AMD Gimping HBM clock on Fiji after v16.12.2 WHQL driver

[gupsterg]

I get artifact at desktop above 545MHz HBM on v16.x.x driver. So I can't give a result of any higher on v16.x.x driver.

Only 1 card out 11 Fiji I had did 600MHz HBM with HBM voltage of 1.4V on v16.x.x driver, I no longer have that card.

There is thread on OCN where I did 250+ runs of 3DM 13 FS FSE FSU at various VGPU/GPU_CLK/HBM_CLK. I did 3 runs of each setup of card and then there where ~9+ or so setups of card plus 3 differing 3DM full bench runs. This was on my i5 4690K rig which did 4.9GHz CPU 4.4GHz Cache 24/7.

www.overclock.net/t/1621676/fury-x-3dmark-13-mega-bench-fs-fse-fsu-over-250-runs-by-gupsterg

 

[gupsterg]

Having been using AMD cards since like forever, rage fury maxx and on with only one hick-up along the way, it's this sort of thing and the mess that was the polaris and vega graphics cards under delivering, was an early adopter for pretty much all ati and amd graphics cards but the past 2 generations have been a complete let down, as was the fury x "being touted" as the overclockers dream which it most certainly was not.

PCB great.

Cooling soloutions air and AIO great on Fiji.

Yeah OC headroom sucked but not totally disappointing for me.

Luv'd the card length and 'package' as a whole. VEGA has no need for as big a PCB IMO. Even AIO I doubt they needed to make as enhanced within shroud, Fury X was nice and simple AIO IMO.

I can live with not much gains in driver development over it's term, but come on AMD don't now block what we had for months and months. Total sham.

 

[Klo]

Does it have a game changing effect on FPS?

If not, who cares. If it does, fair enough with the upset.

 

[Gerard]

e.

Luv'd the card length and 'package' as a whole. VEGA has no need for as big a PCB IMO. Even AIO I doubt they needed to make as enhanced within shroud, Fury X was nice and simple AIO IMO.

Vega could have had a much smaller pcb, there's a lot of empty space on it. Just seems like they kept it large to let third parties use the reference pcb and mount different coolers to it without having to have a framework attached to keep an overhanging heatsink stable.

 

[Andy]

Thankfully overclocking is supported and quite capable on Vega and HBM2.

For how long though?

 

[LtMatt]

For how long though?

It's not going to be removed because it is supported and enabled in the driver for Vega. This is what makes it different to Fiji and HBM1 in our driver.

 

[ubersonic]

It was working on 16.x and its blocked on 17.x drivers

It's not that simple.

It was unsupported and disabled on 16.x however hacks/workarounds/third party apps made it available.

It is also unsupported and disabled on 17.x however the same hacks/workarounds/third party apps aren't compatible with the newer driver design. And instead of badgering the app makers to redesign their apps for the newer drivers people seem to be moaning at AMD over it lol.

 

[Deleted User 234324]

Can't say I really trust what AMD has to say these days, the only thing they have got right in the past decade is Ryzen/Threadripper and even that had issues to start with. Everything else has been a lie or at the very least stretching of the truth.

 

[gupsterg]

It's not that simple.

It was unsupported and disabled on 16.x however hacks/workarounds/third party apps made it available.

It is also unsupported and disabled on 17.x however the same hacks/workarounds/third party apps aren't compatible with the newer driver design. And instead of badgering the app makers to redesign their apps for the newer drivers people seem to be moaning at AMD over it lol.

3rd party apps can't do anything when the driver is blocking performance gain from clock change. VBIOS changes on HBM clock can not do anything on v17.x.x driver, VBIOS is pretty much AMD code. Please understand this.

I have repeated shown that using no third party app and only modifying VBIOS to have clocks, I get no gain on v17.x.x driver.

Here is video for your viewing pleasure, driver is not hacked at all, I can do a video of downing a fresh one and installing live.

This video has only where Fiji VBIOS has increased OverDrive limit, like all other AMD cards have. Notice how I have slider in OD for HBM and changes to HBM clock have an affect. The driver panel for hardware info will show default HBM clock, even when changed, like GPU-Z shows VBIOS default on bottom line, OverDrive will show current HBM on tacho dial/slider when changed, as will GPU-Z on appropriate line. As shown in video there is an affect to GPGPU benchmark with HBM clock change, so slider/driver is applying clock and giving performance gain.

 

[eddyr]

Can't say I really trust what AMD has to say these days, the only they have got right in the past decade is Ryzen/Threadripper and even that had issues to start with. Everything else has been a lie or at the very least stretching of the truth.

I dunno, when they say they don't support something and it is not supported and make software updates as though they don't support it. It seems pretty clear cut. It sucks but is what it is.

3rd party apps can't do anything when the driver is blocking performance gain from clock change. VBIOS changes on HBM clock can not do anything on v17.x.x driver, VBIOS is pretty much AMD code. Please understand this.

I have repeated shown that using no third party app and only modifying VBIOS to have clocks, I get no gain on v17.x.x driver.

Spoiler

Here is video for your viewing pleasure, driver is not hacked at all, I can do a video of downing a fresh one and installing live.

This video has only where Fiji VBIOS has increased OverDrive limit, like all other AMD cards have. Notice how I have slider in OD for HBM and changes to HBM clock have an affect.

Im fairly sure vbios mods aren't officially supported.

 

[gupsterg]

Im fairly sure vbios mods aren't officially supported.

Please understand this, I am not contesting anything regarding official support.

Next understand this any driver from ~ June 15 to Jan 17 I can have HBM performance gain from clock change. But I can't on a driver after Jan 17.

Fiji is not in retail channel.

So why block a performance gain on a product not on sale?

What harm is it doing AMD? it is only frustrating for an owner of Fiji.

 

[LtMatt]

Im fairly sure vbios mods aren't officially supported.

Editing the bios to make the slider appear in Overdrive does not change anything regarding support.

No performance gain was deliberately blocked, it was most likely as a result of Overdrive being replaced and going EOL by Wattman.

As said, it's clear that AMD isn't going to start supporting it, but not actively seeking to disable it is a different matter. If appears that effort was put in to stop this, but it's also possible it was an unintended consequence of another change, and then I can see that it's just what it is and we accept it.

I think this guy gets it.

 

[pandem0nium]

To make the change they'd have to for someone to raise a change request, document the requirements. Then one day of a developers time to make the change, maybe longer to write a unit test as well. Then the QA team would have to build it into their test plan for release testing of the next driver.

It's not worth it and they're just not going to do it.

 

[ubersonic]

To make the change they'd have to for someone to raise a change request, document the requirements. Then one day of a developers time to make the change, maybe longer to write a unit test as well. Then the QA team would have to build it into their test plan for release testing of the next driver.

It's not worth it and they're just not going to do it.

The request would be kinda funny "we need to introduce a bug into the drivers because some people were exploiting one in the old drivers and they're sad it's not there anymore" XD

 

[gupsterg]

Editing the bios to make the slider appear in Overdrive does not change anything regarding support.

Did not mean or imply that. It was to show the driver/OD without 3rd party apps could do the change.

No performance gain was deliberately blocked, it was most likely as a result of Overdrive being replaced and going EOL by Wattman.

No it was not because of OverDrive going EOL.

OverDrive went EOL as Crimson ReLive came out with Wattman support for other cards, ~ June 16. Like stated many a time every driver from June 15 to Jan 17 allows performance gain from HBM clock change. Just to clearly state it I can use a driver from June 16 to Jan 17 which has Wattman and I can change HBM clock and gain performance. May that be VBIOS or 3rd app.

It is Jan 17 onwards drivers that AMD have borked for Fiji.

To make the change they'd have to for someone to raise a change request, document the requirements. Then one day of a developers time to make the change, maybe longer to write a unit test as well. Then the QA team would have to build it into their test plan for release testing of the next driver.

It's not worth it and they're just not going to do it.

If they made no change in v17.x.x driver why do I and several owners on OCN have this issue?

 

[LtMatt]

No it was not because of OverDrive going EOL.

OverDrive went EOL as Crimson ReLive came out with Wattman support for other cards, ~ June 16. Like stated many a time every driver from June 15 to Jan 17 allows performance gain from HBM clock change. Just to clearly state it I can use a driver from June 16 to Jan 17 which has Wattman and I can change HBM clock and gain performance. May that be VBIOS or 3rd app.

Overdrive was still present after 16.12.1 as it was initially only supported on a handful of our GPUs. HD7xxx series users, for example, were still using Overdrive at this time, so Overdrive only went EOL once all supported products were using Wattman.

Take a look back at the release notes.

http://support.amd.com/en-us/kb-art...son-ReLive-Edition-16.12.1-Release-Notes.aspx

• Radeon WattMan(3): Added support for AMD Radeon R9 Fury series, R9 390 series, R9 380 series, R9 290 series, R9 285, R9 260 series, R7 360, and R7 260 desktop graphics products

Even if this stopped working after Wattman was supported on all GPUs, it does not change anything I'm afraid.

 

[gupsterg]

I have Wattman in v16.12.2, I have performance gain from HBM clock change.

It is from v17.x.x AMD have gimped Fiji.

But I roll over Matt.

If I can't get an AMD rep to agree that AMD have blocked performance from 2017 onwards I got no chance getting it stated further up the chain.

Clearly I need to rethink buying AMD products in the future.

Thank you for your time and I hope I did not upset you. I merely wanted my HW still to perform at it's best going on in the future, like it has done for over a year, clearly I can't have that.

All the best chap. I shall stay mute from now on the subject.

At least you can now breath a sigh of relief.

 

[Zeed]

I have Wattman in v16.12.2, I have performance gain from HBM clock change.

It is from v17.x.x AMD have gimped Fiji.

But I roll over Matt.

If I can't get an AMD rep to agree that AMD have blocked performance from 2017 onwards I got no chance getting it stated further up the chain.

Clearly I need to rethink buying AMD products in the future.

Thank you for your time and I hope I did not upset you. I merely wanted my HW still to perform at it's best going on in the future, like it has done for over a year, clearly I can't have that.

All the best chap. I shall stay mute from now on the subject.

At least you can now breath a sigh of relief.

You see why i went Pascal over Vega ??
Now **** is blocked on AMD Vega and NV Pascal but at lest NV delivers performance without need of Bios/driver mod's ofc I could do hard resistors mod on my Titan but tbh... it delivers so much performance its FIRST TIME i dont feel like i need to...

 

[pandem0nium]

Did not mean or imply that. It was to show the driver/OD without 3rd party apps could do the change.
If they made no change in v17.x.x driver why do I and several owners on OCN have this issue?

I haven't said that though, have I?

We don't know if it was deliberately changed or as a side effect of another change to the drivers. I'd give them the benefit of the doubt and go with the latter.

Lets say I own a Greggs store and there's a hole in the wall in the side of the store where people in the know can reach in and grab a steak bake.

After a while we move store to a new premises that doesn't have that hole in the wall. Those that had the free steak bakes will be disappointed but it was never part of the service offered by my Greggs.

 

[LtMatt]

I have Wattman in v16.12.2, I have performance gain from HBM clock change.

It is from v17.x.x AMD have gimped Fiji.

But I roll over Matt.

If I can't get an AMD rep to agree that AMD have blocked performance from 2017 onwards I got no chance getting it stated further up the chain.

Clearly I need to rethink buying AMD products in the future.

Thank you for your time and I hope I did not upset you. I merely wanted my HW still to perform at it's best going on in the future, like it has done for over a year, clearly I can't have that.

All the best chap. I shall stay mute from now on the subject.

At least you can now breath a sigh of relief.

No hard feelings, i know where you are coming from. Sorry it's not the answer you were hoping for.

I haven't said that though, have I?

We don't know if it was deliberately changed or as a side effect of another change to the drivers. I'd give them the benefit of the doubt and go with the latter.

Lets say I own a Greggs store and there's a hole in the wall in the side of the store where people in the know can reach in and grab a steak bake.

After a while we move store to a new premises that doesn't have that hole in the wall. Those that had the free steak bakes will be disappointed but it was never part of the service offered by my Greggs.

Certainly beats the old car analogy.