AMD Fury X Owners' Thread

So Nvidia made the 1080 sound really good on paper and with their graphs boasting twice the power as a 980, but now that these benchmarks are slowing popping up (yes I know they are early benchmarks but gives you a ballpark idea of performance) I must say I am not that impressed.

Ashes Benchmark 1080 (1440p Crazy Settings)

Spoiler

My Fury TriX same settings

Spoiler

Look at the huge 1.5 fps difference.

 

You are Benching it at Ultra, that 1080 is Benchmarked on High.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
And if anyone with Fury X want my OC vBIOS. I can provide it.
https://onedrive.live.com/redir?res...62&authkey=!APfeuiYyrqHYmA4&ithint=folder,rom
I made it month ago. So, it is fully tested. (based on newest official vBIOS from AMD)

Made few sensible changes:
HBM OC limit 600MHz (+20% from 500MHz base)
HBM Voltage: Unchanged as even +50mV did not improve OC by 5MHz (people who made it to 600MHz+ usually did Hard Mod to capacitors)
GPU OC limit 1260MHz (+20% from 1050MHz base; 2000MHz limit looked too stupid to me on Fiji, maybe Vega)
GPU Voltage: +18.75mV as higher may lead to Power Limit Throttling
Power Limit: Bumped by 20% internally t support OC without decreasing performance as clock goes UP.

Could not find OverDrive +50% Power limit to decrease it to +25%, So for safety, do not put Software controlled limit over 25%.
(But from my testing, card refuses to draw more than 350W anyway, so VRMs which can do 450W+ are safe.)

So, it is not big modification, just little one making sense.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Notice:
I made dozens of custom voltage bioses for Fury X. This one above is optimal for me. Maybe someone will get higher GPU OC with higher voltage.

HBM observation:
Higher Voltage does not improve OC without HW Modification of card. Lower voltage does decrease OC potential. But even -50mV (1250mV instead of stock 1300mV will get HBM to stock 500MHz).
There is no reason to undervolt HBM for power efficiency as it eats less 18W at stock. And less than 25W even with bumped voltage and maximum stable OC.

GPU OC observation:
Card does not want to go much over 300W even with vBIOS modification for all power stages. So, voltage bump over +50mV is usually not good idea.

Where did it land my Fury X? Nothing extra: 1135/570MHz (No throttling.)
Passed all tests. I like GPUPI test as well, since single wrong calculation or memory R/W and it will tell you.

 

Which means, at least in this case, the Fury TriX is even that much better. Rather wish I had picked up a XFX Fury Non X but, this 980 Ti will last a couple of years at least before I upgrade anyways. (Came from a 290x so I did see large improvements in games that I was playing. Just wish I would have waited since I was impatient to upgrade in December of 2015.)

 

Performance level of currently available High end cards is fully sufficient for 1080p (100+ fps) and 1440p 60+ fps unless person has some masochistic need to use crazy level of SSAA and some other unwise stuff.

That is main reason why I think it is good, that AMD intends to bring Higher performance to cheaper cards. It holds back game evolution if average gaming PC has card which is 4times weaker than Available high end.

 

I will give it a Flash, I didn't know a higher power limit decreased performance.

 

Don't, It is vBIOS for Full Fiji (Fury X). I did not modify Nano/Fury-non-X.

But it is not Power limit increase which reduces performance, it is voltage increase without proper Power Limit settings.
As card is close to 300W and any bump in clock or Voltage may spike it over 300W at any time.
And GPU in same way as CPU inserts empty cycles.

 

Ahhhh okay I understand, that makes sense.

And I have flashed Fury X roms before, it causes no harm as they are all reference PCB's the only bios I wouldn't flash are the asus stixx fury bios as that card has a different vrm setup.

Edit: I think I have hit my Fury's limit v.v

I am currently stable (kinda, heaven have been on loop for about 24 minutes now) at 1180/570. Max voltage has been 1.288v and max temp so far is 44C. Max power draw according to HwInfro64 has been 385.194 watts between the core and hbm wattage. I have a crap ton of thermal headroom but my card wont draw more than 385 watts. If i push the voltage further and the core clock further I get negative scaling.

All in all I am sad that this is my cards limit it seems, but impressed with in overall.

Edit: Nvm, card just pulled 379.496 watts on the core and 17.875 watts on HBM but I think it is stable.

Spoiler

 

You may not notice few artifacts. Run GPUPI twice, it either shows correct output value, or throws an error.

 

Great job, Here
I will post this on my Thread. (OC is there also, not only RP ;-))

 

This is how OverDrive looks now.


VBIOS default clocks are still 1050/500MHz.

 

Yeah Great Job, im almost always leave Defaults in BIOS when i need to mod things up
And you are good in HEX editing i presume? :nerd:

 

Ran it a few times and it didn't error out.

Also I made a vBios of my own and would like to limit HBM between 525 and 570. And Limit the core the 1225, any chance you could help me with it?

This is the vBios, for the TriX. Voltage increased to 1.298 and power limit increased to a have a power draw of 375 watts. Base core clock set to 1180

https://drive.google.com/file/d/0BwhQfJR0F07hYTF2RUFSaG5kRlU/view?usp=sharing

 

Your vBIOS has currently limits for OverDrive OC:
GPU 2000MHz / HBM: 500MHz

Easiest way to mod that vBIOS for you is to look for:

Code:

[B][COLOR="Red"]40 0D 03[/COLOR][/B] 00 [B][COLOR="Blue"]50 C3 00[/COLOR][/B] 00 32 00 32 00 4D

Red is for GPU OverDrive: 1225 is 84 DE 01 in hex after transformation with 00
Blue is for HBM OverDrive: 525 is 14 CD 00 or 570 is A8 DE 00 in hex after transformation with 00

Hex conversion used is simple one. Take frequency like 500. Add two zeros: 50000.
Convert to Hex by any standard means (windows calc): C350. reverse Order to 50 C3 00

Once you're done editing, load it in Fiji BIOS Editor and save it again. It will fix CRC for you.

 

I did start my hex editing career when I was like 9 y/o on C64.
I learned some assembly, was directly editing data on diskettes. Sometimes I modified games, but usually I did reverse engineer saves.

 

He He lol
I started when i have AMIGA years ago for Moding SAVES for Games.
Usefull with Eye of The Beholder series etc.
Also the Game disk in RAM was originally my Idea in that Times ! (I have AMIGA with 2MB RAM 1MB Chip/1MB Fast so i manage to write into RAM disk the 2x Game Disk for Less Swaping and Faster Load Times)
Now im using pCache for that in PC -> Games have 1.5-2GB in RAM when first Load.
BF:4 or BF:H loads Map in MP in just seconds (3-4sec.)
Also i know "things" :nerd:

first was Pong in 1974 -> I play it on B/W TV in my House, that was gift from my Father :banana:
ATARI 800XL
C64 then 128 and with Disk 1/4
AMIGA
AMIGA 2000
AMIGA 3000
AMIGA 4000 -> That was a dream machine for me i was on her till
PC 486DX with 16MB RAM + GPU and Gravis Ultra Sound !
Then was AMD K2 etc. and 1 Intel P90, in time i had also VooDooo 32MB (Played Quake 2 on it :rpg
and so on till today

 

HH, good. I guessed you have few years on me, but it is like 14~16.

 

You know, im wondering how to find HEX values for GCN GPU's in .ROM file?
What i need to know, and what values i need to check.
I need to MOD my BIOS a little more
eg. i want to add new Power State
and GDDR CL T etc. gRAM Tweaks (as we have it in our BIOS for CPU and RAM)
also i will be glad to have +V for my gRAM (then it can go up to 1650-1700MHz!!! lol)
i will give you my bIOS, please take a look

-> my BIOS on MEGA

My Space in MEGA - lot of stuff there

Also what type of HEX editor i need to use for CRC check and corrections?

And move our conversation here -> http://forums.guru3d.com/showthread.php?t=406016

 

Nice Thank you sir, appreciate it.

 

Could anyone tell me what the best way to check for errors when overclocking my Fury X's memory would be? Something like Unigine Heaven seems to work fine for GPU clock but the memory clock might work okay in that and have issues in Witcher 3 if I set it too high, and then only sometimes. I'm not sure if there's a reliable way to tell quickly if the memory clocks are too high.

 

HBM on Fiji is weird thing. To understand it, do run stock GPU clock.
And OC only HBM. OC it by 10 MHz and benchmark, OC again and benchmark, again...

Then you realize there are certain clock steps where performance actually increases. For me, HBM transfer rate is same in following bands:
473~522
523~572
573~622
So, I have no different performance between 523MHz and 572MHz. In other words, if HBM does not clock 573MHz or more stable, then I can as well sti at 523MHz.

What I noticed:
Bumping Voltage to 1350mV makes HB seemingly stable at 573~623MHz. Games do not artifact. (JEDEC document states HBM should survive up to 1.5V on VDDC, but it is not recommended to use such voltage.)
But luxmark performance goes down/up a lot (unstable). And GPUPI fails to calculate proper result as HBM data are lost.

And I use 'OclMemBench' to see if there is change in actual transfer rate too.

And since there is GPUz 0.8.8 for some time, which can read ASIC for Tonga/Fiji... I have 70.3% ASIC Quality and it states 'Quality is higher than 100% of similar GPUs in database'.