AMD Fury X Memory unlocked, voltage coming

[VanayadGaming]

Hello, 

 

It seems people at  Hardware.fr managed to unlock memory! 

 

 

Update from: @Briggsy

Trixx version 5.0 is available now from Sapphires website (since July 2nd), and provides memory overclocking without any tweaks up to 750Mhz (specifically for HBM). If voltage unlock is possible, then it will be a race between Afterburner and Trixx.

 

screenshot

 

 

Old Story:

 

 

Funnily enough, shortly after the story went live we were approached by Hardware.fr’s editor Damien Triolet. Who shared with us some very interestingly finding. By simply checking the “extend official overclocking limits” option in the MSI Afterburner overclocking utility, the memory frequency slider would appear in AMD’s overdrive utility. Not only that, but by doing so memory overclocking would actually be unlocked and any adjustments made to the memory frequency would in fact take effect and result in performance gains. 

 

 

People at overclock.net have also confirmed this story : http://www.overclock.net/t/1547314/official-amd-r9-radeon-fury-nano-x-x2-fiji-owners-club/1970#post_24118524

 

 

 

 

Personally I am still waiting for something official, but I am really curious on how much this card can overclock with higher voltages...

 

 

main source: http://wccftech.com/unlock-memory-overclocking-amd-r9-fury/

[Grrizz]

You may want to read the article The voltage was not locked its just no-one has software to control the regulator yet and that is still the case but it's being worked on, Hardware.fr only figured out the memory overclocking.

[ZetZet]

What. That makes me wonder. You overclock the fastest memory on the planet and gain performance?? How does that even work. Is it memory bottlenecked?

[Notree]

Yah I'm definitely interested in seeing when they get gpu voltages unlocked.. As someone who overclocks the 980ti is a better value right now .

[VanayadGaming]

You may want to read the article The voltage was not locked its just no-one has software to control the regulator yet and that is still the case but it's being worked on, Hardware.fr only figured out the memory overclocking.

Thanks

[ZetZet]

Thanks

How do you know voltage unlock is coming

[Grrizz]

How do you know voltage unlock is coming

It sounds like the voltage isn't locked its just AMD don't have a standard voltage control in the API like what Nvidia does so people now need to figure out the low-level comms required to communicate with the regulator. There was a mention of a rumour Sapphire are working on it and there are people on the OCN forums also trying to hack it by probing the boards.

[Lays]

What. That makes me wonder. You overclock the fastest memory on the planet and gain performance?? How does that even work. Is it memory bottlenecked?

 

 

That's really not how it works in my experience ANY change to your GPU's performance, whether it's memory or clock speed always increases performance in GPU limited scenarios.  I've overclocked every card I've owned, the original Titan, 3 780's, 2 970's, 2 290x's, a 780 ti and now a 980.  Everytime I was GPU limited, I always gained performance from any types of overclocks at all.

I'm not entirely sure of the scientifics of it, but I'd assume making the memory faster helps with latency + transfering etc.

[Grrizz]

What. That makes me wonder. You overclock the fastest memory on the planet and gain performance?? How does that even work. Is it memory bottlenecked?

I think it comes down to two things, one being the bump in bandwidth is big but its not as big as, at least I, was expecting, the interface is huge (4096bit) but with the lower clocks (500MHz) its actually only a ~30% increase in bandwidth over a 390X so with higher resolutions and a lot of AA you are probably still stressing it, the second thing, which would probably have much less impact is the fact an increase in bandwidth generally comes with a decrease in latency (that's a very general statement but if you are using enough of the bandwidth its mostly true) so you will be shaving some time off each frame just in transfer times.

If the tests were done on 1080p with little or no AA the impact would probably be substantially less but those tests are on 1440p with upwards of 8xMSAA and still most of the time when you look at the fps gain as apposed to the % gain (shown above) its actually only a couple of fps (more often less).

 

[patrickjp93]

What. That makes me wonder. You overclock the fastest memory on the planet and gain performance?? How does that even work. Is it memory bottlenecked?

No, games do not have a bandwidth bottleneck. I suspect this comes down to drivers and pipeline design in the games. The ROPs are the choke point. The only thing overclocking the memory would do is reduce latency. The 290X has now surpassed the 780TI despite having lower bandwidth. If that's not proof enough bandwidth isn't the issue, nothing ever will be.

[ZetZet]

It sounds like the voltage isn't locked its just AMD don't have a standard voltage control in the API like what Nvidia does so people now need to figure out the low-level comms required to communicate with the regulator. There was a mention of a rumour Sapphire are working on it and there are people on the OCN forums also trying to hack it by probing the boards.

so it's not coming, just people trying to make it work. that is exactly what I wanted to hear.

[qwertywarrior]

What. That makes me wonder. You overclock the fastest memory on the planet and gain performance?? How does that even work. Is it memory bottlenecked?

I asked this question on reddit

seems like bandwidth is never enough when it comes to GPU im sure there is a limit but from what i understand its far away

[SubZeroKelvin]

My translation:
 

In order to overclock the memory, which by default is not possible, we had to check the "Extend official overclocking limits" option in MSI Afterburner. Then, after rebooting to windows, we could modify the memory frequency even in the Overdrive section of CCC.

I wonder why nobody tried this before!

[Techhog]

You may want to read the article The voltage was not locked its just no-one has software to control the regulator yet and that is still the case but it's being worked on, Hardware.fr only figured out the memory overclocking.

I fucking called it

[carolkarine]

What. That makes me wonder. You overclock the fastest memory on the planet and gain performance?? How does that even work. Is it memory bottlenecked?

Well, it's only fast in one way. If you think about it, it may have loads of bandwidth, but you've still got more latency in the memory controller due to really slow clocks. I expect that this can be optimized with drivers as nobody has faced this kind of problem with ddr5 before, but for now we can brute-force the problem by overclocking the memory.

[carolkarine]

I asked this question on reddit

seems like bandwidth is never enough when it comes to GPU im sure there is a limit but from what i understand its far away

I assume that despite having assloads of bandwidth, the issue is that there's latency in the memory controller due to low clocks. It's now a large pipe with a slow flow, and by speeding up that flow you solve secondary problems.

[Texhnlyze]

I'm just excited how this card will overclock. We can finally see who wins this time around. Is it going to be the 980ti or FuryX?

[jos]

Done by someone else before

 

The people over at uk.hardware.info have managed to overclock the HBM memory of the R9 Fury X by 20% through a Catalyst Control Center (CCC) glitch.

 

Apparently after every few system reboots, a Memory clock slider in CCC Overdrive became available:

 

Read more at http://linustechtips.com/main/topic/396116-fury-x-memory-overclocked-by-20/

[Doobeedoo]

Well, nothing official yet but still something to look for. Excited to see HBM OC and how it will do in future games, specially when games can elevate more speed and memory.

[Megahurt]

What. That makes me wonder. You overclock the fastest memory on the planet and gain performance?? How does that even work. Is it memory bottlenecked?

My guess would be that it works in series, the gpu has to wait for the data before it can process it. If this is how it works, it means you'll get results regardless of how high the bandwidth is, you just get diminishing returns the higher you go.

[Briggsy]

@VanayadGaming
 
maybe can add this to OP if you want (not worthy of its own thread, but this is official). 
 
Trixx version 5.0 is available now from Sapphires website (since July 2nd), and provides memory overclocking without any tweaks up to 750Mhz (specifically for HBM). If voltage unlock is possible, then it will be a race between Afterburner and Trixx.

 

screenshot

[jos]

@VanayadGaming

 

maybe can add this to OP if you want (not worthy of its own thread, but this is official). 

 

Trixx version 5.0 is available now from Sapphires website (since July 2nd), and provides memory overclocking without any tweaks up to 750Mhz. If voltage unlock is possible, then it will be a race between Afterburner and Trixx.

 

Do you have any idea what HBM is.????

[dalekphalm]

Do you have any idea what HBM is.????

This comment doesn't make sense at all.

 

Yes, @Briggsy knows what HBM is - high bandwidth memory. 500 MHz base clock. He's saying Sapphire's new Trixx OC utility can OC the memory up to 750 MHz.

What remains to be seen, is how far you can push it while remaining stable.

[jos]

This comment doesn't make sense at all.

 

Yes, @Briggsy knows what HBM is - high bandwidth memory. 500 MHz base clock. He's saying Sapphire's new Trixx OC utility can OC the memory up to 750 MHz.

What remains to be seen, is how far you can push it while remaining stable.

Felt like he was just advertising and mentioned as if it is same as overclocking normal card. he didnt specify it support HBM. may it does may be it does not... I thought they used same crystal. why AMD why using 2 different crystals for the chip.. could have used a frequency divider easily...

[Briggsy]

Felt like he was just advertising and mentioned as if it is same as overclocking normal card. he didnt specify it support HBM. may it does may be it does not... I thought they used same crystal. why AMD why using 2 different crystals for the chip.. could have used a frequency divider easily...

 

Its a thread about unlocked memory on the FuryX, and unless someone is completely lost and unaware of what thread they are in, its pretty clear we are talking about HBM. next time ask for clarification if you are confused, instead of posing an insinuating question.