While not as powerful as Intel's per core performance, they are no where remotely close to being similar to hyperthreading. Not even in the same realm of design. Modules in Bulldozer share many things, but they still have two physical processing cores.
They're not fake, just not optimal.
i7-4770k - GTX 780 Ti - 16GB DDR3 Ripjaws - (2) HyperX 120s / Vertex 3 120
ASRock Extreme3 - Sennheiser Momentums - Xonar DG - EVGA Supernova 650G - Corsair H80i
build pics
My original statement to which Romandix responded to was that if the 9000 series had been 12 or 16 core rather than the current 8 core, the price would have been more justified because then it would have been an more of an "enterprise" CPU and more deserving of the $600-900 price tag. In essence, a 12 core 9000 series would be a lot like the i7 extreme edition CPUs, halfway between mainstream FX and enterprise Opterons.
Last edited by yurano; 2013-06-18 at 08:56 AM.
the problem is that you can't maintian a high core count and a high clock speed, intel's 8 cores cap out at 3.2Ghz or somthing, and their 10 cores cap out at 2.8ish(and thats a 5 grand cpu), there is no way that AMD could make a 12-16 core CPU at 5Ghz and make it anywhere remotely affordable, you be talking 5 digit price tag for a CPU
They're fake.
Disregarding caches more than half of the surface area of single core is taken by the floating point unit. One Bulldozer module has two cores but those share only one FPU between them. That is why interger performance is great on Bulldozers and FP performance sucks ass. On very basic level HT duplicates about 5% of full-featured core, while Bulldozer modules duplicates 40% of full-featured core. Both are fuckin' far from ideal, and both are far from two physical real cores.
There's a very good reason why Intel originally started shipping math co-processors (8086 + 8087), becauses the FPU is too important to miss and speeds up too many things.
Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
Trolling should be.
Consider what AMD's doing right now. They're asking a grand for a 5Ghz 8-core CPU.
I'm not saying they should put out 5Ghz 12 core CPUs. For $1000, a 12 core CPU at 4 Ghz makes more sense than 5 Ghz on an 8 Core CPU.
Depends on the task. For Linpack, the performance advantage of HT is 0%. For the average program, I estimate the performance advantage to be about 30%.
Correct me if I'm wrong, but won't the vast majority of 8350s overclock to 5.0 GHz anyway?
You are correct. I would say they could get away with charging 20-30% more for the FX9000 but they are not worth any more than that to gamers. Like someone already mentioned the marketing will still sell these CPU's no matter the price which is a shame.
No, they're not. It's four bulldozer modules that contain the L2 cache, FPUs, fetch, and decode units. With 8 fully pipelined integer cores split between them (IE, 2 cores per module).
No matter how bad of a design or how ideal it is, it doesn't make them fake. The integer cores are physically on the die set. Hyperthreads, however, are not. And in that case, they are 'fake'.
I'm more worried about power consumption and heat. Whilst it's safe to assume that buyers will have more than adequate cooling to keep the temps at bay for such an OC'd chip, I'm sure power consumption will be pretty nasty.
No FPU = not real core. Basically everything uses FPU today (as you can see from Bulldozer benchmarks) so it's effectively four cores for anything that matters.
Also HT is very much on the physical chip. What HT does is duplicate only registers and program stack and can be used for extremely fast task switching.
90% of the clock cycles modern CPU is waiting for memory fetch operations that can take few hundred clocks from DDR3 or IO from ethernet/USB/HDD that can take millions of clock cycles. During that time HT can switch to another task in few clocks (less than 20) and start running that while the first task waits for the data. Effectively HT cuts down the idle time of processors considerably and it's the amount of idle which determines how big benefit you get from HT. For operations that do very large memory fetches or wait for IO HT can show up as nearly 100% gain, while programs that run completely in L1/L2 cache like highly optimized math (Linpack) will never need to switch task while waiting for RAM fetches.
Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
Trolling should be.
Not fully gimped no, but Bulldozer is only four core processor when it comes to FPU performance, not eight core. That was the point all along. Meaning it's eight half-cores while HT is four cores + four core fragments. Both pretend to the OS to be eight cores but both are severely gimped in their own way.
Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
Trolling should be.
They want $1000 for these CPUs? Are they nuts? Yeah enthusiasts like high clock speeds, but they only need that for like, 4 cores tops. The number of people that are both enthusiast enough to want 5ghz AND need to utilize 8 cores for multithreaded performance are very small. And even then, they would probably balk at 220W TDP. A 3570K or 4770K on watercooling would blow the pants off these still at single-threaded, and not be completely far off on multithreaded either.
And if you're willing to spend $1000 on this ... why not just buy one of AMD's 12 or 16 core server CPUs?
When you look at any performance benchmarks between FX-8350/Phenom x6/i5-3570 it's pretty clear in properly multithreaded tests Bulldozer's half-cores almostI makes up for the low IPC, but never really performs up to the levels of eight cores on anything besides integers. Common sense says from reading those numbers that the processor isn't compatible with the way programs are done today, and that's something I've been commenting for years regarding AMD processors ever since x6's launched. They're betting everything on 100% of x86 code running multithreaded which simply isn't happening as soon as they hoped while Intel is focusing on pushing the cuirrent technology to it's limits before adding more cores which is something they obviously can, looking at 8+HT Xeons. AMD came out with first home PC dual core processor ten years ago, and still most AAA games released today don't use even two cores to their fullest potential.
Fastest of those runs at 2.8GHz. More cores = lower speed. FX-9000 line has more cores and crapton more heat and power requirement as previously discussed on this thread, which is the price of that speed and totally unacceptable on server processors.
Last edited by vesseblah; 2013-06-19 at 08:12 AM.
Never going to log into this garbage forum again as long as calling obvious troll obvious troll is the easiest way to get banned.
Trolling should be.
It's not a secret that AMD is shifting resources from the high end PC market to the mobile device market. Intel has no response to Bobcat and Jaguar and their integrated graphics are slow, which makes AMD APUs a vastly better solution for both Xbox One and PS4, as well as pretty much any mobile gaming device.
Last edited by haxartus; 2013-06-19 at 04:28 PM.