1) They're still on 32nm which means worse efficiency.
2) They have far fewer transistors than Intel chips. 1.2B for "8 cores". Even a Sandy Bridge E quad has more with 1.27B. A true six core Sandy Bridge E chip has 2.27B. AMD's "8 core" chips sound massive, but they don't have very many transistors. 8 core Haswell E has 2.6B transistors — twice as many as AMD's "8 core" Vishera.
There is only so much you can do with clockspeed when you're dealing with a chip that has so few transistors. However, Lynnfield quads have just 774M transistors and they're faster per core than Vishera even at a lower clockspeed! A 4 GHz Lynnfield quad will beat a 4 GHz Vishera in gaming benchmarks, and that's a 45nm chip with so few transistors:
FX 8350, 4 GHz
i5 760 quad (Lynnfield), 2.8 GHz
Cinebench R10 single thread benchmark:
FX: 4338
Intel: 4512
Dragon Age Origins 1680 x 1050, Max, no AA or Vsync
FX: 139.2
Intel: 142
Dawn of War II 1680 x 1050, Ultra
FX: 70.5
Intel: 70.8
WoW
FX: 91.5
Intel: 89
Starcraft 2
FX: 47.9
Intel: 44.9
Now, keep in mind that the Lynnfield chip has no hyperthreading, is from 2009, and is running at a 1.2 GHz slower rate. FX is competitive with a gimped quad from 2009. That says a lot.
I realize these are older games that are not heavily threaded like newer ones, but I don't have better benchmarks in front of me (using Anandtech's CPU bench). It still says a lot that single thread performance of a tiny old Lynnfield chip beats FX clocked 1.2 GHz higher, or is around the same performance. Lynnfield also has 7 MB less L2 cache.
The one thing in Anandtech's CPU bench that FX chips seem to really excel at is 7-zip. I assume this is because each of the four modules has two integer threads and one fpu.
FX: 23223
Lynnfield: 11641
Lynnfield also falls well behind when an app can take full advantage of the eight FX threads:
Cinebench R10 multithreaded
FX: 22674
Lynnfield: 15060
But, let's look at 4 core Haswell at 4 GHz... (4790K, 88W)
If you have a water loop and can push enough voltage into an FX to get near 5 GHz then you can game at 4K relatively well because of GPU-bound scenarios. It won't be as fast as Haswell, but it's reasonably competitive with Ivy Bridge E (4930K). But, you're going to pay for the power bill.
I am interested is trying a dual radiator setup similar to this one but with 180mm fans instead. Do you know if anyone has tried that? Eight fans seem as if they would be not only less expensive but also less prone to clicking. However, I'm not sure if the efficiency will be good enough to keep them at low enough speeds to avoid significant noise.
Also, could you use a decibel meter to quiet the skeptics who say there is no way your system is going to be quiet enough?
One other question... Do you think the flow rate from using two 60mm radiators rather than just one would enable a pump to run at a lower rate? Skeptics have said there is no way the pump noise can be low enough to manage to pump through two radiators at a distance like that, especially if only one of them is 60mm.
RichardF got a reaction from Fate in Gaming CPU build $1000
RichardF got a reaction from enthalpy in 23 Noctua fans? Well... yeaa! [Upgraded to 18 Cores]
