Review The Athlon 64 FX, the pinnacle of AMD's consumer processor offerings, now has two cores, to more effectively run a modern OS. There's no surprise - the mystery of whether or not the next FX would be single- or dual-core was dashed a while ago, and really it was obvious if you think about it. Once you understood that it was dual-core, then going on to figure out the target clock frequency was a piece of cake, given Athlon 64 X2 4800+.
With that information, extrapolating performance from the 4800+ is a fairly simple task, one which would at least get you in the ballpark. However, there are still a few important questions to answer...
Firstly, is it better than the single-core Athlon 64 FX-57? Notice the better adjective. Faster is the wrong way to go about the analysis, since single- versus dual-core with non-matching clock frequencies will always throw up different answers, depending on what code you run.
Secondly, does it beat Intel's Pentium Extreme Edition 955 processor?
The FX-60 shares the same transistor count, caches and cache organisation, processor technology and core revision as the Athlon 64 X2 4800+, confirming it as having the same 'Toledo' core. What marks it apart is the 2.6GHz clock speed, versus the 2.4GHz of the 4800+, and the slightly lower operating voltage range of 1.30-1.35V, down from 1.35V-1.40V. It goes faster on less voltage, essentially.
The processor has the same thermal envelope (110W), meaning the same coolers can be used. Under Cool'n'Quiet, AMD's desktop-oriented power-saving technology, the processor drops to 1.2GHz per core with multiplier adjustment, running from just 1.1V.
Being a current E6 revision of the X2 core, badged as FX, it supports all three SSE instructions sets, x86 and AMD64. The processor fits into the same 939-pin socket as all high-end Athlon 64 processors have since the switch from Socket 754. The extra pins facilitate a dual-channel DDR memory controller, which means support for four sticks of memory at the maximum 400MHz speed, the best use of available memory space (introduced first with the 'Venice' revE single core) and support for DIMMs of different sizes per memory channel.
The astute among you will have noticed that the 2.6GHz and 1MB of L2 cache per core make the FX-60 effectively a pair of FX-55s on the same CPU, in the same socket, sharing a memory controller and system link to the outside world. That sums it up nicely. While sharing a memory controller and system link will reduce absolute performance compared to two separate 2.6GHz, 1MB chips, we all know how well AMD dual-core processors can run in that situation.
We mentioned earlier that, given the 110W thermal spec, coolers that effectively keep the X2 4800+ happy will be fine with the FX-60. That also extends to coolers designed for FX-55 and FX-57. The 104W and 110W TDPs of those single-core SKUs mean that their own validated thermal solutions will extend to the FX-60. It's nice to see AMD keeping power and heat in check over the last 14 months or so, despite the significant increases in performance.
As to why the FX-60 isn't a 3GHz single-core product, we can but speculate. It's obvious that AMD's 90nm process technology could probably see it debut such a chip at sane voltage and power, likely with the same coolers we've just discussed, using the best core it has available. However, the 3GHz figure raises expectation with the consumer that the company will follow through with the rest of the 3GHz range, on towards 4GHz.
Secondly, the company is moving down the road of hardware multiprocessor and thread-level parallelism in its CPU products, server, workstation and mainstream consumer alike. Multi-core, including AMD's current dual-core processors, are where it's going to be at for the forseeable near to mid-term future.
So it's 200MHz slower than the 2.8GHz single-core Athlon 64 FX-57 in external CPU frequency, but there are two cores. Remember the, 'better'? Exactamundo...
The chips' OPN code, ADAFX60DAA6CD, tells you that it's an Athlon 64 (ADA), FX-60 (FX60), Socket 939 (D), variable-voltage core (A), variable temperature (A), has 2MB total L2 (6), is an E6 revision 'JH' dual-core processor (D). Turning it over to see the 939 pins would show you nothing you haven't seen before.
A new cooler? Not quite, but close. When FX-57 was released it was shipped solely with a PIB cooler developed by AVC. Since launch, AMD has sought out another model to ship along with FX-57 and its own ~100W CPUs. That other model is what'll ship with FX-60. It's manufactured by Coolermaster and called the CMHK8-8I22A-A2.
Compared to the first FX-57 PIB cooler, the Coolermaster has all four heatpipes aligned on one side of the cooler, rather than two per side. The cooler, including heatpipes, is also around 5mm shorter in major width, although it retains the same 62mm height. The fan is an 80 x 20mm Delta, up from 70 x 15mm on the first FX-57 PIB example, allowing the fan to move a larger air volume at the same fan speed, or the same volume at reduced speed.
Presumably the different cooler design and fan are primarily to achieve the latter, making this cooler even quieter under the same conditions as the AVC. It appears, although we can't confirm it, that the cooler also comes supplied with a Dow Corning thermal material, whereas the AVC cooler is equipped with a Shin-Etsu thermal pad. The heatpipes are outwardly identical with a 5mm width and equivalent length, and we're told they each have a 30W capacity, making the CMHK8-8I22A-A2 a 120W capacity device.
The differences add up to something that's smaller overall, but has the ability to cool better at the same fan speed and equivalent levels of noise. We'd hazard a guess that this PIB cooler revision will also see service with the next speed grade of Athlon 64, Athlon 64 FX and Athlon 64 X2, if and when they appear.
The processor driver allows the OS to adjust the performance level of the CPU cores via multiplier and voltage (P-states), based on CPU demand. The driver is therefore a recommended installation, giving rise to power savings and less heat output when the CPU is idle.
However, Windows XP Service Pack 2 doesn't correctly identify CPU load in all cases, which can lead to it setting a lesser P-state than would normally be required, reducing absolute performance with the side-effect of better environmental performance. A fix, not generally available for download from microsoft, changes that to affect proper adjust of the CPU P-states for the right performance levels, trading off the environmental sides previously mentioned.
On top of that is Cool 'n' Quiet, which, at least initially, may not even be available for FX-60 until BIOS engineers add in the correct IDs to the CnQ tables to let that feature work correctly. So, depending on BIOS revision, CPU driver installation and presence of the essentially unavailable hotfix, differing performance and environmental levels will be experienced. Something to keep an eye on.
With the same E6 revision Toledo core as the Athlon 64 X2 4800+, you'd expect the FX-60 to have largely similar memory sub-system performance, given the memory controller on the processor. ScienceMark 2.0 largely tells us if that supposition is true or not, helping us to measure memory bandwidth and memory access latency. Bandwidth first.
The dual-core AMD processors, including FX-60, don't have quite the bandwidth of the single-core CPUs, despite sharing the same basic configuration on the memory controller. Shared access to the controller by the cores sees to that. The FX-60 performs just like the 4800+ in this test, as expected.
Still, 85 per cent utilisation is much better than the Intel systems can manage. The 955 and 3.73GHz XEs need a 266MHz CPU-to-system bus to grab comparative bandwidth.
Likewise, the dual-core AMD CPUs are slower at going out to main memory than their single-core counterparts, by around ten per cent. Their access latency is still years ahead of current Intel systems with off-chip MCHs.
Our CPU test suite is nicely split into single- and multi-threaded benchmarks and applications. Single-threaded ones first.
You can see the AMD processors scale almost linearly in their 200MHz steps, from X2 4800+ and 4000+ (both 2.4GHz, 1MB L2) to FX-60 (2.6GHz, 1MB), and finally FX-57 (2.8GHz, 1MB). Cache-agnostic on the Netburst core, Realstorm also scales with CPU frequency with the Intel processors.
The performance difference between AMD and Intel here is a measure of the performance of the FPU for the most part, AMD processors having the stronger one when doing lots of 3D-oriented calculations as in Realstorm.
The result graph follows the pattern of the Realstorm results, for the same reasons. The AMD processors generally have the stronger main integer pipe and FPU, depending on the calculation being performed of course.
Our multi-threaded CPU-bound tests split into two groups: media processing, and 3D and image processing. Media processors first. We use a version of LAME - 3.97 - that's compiled with Intel's multi-threading C compiler.
The FX-60 ties with 955XE in this test, 840XE further tying with X2 4800+. Intel processors with HyperThreading active see their performance scale with CPU frequency after it scales with the number of possible threads. This test shows that the compiler and software application itself has a strong bearing on overall performance from a processor.
DivX 6.1 Pro lets us encode a large DV file into a 1.7Mbps DivX video stream, using the encoder's Insane profile. It supports AMD and Intel multi-core, Intel HyperThreading, and supports more than two execution threads.
Our video encoding test shows the same pattern. 955XE is competitive with FX-60 and 840XE holds off the X2 4800+, while the single core CPUs bring up the rear. It's clear that a hardware multi-core CPU is a benefit to media processing tasks.
It's generally the done and easy thing to rubbish Intel processors these days, and indeed I'm happy to put the boot in when it's deserved, but its clear that they're well suited to these two tasks when the encoder is favourable.
Software 3D renderers and their atypical code make-up can be found in digital content creation apps, offline renderers, 3D simulators and the like. FP-heavy, more so than any other type of common code, these tests give the multiple FPUs of modern processors a real hammering. These particular tests are also multi-threaded.
The Cinema 4D engine, although in our tests its from the R8 release in 2003, is used in more places than you'd think, from film rendering to game development toolchains.
Dual-core is a big win, nearly doubling the score on AMD at the same CPU frequency. 666 is the fastest score from a single CPU we've yet seen here, Opteron and Xeon included. Much the same is seen with Kribibench. Dual-core is a large win, offering almost 2x the speed of a single-core chip at the same frequency, and the 955XE is faster than FX-60 in this particular test.
picCOLOR is our image analysis benchmark and uses a range of integer and FP calculations, often using the SIMD logic, to process image data for a wide range of applications.
6.15 is the fastest single CPU score we've seen with this build of picCOLOR, FX-60 besting X2 4800+ by the expected margin.
While the DFI is likely the reason why FX-60 comes out on top here, it shows that the 200MHz deficit can matter little when the graphics hardware is in on the act. Let's confirm that with some slightly CPU-bound real games, at 1024x768.
The FX-60 is less than one percentage point slower than the FX-57, despite a 200MHz deficit, and with the same-speed pair of X2 4800+ and A64 4000+, the dual-core processor is actually the slight victor.
There's a couple of percent between 57 and 60. It's all down to whether that couple of percentage points matter to the owner of the processor. If it were me...
AMD Athlon 64 FX-60 is the best consumer processor AMD has ever produced. With effectively a pair of FX-55s sat in the same socket, sharing an efficient memory controller, it's close enough to FX-57 in single-threaded apps that the multi-threaded advantage makes that slender gap moot. Targetted at the well-heeled enthusiast, the new dual-core processor should be a shoo-in for those with FX-57s already, and those with the required readies to drop on the latest and greatest.
Gaming wise, the FX-60 is a minor step down from an FX-57, both at their default clocks. But the step is a small one, and the benefits of the extra core in system that's being used by other applications while you play are what the FX-60, and all performance dual-core processors, are about. The FX-60 is a gamer's chip, make no mistake.
The FX-60 is simply the most desirable processor to slide into a 939-pin processor socket to date. Dual-core is that good, even though our test suite could use a little work to really show you how a multi-processor system can chew through modern workloads.
However, for the first time in a long time, we're left pondering whether Intel have a worthy contender to a new Athlon 64 FX processor. The 955 'Presler' processor has a fine turn of speed when it's worked in multi-threaded applications, and it's no gaming slouch either. That Intel insiders say its a 4.26GHz processor in disguise is telling, too.
At similar prices - $1000 or so in volume - there's nothing much in it when it comes to price, especially if FX-60 turns out to be close to £800 today, rather than the conversion at current exchange rates. What edges it for the FX-60 is the excellence of the supporting platform and the environmental performance, on top of the slender overall performance win. It runs cooler than FX-57, consumes the same under our load conditions, and bests 955XE in both those metrics by quite some margin.
It won't be cheap, but then for most FX consumers that matters little. They want the best CPU for their systems and under default conditions, the FX-60 is that microprocessor. The particular review sample used for this article does a nice and easy 2.8GHz at stock voltage, and just shy of 3GHz when voltage is upped a little, all on the default PIB cooler. Unlocked multipliers means adjustment of the requisites is a piece of cake.