Overclock.net - An Overclocking Community - View Single Post - FX giving bad FPS? Here's how to tune your old FX to give better in game performance...

View Single Post
post #1 of (permalink) Old 08-31-2017, 08:51 PM - Thread Starter
Minotaurtoo
Useful idiot
 
Minotaurtoo's Avatar
 
Join Date: Jun 2013
Location: Florence, AL
Posts: 2,396
Rep: 80 (Unique: 46)
Here is a post I put in the 83xx FX owners club... I was going to re-write it, but it's pretty cut and dry already so I just copied / pasted it here to give it a stand alone title.

The problem seems to be that many FX users are stuck with their cpu or are wondering if they "need" to upgrade to Ryzen or Intel... well I say not really... not yet anyway... If you want to and have the money by all means go ahead, but if you are like me and kinda broke for the moment there is good news...FX can be tuned... if you have already tuned yours like described below and are still having trouble with cpu bottle-necking in games I would suggest finding out if anyone you know who has a faster cpu is having trouble as well because this literally gave me a 29% performance increase in the game tested... I haven't had time to test all games thoroughly yet, but results are already looking good in what I have tested... Enough of my rambling, here's the post:

I have been hearing people complaining or lamenting about FX not performing well in games and one person even said they couldn't get good fps in games...personally I haven't had much if any issues... the fps they were quoting were much lower than I was seeing in games so I got to wondering why that was... I saw from their benches and from some online that I was doing much better even with them having better gpu's... so I decided to try something...I went back to stock.... been so long since I OC'd this thing I actually forgot what all I had done redface.gif

Note: before doing any overclocking be sure you have adequate cooling... also if you haven't upgraded from stock cooler, that might be your problem as FX cpu's under load will generally thermally throttle under load on stock coolers.

All tests were run at 1080p ultra preset with CMAA on Dirt Rally

Dirt 4 was mentioned... and since Dirt 4 has no canned benchmark and since Dirt Rally isn't that much older and I get similar FPS in it as I do in Dirt 4 I used the benchmark in it to test why others were having so much trouble.... turns out when I went to stock even on a 9590 with 4.7ghz base I had mins around 59 and average around 80... this literally blew my mind because those were horrible compared to what I had been seeing even at 4k it never dropped below 60 unless I had vsync on...

I then loaded my 5ghz profile and tested it... got 108 average and 75 min.... ok that was expected... I looked in my OC profiles and discovered that I had overclocked more than just the cores.... I really have forgotten since it had been a long time ago... but the cpu/nb was at 2600mhz along with the ram and HT being OC'd as well... then I decided to push for all it was worth... at same core speeds I pushed cpu/nb to 2700mhz and ht to 3000 mhz... then I got 111 average and 76 min

I tried for more, even pushed up to 2800 cpu/nb 2500 on ram and 5.2ghz on cores with no more gain... so looks like I might have gotten rid of the cpu bottleneck before it happened lol... no wonder I didn't feel like FX was that old yet.....


btw, I rounded off those numbers.

I'm posting it here just in case anyone needs to get more performance and haven't considered those options... hope it helps someone

proofs:

stock 4.7ghz (5ghz boost) with 2400mhz ram so this will be worse if you have less than 2400mhz ram


I'm adding this one to show what just oc'ing the cpu/nb and manually setting clocks does
stock 4.7ghz with cpu/nb OC of 2600mhz and DOCP ram profile selected and turbo core disabled



5ghz with 2400 mhz ram and 2600 cpu/nb that's a 29% improvement on mins over pure stock with only a 6% core clock increase... so that cpu/nb really makes a difference... averages improved by 36%...

5ghz with bus overclock, 3000mhz ht, 2700mhz cpu/nb and 2440mhz ram


I think this is the best I'm going to get overall... 5.067ghz 2432mhz ram 2635 cpu/nb ...


verification of settings:




Here is the Cinebench results... it's a little confusing as all are shown together but here is a little guide
10. is bone stock with only memory setting plugged in manually
9. is stock clocks with turbocore disabled and DOCP profile activated and cpu/nb changed to 2600mhz
8. is a test OC using only the base clock to achieve 5ghz...mem and such were kept as close to stock as possible but I think because of the way multi's worked out the cpu/nb was near 2500mhz to run the ram at 24xx
7. is a tuned straight up 5ghz OC with 2600 cpu/nb and 2400mhz ram settings
6. is a tuned version of #8 with 2700mhz cpu/nb and 2440 on the ram with 3000 on HT
5. is a straight 5.1ghz OC with only ram speed and timings manually entered
3. is the one I'm proud of... I made it beat out a 4770k... core is 5.117ghz with 2600 cpu/nb and 2400mhz on ram I could run this 24/7, but the heat was a bit edgy for me... once I get ready to ditch this rig I'll push it lol.



Timespy with stock clocks (optimized defaults)


Timespy with stock clocks but cpu/nb 2600mhz and 2400mhz ram:


Timespy with 5ghz core clock 2600 cpu/nb and 2400 MHz ram:


The part I find interesting here is how just overclocking the cpu/nb and the ram gained more performance than overclocking the cores by 300 mhz. I may go for a "balls to the walls" run on the cpu/nb to see just what I can get out of stock core clocks... I'll try to do some more testing, but it's looking pretty conclusive that cpu/nb clocks matter more than core clocks.... at least when at 4.7ghz or above...


Here is a bit of a random link... only one bench here at my daily run settings... but according to the cpu test mine is in the 100th percentile of all 9590's tested...that should say something about my "tuning" since there are many people who achieve 5 ghz + on 9590's... http://www.userbenchmark.com/UserRun/4758334


Geekbench results here:

stock 4.7ghz

stock 4.7ghz with cpu/nb 2600mhz

5ghz with cpu/nb 2600

I think the memory test at the bottom tells the story the best... the differences are astounding for such a low overclock of the cpu/nb

API test using 3dmark... I find this interesting as well... seems memory speeds and cpu/nb clocks have a dramatic affect on API

4.7ghz stock (optimized defaults)


4.7ghz with cpu/nb 2600mhz and 2400mhz ram


5.0 ghz with cpu/nb 2600mhz and 2400mhz ram


Another Impressive result from Gapottberg... very good research on his part:
https://www.overclock.net/t/1637388/fx-giving-bad-fps-heres-how-to-tune-your-old-fx-to-give-better-in-game-performance/220#post_26484039

From Jaredismee: https://www.overclock.net/t/1637388/fx-giving-bad-fps-heres-how-to-tune-your-old-fx-to-give-better-in-game-performance/230#post_26491850


here is a post where someone else has achieved good results from cpu/nb oc'ing pared with core oc https://www.overclock.net/t/1637388/fx-giving-bad-fps-heres-how-to-tune-your-old-fx-to-give-better-in-game-performance/30#post_26334833

Just thought this was a cool explanation.. not perfect, but still pretty accurate: https://www.overclock.net/t/1637388/fx-giving-bad-fps-heres-how-to-tune-your-old-fx-to-give-better-in-game-performance/210#post_26469414

Wonder if I should do a youtube video showing how to add life to your old fx cpu lol...

Intel who?


Minotaurtoo is offline