Hardware AMD FX-8120 review

Fair enough. Although the trace cache and Hyperthreading were things introduced by Pentium 4. The missteps were byproucts of trying to fit in too much too early.

Mind you you still see traces of Pentium 4 and Pentium M in Nehalem and Sandybridge. Albeit without the deficiencies of what you stated.

As for Bulldozer right now, I'm inclined to agree with you, BD needs quite a bit of refinement and in maybe a weird world where it would be easy to combine technologies but I guess i'm over-simplifying it. You've pretty much pointed out BD's weaknesses, the only problem now is how will AMD address the misprediction penalty?

 

Who knows? At this point in time; they're just upping clock speeds and IPC, which means they'll have to alleviate some of that branch prediction penalty that's kicking the IPC in the teeth.

Does mean, however, that Piledriver might have lower temps. Bulldozer is wieeeerd with the temperature scales. Phenom II Chips went through reasonable curves. BD chips just jump. There's no curve there. 30-*load*-50. 51. 52. So on and so forth. Even did the same thing under water, cheeky bugger! (Yes. I had a good application of thermal paste. Yes. The block was unclogged. And yes, I checked all of this on first install on air anyhow. BD is just weird.)

 

Ever OC'ed Pentium 4? While it was hot from the beginning, the leakage on some chips that occured at around 3.9+GHz was so obscene it beggared belief. Although I will admit, BD on load reveals an inherent issue since it's idle throttling is mighty impressive. Remember that this chip consumes a boatload of power. Couple it with an architecture that was desgined to run at higher clocks (Pentium 4 eh?, okay not really) but didn't make it to those clocks and you have some pretty crazy leakage.

 

Yup. That said; I can push this chip to well over the 4.65ghz that they achieved, I could even hit 5ghz if I balanced the Vcore enough, but it just gets so damn hot.

I would like to see the exact bios setttings used while benching, though. I'm curious. AMD have a load of safeguards in the Bulldozer chips that you have to disable to get them to move anywhere near quickly, from Application Power Management, which is a plug-in for turbo, which means it'll actually friggin' downclock if too many cores are loaded, to the typical Hardware Thermal Control, which just insta-throttles if the CPU even gets near 50 degrees centigrade. Which it will.

Also pushing the motherboard can yield better performance. The HTT link is plugged into the Memory Controller, so the higher you can get the motherboard and northbridge; the more memory bandwidth you open up for BD to work with, which can really boost scores in some tests.

 

I gotta take down notes for when I get myself an AMD build. I think it's running into the same frequency scaling issues that all chips ran into. With the higher frequency leads to higher voltages and leakage, and with that comes increased heat. But if anything, there is some truth and yet still some falsehoods about GloFlo's transistor manufacturing process. On one hand it's made to spec. On the other, the spec has dramatically upscaling leakage.

 

i cant see Piledriver being a huge step from BD - its the one after which should fix the issues.

 

It isn't, it's claimed to have 7-15% better performance. But that's to be expected. It's still using the same node, the same general architecture. The issues can only be incrementally concluded anyways.

 

A license to produce X86 is worth its weight in gold if done right.

 

any company would have to start from the ground up

 

Not really. AMD has a lot of expertise within them, it's just a long complicated chain of decisions that's lead them down this path. Once they're on the path; they need to continue. You cannot afford, in this line of business, to turn around and change your mind unless the resultant item is so ungodly awful that it's worth more if you go "Screw that. We need something brand new."

Bulldozer isn't that bad. It's slow, it's hot, but it has the ability to evolve.

 

This isn't true - AMD's FPU shrank area by 30% and reduced critical path by 20% when they switched to using synthesis tools.

 

Part of that is just the dieshrink, too.

 

would AMD be better making cpu's that compete with i3, and maybe leave the higher -end chips to intel?

 

Not really. Even the i3 is pretty damn fast in relation to some stuff. They need something at the high-end, too, because that's where most of the Server research goes, too.

Bulldozer is basically just a server chip, after all.

No, AMD's current best product is Llano. It's just a shame you can't get good FM1 motherboards. They're all low-end cheapy stuff. Give us a slightly more expensive, but feature packed FM1 motherboard and it'll be cool.

 

This is turning into the Ebay situation when that company decided it didn't want all those pesky loft-clearing casual sellers anymore, just the big boys and SMEs.

The equivalent here is AMD not wanting all us pesky but loyal/tight consumer desktop CPU buyers, only the GPU speed freaks.

I never liked paying Intel's motherboard tax, but if it gets me a working CPU with all cores [EDIT] effectively threaded then I'll just save up more dough for it and eke it out to 6 years instead of 5, and as AMD wants, just stick to Radeons within the next couple of years.

 

All physical cores? I think you're getting confused about the state of Hyperthreading. Even the I3 is just a Dual core, but hyperthreaded.

 

You caught me out - seems like Intel are able to make their hyperthreading more effective, so now I'm just dispassionate about CPUs, always used to set a £150 limit on them, now it's just who's got the best one, and any time before 2014 that's going to be Intel.

 

Part of it is that Windows 7 currently doesn't schedule properly for the Bulldozer cores without an optional Hotfix, and another part is that while bulldozer has more potential; It's limited by the much lower IPC over the Intel chip, so the bottlenecks where Hypertheading slows things down, (Hence the "Only" 30% boost in performance) Aren't so obvious as they are with the module design in BD.

 

Worng. The INT = doubled, the FP(u) unit is NOT.
http://nl.hardware.info/reviews/2382/3/amd-fx-8150--8120--6100--4100-bulldozer-review-modules

 

Whoops. I always get the two confused, t'ever. It's still a flaw in the design, you'd think with all the other doubled points; they'd alleviate that choking-point in the design.