AMD silently nerfing the performance of older GCN cards

[IGJoe2192]

40 minutes ago, Humbug said:

LOL Not supporting GPUs means that you don't spend development time on them getting new features working or optimizing for newer games. In that case what AMD does is they tell us GCN 1 support is end of life. And then future driver updates do not support them.

 

This is different. We have a feature which was already working. They released a driver which supports GCN 1 and they disabled said feature.
 

There is a problem that needs to be fixed it has to be fixed by AMD. The community will ask them to fix it. The product is not end of life, market share talk should not come up.. even if only 1 user remains it is AMD's responsibility to fix it.

 

Besides for AMD do you think that this is a tiny percentage of their userbase? No, it's a noticeable portion of the AMD GPU userbase. Only when you look at the total GPU market it will be small because AMD's market share is small.

 

You are assuming a lot... Have a nice day.

[79wjd]

18 minutes ago, IGJoe2192 said:

Nope, not enough. Move along.

So Nvidia should just stop supporting Pascal? The 1060, 1070, and 1080 only have a 4% share on Steam. 

[IGJoe2192]

50 minutes ago, djdwosk97 said:

So Nvidia should just stop supporting Pascal? The 1060, 1070, and 1080 only have a 4% share on Steam. 

They don't have a decreasing market share. in DX12 GPU's alone the HD7900 series itself has gone from 1.98% down to 1.69% share in the last 5 months... AMD doesn't see money for HD7900 sales anymore, and they are declining in the market. What all of you also don't realize is that while these old cards may be in systems on Steam's survey that does not mean they are using the limited number of games that leverage async compute. So in all honesty how many customers does this "hurt"? Why are you so pissed about if you do not have that hardware...

[79wjd]

19 minutes ago, IGJoe2192 said:

They don't have a decreasing market share. in DX12 GPU's alone the HD7900 series itself has gone from 1.98% down to 1.69% share in the last 5 months... AMD doesn't see money for HD7900 sales anymore, and they are declining in the market. What all of you also don't realize is that while these old cards may be in systems on Steam's survey that does not mean they are using the limited number of games that leverage async compute. So in all honesty how many customers does this "hurt"? Why are you so pissed about if you do not have that hardware...

I don't have an issue with them cutting driver support, but removing a feature that was once there is a very different thing. 

 

I don't have a lot of things, that doesn't mean I shouldn't care. Now if I had no intention of every buying a GPU or an AMD GPU, then sure, I could see the argument that I shouldn't care, but that's not the case. 

[Notional]

So according to AMD, this is an issue they are looking into. If it is indeed an issue, it will be solved, as it would be considered a bug. Source: 

At the very end.

[Shahnewaz]

http://gpuopen.com/concurrent-execution-asynchronous-queues/#comment-6937

Quote

Alessio T says:

December 12, 2016 at 8:49 pm

Could you please re-enable asynchronous compute queues on GCN1 GPUs (7700-7900, 280, 370 etc…) under Direct3D 12? It worked fine without major issues compared with newer GCN iterations. But since many months it does not work at all under Direct3D 12 with GCN 1 (while there are no issues under Vulkan). Moreover the hijack to the direct-queue is causing stuttering, which force developers to create and patch with ad hoc path for GCN 1 GPUs and remove queues concurrency.. This simply kinda sucks and reminds me how bad serialization of concurrent queues were under NVIDIA Maxwell GPUs until some months ago. GCN1 GPUs are still sold around the word, and most of them (Tahiti and Pitcairn and their re-brands) are still very-performance valid GPUs. Moreover there is no reason to disable a feature that just worked fine. Thank you for the attention.

Quote

Stephan Hodes says:

December 13, 2016 at 9:41 am

While async compute worked fine in *almost* all cases on GCN1, we identified some corner cases where it could cause severe issues. We’ve discussed several ideas how the driver could identify and work around those corner cases, but for now the safest solution is to disable async compute on the affected cards. Sorry.

 

[leadeater]

1 hour ago, IGJoe2192 said:

They don't have a decreasing market share. in DX12 GPU's alone the HD7900 series itself has gone from 1.98% down to 1.69% share in the last 5 months... AMD doesn't see money for HD7900 sales anymore, and they are declining in the market. What all of you also don't realize is that while these old cards may be in systems on Steam's survey that does not mean they are using the limited number of games that leverage async compute. So in all honesty how many customers does this "hurt"? Why are you so pissed about if you do not have that hardware...

AMD doesn't have a decreasing market share and this isn't just for HD7900 series cards, tons of current cards you can buy right now use the same GCN 1.0 core so solely looking at HD7900 series cards is incorrect.

 

And not having the product is irrelevant, everyone should point out and protect consumers when there is something wrong. Idly standing by letting this sort of thing happen sets a precedent, one that my hurt you later.

[Humbug]

11 minutes ago, Shahnewaz said:

snip

Doing that without informing users is ridiculous.

Should have been stated in the driver release notes.

[IGJoe2192]

22 minutes ago, leadeater said:

AMD doesn't have a decreasing market share and this isn't just for HD7900 series cards, tons of current cards you can buy right now use the same GCN 1.0 core so solely looking at HD7900 series cards is incorrect.

 

And not having the product is irrelevant, everyone should point out and protect consumers when there is something wrong. Idly standing by letting this sort of thing happen sets a precedent, one that my hurt you later.

Never said that.

 

Look at the all mighty god "The Steam Hardware Survey" an overwhelming majority of GCN 1 products are in the decline.

 

"idly" sitting by is better than crying wolf as soon as you see the most minute disruption. Sit back, chill out, get all of the information and then act. As quoted above, it looks like they disabled it for a reason and are looking at ways to correct the issues they have found.

 

 

[leadeater]

26 minutes ago, IGJoe2192 said:

"idly" sitting by is better than crying wolf as soon as you see the most minute disruption. Sit back, chill out, get all of the information and then act. As quoted above, it looks like they disabled it for a reason and are looking at ways to correct the issues they have found.

No body cried wolf, they cried "what just happened please tell us" and AMD in tern eventually did. "Sit back and chill out" would have meant nothing would have happened and why it was disabled would have never been known. I was quite sure it was disabled for a reason but that is different than actually knowing.

 

Having the featured getting disabled was not the issue, not being told was.

 

26 minutes ago, IGJoe2192 said:

Never said that.

 

3 hours ago, djdwosk97 said:

So Nvidia should just stop supporting Pascal? The 1060, 1070, and 1080 only have a 4% share on Steam.

 

2 hours ago, IGJoe2192 said:

They don't have a decreasing market share. in DX12 GPU's alone the HD7900 series itself has gone from 1.98% down to 1.69%

Not sure how to read that other than what is there in the text??

 

My only issue is you seem very aggressive in your stance that this was never an issue and no one should have tried to get answers, and that the AMD user base is too small to care anyway. If you're not effected and don't care then say your piece and move on, your entitled to your opinion but it has no more value than anyone else's.

[IGJoe2192]

47 minutes ago, leadeater said:

No body cried wolf, they cried "what just happened please tell us" and AMD in tern eventually did. "Sit back and chill out" would have meant nothing would have happened and why it was disabled would have never been known. I was quite sure it was disabled for a reason but that is different than actually knowing.

 

Having the featured getting disabled was not the issue, not being told was.

 

 

 

Not sure how to read that other than what is there in the text??

 

My only issue is you seem very aggressive in your stance that this was never an issue and no one should have tried to get answers, and that the AMD user base is too small to care anyway. If you're not effected and don't care then say your piece and move on, your entitled to your opinion but it has no more value than anyone else's.

You said "AMD doesn't have a decreasing market share."

 

I said "DX12 GPU's alone the HD7900 series itself has gone from 1.98% down to 1.69%" (talking specifically about 7900 series under the DX12 GPU category, not AMD in its entirety. This is in reference to the Steam Hardware Survey) "They don't have a decreasing market share" is in reference to a person asking if Nvidia should drop support on the Pascal lineup due to their current market share value, which is still growing and making them money.

 

What part of "Sit back, chill out, get all of the information and then act" says to do nothing? I never said it was a non-issue, just that it isn't something worth writing home about. People are making a mountain out of a mole hill. AMD turned off a feature for the greater good. This is so they could come up with a fix before people tried to use it, had real issues and started bashing them as they did for turning it off. I am one for getting answers but not for bashing a company until I know the reason they have done something.

 

This isn't something that has effected a majority of people and has not ruined their gaming experience. This is a small feature that is relevant to a dwindling number of specific hardware users. Therefore it is no big deal, not some marketing scheme that is going to ruin the industry for consumers.

 

One last bit and that has to do with your perception of my aggressiveness. That is on you for assuming ones tone based on text and is an internal reflection on how you feel.

[leadeater]

41 minutes ago, IGJoe2192 said:

I never said it was a non-issue, just that it isn't something worth writing home about. People are making a mountain out of a mole hill. AMD turned off a feature for the greater good. This is so they could come up with a fix before people tried to use it, had real issues and started bashing them as they did for turning it off. I am one for getting answers but not for bashing a company until I know the reason they have done something.

 

This isn't something that has effected a majority of people and has not ruined their gaming experience. This is a small feature that is relevant to a dwindling number of specific hardware users. Therefore it is no big deal, not some marketing scheme that is going to ruin the industry for consumers.

 

One last bit and that has to do with your perception of my aggressiveness. That is on you for assuming ones tone based on text and is an internal reflection on how you feel.

True only reading text is a wicked thing for interpretation but you clearly said a potential 1.1 million people is not enough to care about, so context here is key. With your responses there really isn't any other way to interpret your intentions other than overly hostile. Can you yourself say you weren't being overly dismissive of the issue, it's one thing to say for me to reflect on this but maybe you should think about why not just me responded to you with the very same issue.

 

If no one "wrote home about it" this information would have never been disclosed so it's a bit disingenuous to say anything to the contrary. You can't get answers without asking for them so what was supposed to happen, we continue to say nothing after months of it being disabled and never been told. It was luck and inquisitiveness of users that found out that the feature was disabled, not from any official communication from AMD till now. Maybe the chain of events isn't that clear.

 

I will completely agree the impact was small and the list of games that used the feature on the effected hardware was even smaller but again that wasn't the issue. Also we have to remember async compute was never originally enabled on GCN 1.0 it came later with API and driver support which was documented so it's not really that much to ask for to do the same when taking it away again.

[Prysin]

15 hours ago, leadeater said:

AMD doesn't have a decreasing market share and this isn't just for HD7900 series cards, tons of current cards you can buy right now use the same GCN 1.0 core so solely looking at HD7900 series cards is incorrect.

 

And not having the product is irrelevant, everyone should point out and protect consumers when there is something wrong. Idly standing by letting this sort of thing happen sets a precedent, one that my hurt you later.

Tons of current cars

 

GCN 1.0 cards

HD 7850 Pitcairn -> EOL

HD 7870 Pitcairn -> EOL

HD 7950 Tahiti -> EOL

HD 7970 Tahiti -> EOL

HD 7970GHz Tahiti -> EOL

HD 7990 Tahiti -> EOL

R9 265 Pitcairn -> EOL

R9 270 Pitcairn -> EOL

R9 270X Pitcairn -> EOL

R9 280 Tahiti -> may find some still in retail channels

R9 280X Tahiti -> may find some still in retail channels

R9 370 Pitcairn -> may find some still in retail channels

XBOX 1 Pitcairn

PS4 "OG" Pitcairn 

 

Yes, tons of them....

 

In comparison

GCN 2, 3 and 4 consists of

R9 285 Tonga

R9 380 Antigua

R9 380X Antigua

R9 290 Hawaii

R9 290X Hawaii

R9 295x2 Hawaii/Vesuvius

R9 390 Grenada

R9 390X Grenada

R9 390x2 (dual 390, powercolor only) Grenada

Fury Fiji

Fury X Fiji

Nano Fiji

RX 460 Polaris 11

RX 470 Polaris 10

RX 480 Polaris 10

ALL Kaveri iGPUs  Hawaii

ALL Carrizo iGPUs  Tonga

ALL Bristol Ridge iGPUs  Tonga

PS4 Pro Polaris 10

 

 

GCN1 has just been recycled ALOT

[xAcid9]

15 hours ago, Shahnewaz said:

http://gpuopen.com/concurrent-execution-asynchronous-queues/#comment-6937

 

Wait.. isn't that Alessio from BitsAndChips? They jumped into conclusion too early previously? lel

On 12/8/2016 at 10:18 PM, Notional said:

 

 

15 hours ago, Humbug said:

Doing that without informing users is ridiculous.

Should have been stated in the driver release notes.

Indeed.

 

[leadeater]

4 hours ago, Prysin said:

-snip-

Yea true a lot of them are old but they are actually still being sold and a lot of them are still in pre-builds and laptops, plus I wouldn't really say EOL if they are still being supported in the current driver branch. My old 6970 is EOL, so is my 4850 etc but being end of sale doesn't mean end of life. Anyway now we are just nit picking and still dancing round the actual point and just arguing about how many cards are out their when that isn't the point.

 

Here's a question, would it have been ok for AMD to have disabled the feature on any of the later GCN architectures and not have said so. If for any one of them the answer is no that is not ok then it's not for any of them. Supported is supported period, it doesn't matter how old something is or how few there are or if you can still buy the product. If AMD had actually EOL'd all these cards a couple of months ago that would have actually been fine since we would have known.

 

Quote

AMD Desktop Product Family Compatibility
Radeon™ RX 400 Series Graphics	AMD Radeon™ R7 300 Series Graphics
AMD Radeon™ Pro Duo	AMD Radeon™ R7 200 Series Graphics
AMD Radeon™ R9 Fury Series Graphics	AMD Radeon™ R5 300 Series Graphics
AMD Radeon™ R9 Nano Series Graphics	AMD Radeon™ R5 200 Series Graphics
AMD Radeon™ R9 300 Series Graphics	AMD Radeon™ HD 8500 - HD 8900 Series Graphics
AMD Radeon™ R9 200 Series Graphics	AMD Radeon™ HD 7700 - HD 7900 Series Graphics

 

[goodtofufriday]

This thread just reminded me i had to sell my two 6990s because last yesr they went EOL and the very first crimson driver was the last one these cards would get.  

 

And i was okay with that. It was an old card on an older architecture. By that time it was still managing by sheer brute force of the hardware spec. 

 

What am i getting at?  Its okay for very old hardware to gain n then lose a feature set. It's not ideal but we dont want to hold back progress. 

 

As for that user saying nvidia should not support gtx 10 series because of the small user base. Get real man, this isnt a troll discussion. A brand new product will have a smaller user base amd is expected to grow. Not the same as a 5 year old card thats being phased out and has a continually dimishing user base. And you know that,  so why even bring it to the table. 

[xAcid9]

19 minutes ago, goodtofufriday said:

This thread just reminded me i had to sell my two 6990s because last yesr they went EOL and the very first crimson driver was the last one these cards would get.  

Did AMD announce they dropping support for pre-GCN or they silently drop driver support when that happen?

[goodtofufriday]

1 minute ago, xAcid9 said:

Did AMD announce they dropping support for pre-GCN or they silently drop driver support when that happen?

They officially announced anything pre-gcn was going EOL. It was unfortunate since some not too old cards were pre-gcn rebrands

[79wjd]

1 hour ago, goodtofufriday said:

This thread just reminded me i had to sell my two 6990s because last yesr they went EOL and the very first crimson driver was the last one these cards would get.  

 

And i was okay with that. It was an old card on an older architecture. By that time it was still managing by sheer brute force of the hardware spec. 

 

What am i getting at?  Its okay for very old hardware to gain n then lose a feature set. It's not ideal but we dont want to hold back progress. 

 

As for that user saying nvidia should not support gtx 10 series because of the small user base. Get real man, this isnt a troll discussion. A brand new product will have a smaller user base amd is expected to grow. Not the same as a 5 year old card thats being phased out and has a continually dimishing user base. And you know that,  so why even bring it to the table. 

Again, there is nothing wrong with dropping support for an older generation. There is, however, a problem with removing a feature that was once present. And the reason for the comparison to the 10 series is that people were adamant about the percentage of GCN 1 cards being too low....well, that percentage is still higher than a lot of supported cards.

 

Now, apparently this is just an error in the driver (going by the AMD interview on PCPer). So if that's the case, then it's not an issue. 

[Kwee]

If someone ask, Async Compute still disabled in new driver (16.12.2)

[Blake]

Hey guys, I have a great idea, lets continue to argue about semantics and minute and insignificant details rather the the actual topic? Great idea hey /everythreadever

[Humbug]

On ‎12‎/‎14‎/‎2016 at 8:24 PM, goodtofufriday said:

 

What am i getting at?  Its okay for very old hardware to gain n then lose a feature set. It's not ideal but we dont want to hold back progress. 

it is not ok at all to lose a feature.

it is ok to lose driver support.

For example AMD can make an announcement that the next driver will be the last one which supports GCN 1.0 GPUs. Then the newer drivers will not be compatible with GCN 1.0 hardware. That is ok.

 

what you DO NOT do is silently turn off features via a driver update. That's what has happened here. And btw GCN 1.0 is still actively supported by AMD, they did not announce end of life. And if they did you will not lose features or have your GPU nerfed, you will simply lose the ability to install/use future drivers on your hardware. It will not make your card slower on any existing workload.

[Prysin]

20 minutes ago, Humbug said:

it is not ok at all to lose a feature.

it is ok to lose driver support.

For example AMD can make an announcement that the next driver will be the last one which supports GCN 1.0 GPUs. Then the newer drivers will not be compatible with GCN 1.0 hardware. That is ok.

 

what you DO NOT do is silently turn off features via a driver update. That's what has happened here. And btw GCN 1.0 is still actively supported by AMD, they did not announce end of life. And if they did you will not lose features or have your GPU nerfed, you will simply lose the ability to install/use future drivers on your hardware. It will not make your card slower on any existing workload.

let us go out on a limb here and assume there is a reason for AS/AC not working. Let us say there has been a updated method in how it is used, perhaps said method provides benefit to Nvidia users? I don't know, and OP has no clue at all. So let us assume there is a reason.

 

Would you say it is NOT OK, to disable a feature for a very small family of GPUs, if the execution of said feature is holding back development of said feature globally?

 

I dont. If there has been a underlying change to DX12 AS/AC execution in order to improve global performance, and said change is not playing ball with older GCN cards, then fuck it. Disable it for those cards until you can fix it, or dont fix it.

 

GCN1.0 is a 4.5 year old architecture. The only customers i have any pity for is the fools who bought a R9 370 and R7 360. However said fools should have bought a GTX 950 or 750Ti, as those cards were clearly faster in 9/10 tests and still is. 

[Blake]

27 minutes ago, Humbug said:

it is not ok at all to lose a feature.

it is ok to lose driver support.

For example AMD can make an announcement that the next driver will be the last one which supports GCN 1.0 GPUs. Then the newer drivers will not be compatible with GCN 1.0 hardware. That is ok.

 

what you DO NOT do is silently turn off features via a driver update. That's what has happened here. And btw GCN 1.0 is still actively supported by AMD, they did not announce end of life. And if they did you will not lose features or have your GPU nerfed, you will simply lose the ability to install/use future drivers on your hardware. It will not make your card slower on any existing workload.

Whole lot of holes in your arguments here:

 

Async compute under dx11 hasn't been touched (and never was by anyone, don't thing a single dev even looked at it) GCN 1.0 was dx11 (wasn't even sold as dx12). This feature that was promised at time of purchase is still enabled. dx12 Async, which wasn't promised at time of purchase as added on later.

Driver support hasn't been dropped.

 

FYI alot of people have the opposite reaction. a good example is every single java exploit and OSX, Apple silently kills That version of java and not a single person complains, "it's good because people are stupid", look they know their customer base. This is exactly the same as the 'amd/console/nvidia gamer" customer base fucking mouth breathers the lot of em, but they think they are smart... thats the difference. AMD comes out as says it breaks enough to be a concern, we are looking at it, but have disabled it in the next driver till it can be looked at and while this happens everyone looks at the Stephan Hodes quote and assumes it is official, forcing him to go back to the post and say "not acting as an amd official here" then later "the driver team already said they are looking into it guys".

 

Seriously, it is just a cluster fuck and there are whole in both side of the arguments

 

 

[leadeater]

8 minutes ago, Prysin said:

Would you say it is NOT OK, to disable a feature for a very small family of GPUs, if the execution of said feature is holding back development of said feature globally?

My vote would have been to End of Life all GCN 1.0 card instantly if this was the case. If an architecture is hindering future progression then that pretty much is the definition of End of Life. Release a final legacy support driver that allows turning async compute off/on and then let game profiles take care of it to address any bugs, so it can stay globally on for all games with the ability to disable it.

 

It was a very poor choice on AMD's behalf to use any GCN 1.0 architecture in the R7/R9 300 lineup, locking out stream processors from a GCN 1.1(2.0, hell what ever) for these models would have been less stupid.