Crazy old overclocking, real world?

[DARBADARBZ]

https://www.asrock.com/news/index.asp?cat=News&ID=957

 

Would this make this CPU good by today's standards? How much better does making this chip go so much faster increase real world speeds? Am I the only one super curious to see how much better +4.1Ghz looks like in real life?

[Levent]

12 minutes ago, DARBADARBZ said:

Would this make this CPU good by today's standards

Absolutely not.

[DARBADARBZ]

4 minutes ago, Levent said:

Absolutely not.

how/why?

[8tg]

Raw clock speed overclocking like this isn’t the same as how modern CPUs function. If you want 7ghz+ you’re looking at one functional core, on liquid nitrogen, it’s incredibly unstable and is done solely for the verification and then usually that’s game over for the hardware after a few runs of that.

 

If you had some absurdly cpu intensive but single core workload, an old FM2+ apu at nearly 8ghz is still going to be trounced by a current i3 at stock speeds.

[Levent]

2 minutes ago, DARBADARBZ said:

how/why?

Just because you clock your CPU to 8ghz doesn’t mean it’s stable or performance scales as much. Usually these CPUS are just stable enough for CPUZ validation. 

[starsmine]

19 minutes ago, DARBADARBZ said:

how/why?

because the IPC of that processor is so low

Thats a 4 core piledriver without l3 cache. at that clock speed it will be memory bottlenecking stalling a lot of those clocks. And 4 cores in the bulldozer line is 4 cores (im sorry, it is, i know people want to argue its not) but its only has two front ends (hence why people argue its not). 

bulldozer pick, no one cared about getting pictures of later generations of it 


Basicly it bottlenecks itself in getting data unless you can have your program stay inside the 4MB l2

And even when it can get that, anything with floating point it will only act like one core. So almost anything AI related(which would be bigger then 2MB anyways), even when doing FP8/16 (which the processesor can not natively handle)

Here if you want to know more about piledriver https://en.wikipedia.org/wiki/Bulldozer_(microarchitecture)#Bulldozer_core

Notice how one core can only dwo two ALU and two AGU opperations per clock per core


A single Zen 2 core can do 4 ALU + 3 AGU per clock per core, and a far more robust front end to prevent bubbles in the pipeline with better branch prediction

[jaslion]

Performance doesnt scale linearly.

 

It often happens rhat when you oc performance goes down as its a bad overclock that is stable enough.

 

Even if it would br perfect an i3 8100 would still beat it with ease and it not like those are super usable these days.

[Somerandomtechyboi]

21 minutes ago, starsmine said:

because the IPC of that processor is so low

^^^

even 1st gen core beats zambezi/vishera let alone the cacheless trinity apus

 

23 minutes ago, starsmine said:

Thats a 4 core piledriver without l3 cache. at that clock speed it will be memory bottlenecking stalling a lot of those clocks

im curious if ram oc ever helped amd fx or the old apus cpu performance wise, imcs were pretty decent with zambezi/llano and ddr3 3200+ seems like its possible though vishera/trinity stuck to ~3000 max

 

i do have fx but my 970 extreme3 is crippled for some reason maybe due to socket cover swap, used to be able to hit 2700 ish on 2 sticks for a frequency run (likely higher had i used better sticks cause i might have used one of my worst psc sticks) but now struggles to get over 2400, so im gonna have to buy another board if i wanna test for myself and im kinda focused on trying to get over the 3400 mark on x58 with a bloomfield chip atm so id rather spend my cash binning more w3503 or buying a gulftown chip

 

also got fm2(+) but boards are all trash and yeah id rather play with fm1 llano that has a better imc for 3200+ shenanigans instead of being stuck at ~3000 and getting bored instantly