade items you’ve collected back and forth between characters, or have one character fighting while the other is healing him.
This is where having four processing cores really shines.
Debuting alongside the new 4x4 CPUs are a new chipset (NVIDIA’s nForce 680a) and a new motherboard based on that chipset (the ASUS L1N64-SLI WS), AMD dubs all this the Quad FX dual socket direct connect platform. Let’s see what’s new with Quad FX…
In many ways AMD’s new 4x4 CPUs (the Athlon 64 FX-70, FX-72, and FX-74), share more in common with AMD’s server-oriented Opteron line of CPUs than the Athlon 64 X2. Considering the origins of the FX line (and the Athlon 64 in general for that matter) date back to the Opteron, you could say that the new FX-7x CPUs are going back to their roots, but more on that later. Let’s take a look at the new CPU specs:
Effective Data Bandwidth: 33.6GB/sec (8GB/sec HyperTransport link + 25.6GB/sec Memory)
FX-70: $599 per pair
FX-72: $799 per pair
FX-74: $999 per pair
Like the Athlon 64 X2 CPUs, the Athlon 64 FX-70, FX-72, and FX-74 support conventional unbuffered non-ECC memory. This means you can use the same off-the-shelf DDR2-800 or DDR2-667 memory used previously on your AM2 setup on the new 4x4 CPUs. Everywhere else though the new FX CPUs more closely resemble AMD’s Opteron CPUs.
Keep in mind though that in order to take advantage of this, you’ll need to populate all four DIMM slots with memory. Also like Opteron, 4x4 systems also follow a non-uniform memory access model (NUMA), so you’ll need a NUMA-aware OS such as newer Linux builds, Windows Server 2003 64-bit Edition, or Windows Vista.
Because we’re testing with the GeForce 8800 GTX in this article and don’t have Vista drivers for this GPU, we conducted our tests under Windows XP Pro. We’ve been told by AMD that even though this isn’t a NUMA-aware OS, it shouldn’t affect our benchmark results. Only memory-sensitive applications such as SiSoft Sandra’s memory tests and other memory benchmarks are affected by this.
The other key difference between the new FX CPUs and their predecessors is AMD’s new 1207-pin socket, frequently referred to in the past as Socket F. This is the same socket AMD uses on newer Opteron processors, and as you can see in the photos, the pins are located on the motherboard itself, rather than the underside of the processor.
With 1207 pins on the new socket, frankly we’re glad to see that AMD has moved the pins from the CPU to the motherboard – could you imagine trying to fix a bent pin on your $600 or $1,000 CPU?
The chipset supports NVIDIA technologies such as MediaShield, with support for RAID 0, 1, 0+1, 5, and JBOD (just a bunch of disks) using Serial ATA drives, in fact, one feature that’s been touted is the ability to run up to two RAID 5 arrays. As we mentioned previously, the nForce 680a chipset also supports up to four DDR2 memory channels (2 per CPU).
On the networking side, NVIDIA’s FirstPacket technology is included, providing prioritized networking for apps like gaming and BitTorrent. With quad Gigabit Ethernet support, we were a bit surprised however to hear that the chipset doesn’t support NVIDIA’s teaming technology, which allows the Ethernet ports to be combined as one for increased throughput.
For graphics duties, the nForce 680a SLI chipset supports up to four PCI Express graphics slots. Only two of the slots run at full x16 speeds, the other two slots are x8. With four PCI Express graphics slots onboard, the nForce 680a SLI chipset is ready to support not only SLI, but also NVIDIA’s GPU-based physics solution, or you can use the four graphics slots to connect up to 8 displays.
As you can see in the above diagram, the nForce 680a SLI MCPs are linked to one of the FX processors via an x16 HyperTransport link, while the two CPUs are linked by a x16 coherent HyperTransport link. This configuration allows the system to function when one CPU is installed. Each MCP supports six SATA ports, 10 USB ports, and dual GigE, as well as the PCIe slots. Among the two MCPs, the chipset supports up to 56 PCI Express lanes total.
The beauty of this platform is not only its sheer specs, it’s also forward-looking. AMD boasts that today’s Quad FX platform will be upgradeable to eight cores (two quad-core CPUs) when AMD’s quad-core FX CPUs are released in the second half of 2007.
Like all of ASUS’ high-end motherboards, the L1N64-SLI WS is entirely passively cooled. You’ll find no active fans on the board’s nForce 680a MCPs or VRM circuitry. Instead ASUS relies on copper-based heat pipes: sitting directly atop the nForce 680a SLI MCPs is a massive copper plate with twin heat pipes connected to it. A copper heatsink with rolled fins helps keep the heat pipes cool.
For powering both CPUs, ASUS has integrated an 8-phase power solution for each FX CPU. Resting above the power circuitry for both CPUs are large copper heatsinks.
For additional cooling, you can mount auxiliary cooling fans to each of the heatsink assemblies. ASUS provides three fans with the motherboard, so one fan for each area.
In between both CPU sockets you’ll find four DIMM slots – two per processor. Again, as we mentioned previously, for optimal performance you’ll want to popular all four DIMMs, or else the chipset will fall back to single-channel mode.
Keeping to the specs of the chipset, ASUS provides four PCI Express graphics slots on the L1N64-SLI WS. The blue slots are the x16 graphics slots, while the black slots are the secondary x8 graphics slots. ASUS provides plenty of room for dual-slot graphics cards like NVIDIA’s GeForce 7900 GTX and 8800 GTS/GTX, in fact ASUS leaves enough room between all the PEG slots for dual slot cards except the bottom x8 slot.
ASUS also equips the L1N64-SLI WS with 12 SATA ports.
The only area where ASUS falls a little short is in the USB and networking department. Out-of-the-box the board only supports up to 10 USBs: four via backplate and six via external USB headers, while the board only ships with two GigE Ethernet ports. You’ll also notice that legacy ports are still provided. We honestly would gladly swap out the PS/2 and parallel ports for a few more USBs on the board’s back plate. For digital audio duties, ASUS does provide both optical and coaxial S/PDIF outputs.
With so many features onboard, the L1N64-SLI WS is a little larger than most ATX motherboards and thus uses the extended ATX form factor. Because of this, the motherboard may not fit in some cases. You’ll also notice the area around the CPU socket is rather cramped – we just barely had enough room to get Zalman’s CNPS9500 LED cooler to fit on the motherboard (the heatsink’s fins barely missed brushing up against the memory modules), you’d be hard-pressed to get a larger cooler to fit.
For overclocking, the L1N64-SLI WS offers HyperTransport speeds up to 400MHz in 1MHz increments, and memory speeds of 400, 533, 667, and 800MHz. CPU voltages up to 1.45V are also provided in BIOS and with memory voltages topping out at 2.5V.
The L1N64-SLI WS officially carries an MSRP from ASUS of $349.99. That places it a little higher than today’s high-end Core 2 motherboards, but not by much considering the extra features this board carries. With 4x4 kits starting at $599, this means that you should be able to get AMD’s baseline 4x4 platform for a little less than the price of one quad-core Core 2 Extreme QX6700, which carries a list price of $999.
LAME MT MP3 Encoding (MS Compiler)
We really, really wanted to provide FX-70, FX-72, and FX-74 numbers, but for whatever reason whenever we downclocked our FX-74 CPU to FX-72 levels we encountered a number of BSODs and stability issues in Windows. Strangely enough, downclocking the FX-74 to FX-70 speeds worked okay, but the system just wouldn’t work very well at the 2.8GHz required for the FX-72. Of course, keep in mind that we’re testing with a preproduction board on a beta BIOS, hopefully ASUS and AMD will have the kinks worked out on final retail boards. We spent so much time troubleshooting the FX-72 problems however that we didn’t have time to test with other AMD CPUs such as the FX-62 and 5000+.
Microsoft Windows Media Encoder 9
LAME MT MP3 Encoding
AMD’s Athlon 64 FX-74 falls just three seconds behind the Core 2 QX6700 in our video encoding tests with Windows Media Encoder 9, while the X6800 and FX-70 finish in a dead heat. We use the Microsoft compiler for our LAME MT MP3 encoding tests, here Intel’s Core 2 CPUs excelled. We noted previously that LAME MT doesn’t take advantage of quad-core processing, which helps explain the FX CPUs poor showing in this particular benchmark.
FutureMark’s 3DMark 06 is another synthetic benchmark that takes advantage of multi-core. FutureMark’s CPU test gives the edge to the Core 2 QX6700, with AMD’s FX-74 finishing 9% behind the Intel CPU. The FX-70 falls to third, trailing the FX-74 by 14%.
Valve is one of many game developers that plans to incorporate support for multi-core in their latest games. In fact, multi-core optimizations will be incorporated into the next Source engine update just ahead of Half-Life 2 Episode 2.
For our gaming tests, we’re running a mixture of high and low-res testing. We realize that most of you don’t game at 800x600, which is the ideal res to test CPUs, so we also include results at 1600x1200 with 4xAA/8xAF, which are graphics settings more typical of someone with a GeForce 8800 GTX card.
Quake 4 is one of the few games on the market that’s been programmed to take advantage of dual threads, but unfortunately not multi-threading. As a result, the Core 2 Extreme X6800 delivers the fastest performance thanks to its high clock speed, with the QX6700 falling behind by 4%. Like our performance results with F.E.A.R., the FX CPUs continue to fall behind the Intel CPUs, even at 1600x1200 the GeForce 8800 GTX graphics card is so powerful that the GPU isn’t bottlenecked and therefore the Core 2 CPUs are comfortably ahead of the FX processors.
Company of Heroes
For whatever reason the Core 2 CPUs put up a particularly strong showing in Company of Heroes at 800x600, with the Core 2 Extreme X6800 delivering up to 1.5 times the performance of the Athlon 64 FX-74 at that resolution. Once you crank up the screen res and turn on the AA/AF however, the GPU is the limiting factor and thus performance between all the platforms we tested was the same.
Under our multi-tasked gaming scenarios we run our MP3 and WME9 encoding tests in the background while also gaming. This puts an enormous burden on the CPU, particularly at lower resolutions, and thus performance drops significantly for the dual-core CPUs and the AMD Quad FX processors finally begin to shine. Intel’s Core 2 Extreme QX6700 still put up a stronger showing in F.E.A.R. however, outperforming the FX-74 by 15% at 800x600 although the margin of victory declines once more of the burden is placed on the graphics card at 1600x1200 with 4xAA/8xAF.
Under Quake 4 the Intel quad-core processor still enjoys a nice lead over the quad-core AMD platform, in fact the Core 2 Extreme begins to encroach on FX-74 territory at 800x600.
Officially, AMD’s new Quad FX platform isn’t AMD’s response to Intel’s quad-core Kentsfield CPU – as we outlined in the intro, Quad FX has been in the works for quite some time – but it’s hard not to make comparisons between the two.
Obviously there are differences in the implementation: AMD’s quad-processing solution is a two-socket, two-processor setup while Intel’s Kentsfield CPU consumes only a single CPU socket, with four processing cores integrated into the CPU’s packaging. Intel’s solution isn’t a native design, but as we’ve seen in today’s benchmarks as well as our Core 2 QX6700 Performance Preview earlier this month, this doesn’t seem to negatively impact performance in real-world apps that much.
The real beauty of Kentsfield is that it’s compatible with a wide range of existing motherboards; in comparison AMD’s Quad FX platform requires a new motherboard and chipset, ASUS’ L1N64-SLI WS with the nForce 680a chipset. The new NVIDIA chipset is certainly powerful, supporting features previously unheard of such as 12 SATA ports and four PCI Express graphics slots, but with ASUS being the only motherboard provider at the moment, that could keep street prices on L1N64-SLI WS motherboards artificially high.
Adding to the negatives is the enormous amount of heat two FX-74 processors put out: at idle we were seeing temps in the 45 degree Celsius range!
In terms of performance, the Quad FX platform is certainly no slouch, but it’s not capable of outperforming Intel’s Kentsfield CPU. With the exception of Cinebench 9.5, the FX-74 trailed the Core 2 Extreme QX6700 in all of our testing, in some cases pretty severely. In our high-resolution megatasking scenarios the graphics card becomes more of a bottleneck and as a result, the two platforms perform more closely to each other, but Intel’s quad-core CPU still held a decisive advantage.
There is a silver lining for Quad FX however. AMD’s priced the FX-70 and FX-72 CPUs so aggressively that if you go by official list prices, you can get the FX-72 CPU in particular and ASUS’ L1N64-SLI WS for less than the price of one Core 2 Extreme QX6700 CPU: technically this combo is the cheapest way to get to four processing cores at the moment.
When you consider that like all FX CPUs, the FX-70 has an unlocked multiplier, the package becomes even more enticing – with a few keystrokes in the L1N64-SLI WS BIOS you could overclock your FX-70 CPU to FX-74 (or greater) levels.
Unfortunately, we don’t have an FX-70 CPU to test this theory, but we have no doubt that enterprising AMD enthusiasts will be trying this out shortly. When you factor in that today’s Quad FX platform will eventually scale to eight cores in 2007, this potential upgrade route becomes even more attractive.
AMD’s Quad FX platform doesn’t roll over the competition like AMD had hoped, mainly because Intel accelerated the launch of Kentsfield from Q1’07 to Q4’06, but it does lay the groundwork for AMD’s next-generation K8 processor, codenamed Barcelona. For the hardcore crowd that could really use the extra SATA ports or would like to setup a mega 8-display system, AMD’s Quad FX platform may be tempting, but for the rest of you, we can’t help but feel that all this may be a little too much until more apps are available that take advantage of multi-core.
That day is coming soon though, and when it arrives Quad FX will be ready.
|© Copyright 2003 FS Media, Inc.|