I finally finished putting together a new bulldozer based system for my wife using: an AMD FX-8120 Zambezi processor, ASUS M5A97 MB, and 16 GB of Patriot DDR 3 1866 memory. I was wondering how it would hold up to the i5 2500k.
The Core i5 2500k based system I built for myself in March 2011 won't be exactly apples to apples. It is running only with 1333 memory, but I did install a huge air cooler (Scythe SCMG-2100 MUGEN-2 Rev.B) to replace the stock one; the AMD machine still only has the stock cooler. Since both are similarly price offerings from Intel and AMD targeting the same customers, I thought it would be interesting to put them through their passes.
Both have unlocked multipliers. I overclocked the i5 to a conservative 4.5 GHz, but it ran stabling overnight at 4.8 GHz as well. The FX-8120 only made it to 2.2 with the stock cooler before overheating became a problem; I am not even confident at 2.2 GHz it could run all night under a full load. I would suspect that I would get closer to the 5 GHz range with the FX-8120 with a high quality air cooler.
i5's 4 cores vs FX-8120's 8: First off AMD's marketing team calling it 8 cores isn't really true. A module of two "cores" share "early pipeline stages (eg. instruction fetch, decode), the FPUs, and the L2 cache with the sibling in the module." (wikipedia) So, the are not really full independent cores, but they are much more so than the virtual cores that are enabled through Intel's hyperthreading which give two threads of execution per core, but share a lot more of the hardware.
For an extra point of reference, I will also benchmark my work PC, a HP Z400 workstation with a Xeon W3520. The W3520 is basically the server version of the i7 920 with extra things like virtualization enabled. My workstation is about a year old, and is used daily for software development.
So we have three machines to look at: i5 2500k based one, AMD FX-8120, and an HP Z400 workstation. The two put-together machines are mainstream systems with unlocked processors that a budget minded enthusiast would build. They both hit the sweet spot of being the cheapest models that you can get that don't sacrifice cores or lock multipliers.
Machine Setups:
Custom Intel i5:
Processor: Intel Core i5 2500k Sandy Bridge @ 3.3 GHz overclocked @ 4.5 GHz
Motherboard: Asus P8P67 LE
Chipset: P67
Memory: 8 GB (4GB x 2) DDR 3 1600
OS: Windows 8 AMD64 (build 8174)
Custom AMD FX
Processor: AMD FX-8120 Zambezi @ 3.1 GHz overclocked @ 4.2 GHz
Motherboard: Asus M5A97
Chipset: AMD 970
Memory: 16 GB (4GB x 4) Patriot Viper Xtreme DDR 3 1866
OS: Windows 8 AMD64 (build 8174)
HP Z400
Processor: Xeon W3520 @ 2.66 GHz
Chipset: X58
Memory: 8 GB (2 GB x 4) ECC DDR 3 1333
OS: Windows 8 AMD64 (build 8174)
Price as tested: ~$1200
Benchmarks:
Cinebench 11.5 (AMD64):
FX - 6.71
i5 2500k - 7.10
Xeon W3520 - 3.99
Sandra 2012 (AMD64):
Arithmetic and floating point:
FX - 91.4 GOPS, 111 GIPS, 75.32 GFLOPS
i5 2500k - 86.3 GOPS, 113.43 GIPS, 65.67 GFLOPS
Xeon W3520 - 70 GOPS, 79.43 GIPS, 61.63 GFLOPS
Multicore efficiency:
FX - 11.83 GB/s
i5 - 8.63 GB/s
Xeon W3520 - 15.79 GB/s
Memory bandwidth:
FX - 19.3 GB/s - is this just because of the 1866 memory vs 1333, or is the memory controller that much better? I am not sure.
i5 - 10.73 GB/s
Xeon W3520 - 12.52 GB/s (You can see the Xeon's triple channel memory controller at work vs the double channel of the i5. It is roughly 50% faster as you would expect.)
From the reviews I have been reading I was expecting more of a massacre, but the FX held its own. With a better cooler than the free OEM one, I am sure I could have got another 10-20% out of it. My guess is that the newer faster memory I had in the FX machine had a lot to do with its surprisingly better numbers. If I have some time, I will up the speed of the memory in the i5 system and rerun the tests. Update: correctly setting the i5's memory speed only improved the memory bandwidth and multicore efficiency scores.
As we have seen in the past, some architectures are more efficient per clock. Ironically it was AMD that was the efficiency king back when Intel was pushing the clockspeed of their netburst Pentium 4s. Roles seem to have been reversed with AMD's so-called 8 core FX processor, and Intel's focus on efficiency with the core architectures. The FX is roughly keeping up with the i5, but it is 9 months newer. After 9 months, has AMD delivered? I don't think bulldozer was the leap forward that AMD had hoped, but it did better that I thought.
I can see a few reasons why someone would want to buy one of these AMD FX processors over the Intel sandy bridge k SKUs. I personally bought it because it has hardware virtualization built in. You have to go with the Xeon parts with Intel to get hardware virtualization, and they are much more expensive. At work, my workstation has a Xeon W3520 for running Hyper-V, but they retail at around $320. At almost $200, the AMD parts seem much more reasonable. I also like support AMD because they are the underdog, and I would like them to survive. Although the FX processor did better that I expected against the i5, it did fall below my expectations.
