The New AMD FX Lineup
The launch of “Vishera” includes four new CPU models, ranging from $130 up to $220 at launch.
FPU Modules/Integer Cores
Price Oct 23, 2012
|FX 8350||4/8||4.0 GHz||4.2 GHz||125 W||8 MB||8 MB||2200 MHz||$220|
|FX 8320||3.5 GHz||4.0 GHz||$180|
|FX 6300||3/6||3.5 GHz||4.1 GHz||95W||6 MB||2000 MHz||$140|
|FX 4300||2/4||3.8 GHz||4.2 GHz||4 MB||4 MB||$130|
The chart above includes the launch retail prices at Newegg. As you can see, the models’ prices are reflected mostly in clock speed and core changes, with very little else differentiating them. This means that even though you bought a $150 or $130 CPU, you will still benefit from features like AES and FMA instructions and a pretty substantial L3 cache. Furthermore, all models are completely unlocked. This is huge, as it allows one to spend no more than $130 or so on a CPU, with the potential of running it at much higher speeds than stock.
AMD have clearly targeted several specific Intel models with their pricing, so it will be interesting to see how they compare. We have included these direct competitors, to offer a more focused review. Here is how they compare on paper:
Intel Core i5 3570K
Intel Core i5 3450
Intel Core i3 3240
|Code Name||Vishera||Ivy Bridge||Vishera||Ivy Bridge||Vishera||Ivy Bridge||Zambezi|
(Oct 23, 2012)
|Base Clock||4.0 GHz||3.4 GHz||3.5 GHz||3.1 GHz||3.5 GHz||3.4 GHz||3.6 GHz|
|Turbo Clock||4.2 GHz||3.8 GHz||4.0 GHz||3.5 GHz||4.1 GHz||NA||4.2 GHz|
|L2 Cache||8 MB||1 MB||8 MB||1 MB||6 MB||NA||8 MB|
|L3 Cache||8 MB||6 MB||8 MB||6 MB||8 MB||3 MB||8 MB|
|New Instructions||AVX, AES, FMA3, F16C||AVX, AES, FMA3, F16C||AVX, AES, FMA3, F16C||AVX, AES, FMA3, F16C||AVX, AES, FMA3, F16C||AVX||AVX, AES|
|TDP||125 W||77 W||125 W||55 W||95 W||55 W||125 W|
|Process||32 nm||22 nm||32 nm||22 nm||32 nm||22 nm||32 nm|
These are all the models we’re including for the review. The Bulldozer based FX-8150 is included just to verify the performance improvements of Piledriver. It can probably be considered obsolete now (and I’ll refrain from making the obvious joke) and may go on a fire-sale soon. Note that because the top three models all have 8MB of L3 cache, it is easy to replicate all of them with the FX-8350, including the FX-6300 by simply disabling a pair of modules (this disables their low level cache as well). However, I could not replicate the FX-4300, so that will have to sit this one out. That’s unfortunate, because I think it has the potential to be a great product.
On the Intel side, the competing products are actually priced a bit higher than the new FX chips that targeted them. The 3570K costs $230 now, although I expect that price to come down in reaction to the FX-8350. Likewise, the Core i5 3450 is still $15 more than the FX-8320, and that model is completely locked down – no overclocking.
Intel’s Core i5 lineup doesn’t reach below $190, so at $150, the FX-6300 is competing with the top Ivy Bridge Core i3. It is instantly apparent just how much Intel strips out of their products for the sake of product segmentation. The Core i3 lacks turbo frequency, an L2 cache, a pair of physical cores, and even some instruction sets. All this stuff is right there in the chip – it is just disabled.
Core i5 3570K
Core i5 3450
Core i3 3240
AMD 990FX Chipset
Intel Z77 Chipset
|Memory||8GB (2x4GB) G.Skill RipjawsX 1833 MHz 9-10-9-28|
|Hard Drive||OCZ Vertex 4 256GB|
|Video Card||MSI Hawk|
ATI Radeon HD6870 1GB
(AMD HD6870 vs Nvidia 560 GTX)
|Motherboard Drivers||Catalyst 12.8 Chipset|
Catalyst 12.8 AHCI
|Video Drivers||Catalyst 12.8|
|Operating System||Windows 7 Professional x64 Edition SP1|
|FX-8320 replicated from an FX-8350 by adjusting multiplier to 17.5x/20x|
FX-6300 replicated from an FX-8350 by disabling 2 modules, adjusting multiplier to 17.5x/20.5x, lowering NB to 2000 MHz
Core i5 3450 replicated from a Core i5 3570K by adjusting multiplier to 31x/35x
Core i3 3240 replicated from a Core i3 3220 by adjusting BCLK to 103 MHz
Finally, we have paired up all the models in our charts by price – at the top of each chart, you have the FX-8350 and i5 3570K duking it out at around $210-220. Below that, you have the FX-8150 just to compare to the FX-8350. Below that you have the FX-8320 and i5 3450 at $180-195, and below that are the FX-6300 and i3 3240 at $150 a piece. That one should be interesting.
We’ll start things off with SiSoft Sandra, to confirm FPU and Integer performance in pure math tests as well as accelerated ‘multimedia’ tests. Starting with the classic Whetstone test. This is one of the first synthetic CPU benchmarks ever created (in 1972!). This is a floating point test, so the Piledriver cores will be less than optimized due to the shared FPU design of the modules:
Despite only running in ‘quad core’ mode with four FPU engines, the Piledrivers have no problem beating their Intel competition. Even Bulldozer did well in this test, so the bump from Zambezi to Vishera is 10% here.
Next up is the cleverly titled “Dhrystone” which is the Integer based benchmark made to compliment Whetstone.
This time it’s a dead heat, although the FX-6300 has no problem handling the i3 3240 with its two cores and small cache. The Zambezi > Vishera bump is 13% this time.
Those ‘pure math’ tests aren’t exactly optimized for today’s CPUs. SiSoft Sandra’s “Multimedia” tests are more like real-world based benchmarks, which make full use of multithreading and the latest instruction sets. Starting again with FPU performance, which should be AMD’s weak spot:
This time the tables have turned a bit, as Intel’s superior per-clock performance can overcome AMD’s shared FPU design in this test. The FX-6300 has an extra core to easily handle the i3 3240 despite this. The Piledriver speed increase is 10% once again.
The streaming Integer based test allows the FX architecture to really flex its muscles, and is a ‘best case scenario’ for the chip. Because of this, the AMD CPUs destroy Intel here, especially at the $150 price range. 11% bump this time.
As we saw in our Trinity review, the AMD chips can outperform competing Intel chips, or perform slightly slower, based on computation type. However even in the ‘worst case scenario’ the performance is not too far off, which bodes well for real-world performance.
Let’s kick off our real-world performance testing with some games: