[Rumour MLID] AMD engineers were battling artifacting issues in Navi 31 just before launch reveal

[Marees]

Claim from Moores Law Is Dead:

AMD engineers were battling artifacting issues in Navi 31 just before launch reveal. The fix for this issue resulted in lower performance. The issue likely happened due to issue in GCD/MCD chip communications. (If that is the case then this bug should not impact the upcoming Navi 33 7600 XT chip as it is monoloithic)

 

[UnknownSouljer]

It's definitely an interesting video on a lot of fronts. He talks about 4 or 5 really big topics in a short period of time.

It begs the question on whether or not more can be unlocked and whether or not drivers can come in to fix the issue. And it also brings up questions for the future of chiplets and manufacturing. It is a major hurdle.

It also makes me wonder if they will decide to release another one that has a fix in hardware and call it 7900xtfx or whatever that has that additional 20% on top of the 7900xtx and whether or not it is valuable for them to do so.

 

[sfsuphysics]

It begs the question on whether or not more can be unlocked and whether or not drivers can come in to fix the issue. And it also brings up questions for the future of chiplets and manufacturing. It is a major hurdle.

IMO an even bigger question is whether old tech is going to continue to largely be ignored as he states that them trying to "fix the issue" is the reason why the amd 6k drivers never got a single update for months.

 

[UnknownSouljer]

IMO an even bigger question is whether old tech is going to continue to largely be ignored as he states that them trying to "fix the issue" is the reason why the amd 6k drivers never got a single update for months.

That was a resourcing issue. 6k series cards did get updates. They just kept getting pushed back as they were trying to fix a problem they apparently couldn’t fix.

 

[Lakados]

I see a lot of people being far more hopeful than I am about a software fix for this, I think the fix is already in place it was the mitigation.

There are a lot of inherent problems with moving as much data as AMD is around the silicon and keeping everything organized and in check with no latency, cross-talk, or interference.
I think the "fix" will come in the 8000 series, the 7000 has been delivered as is and is doing fine.

 

[UnknownSouljer]

I see a lot of people being far more hopeful than I am about a software fix for this, I think the fix is already in place it was the mitigation.

There are a lot of inherent problems with moving as much data as AMD is around the silicon and keeping everything organized and in check with no latency, cross-talk, or interference.
I think the "fix" will come in the 8000 series, the 7000 has been delivered as is and is doing fine.

It's more or less assumed that nVidia will release a "mid-generation" refresh that will include a 4090Ti or Super or whatever they want to call a bumped up version. If AMD could do a hardware fix on the 7900XTX they could release that plus a frequency bump to combat at the top end if they wanted to. At least it seems that way from over here as an arm-chair observer.

 

[Lakados]

It's more or less assumed that nVidia will release a "mid-generation" refresh that will include a 4090Ti or Super or whatever they want to call a bumped up version. If AMD could do a hardware fix on the 7900XTX they could release that plus a frequency bump to combat at the top end if they wanted to. At least it seems that way from over here as an arm-chair observer.

The biggest issue is the conflicting rumors of RDNA 3+ existing and being canceled, so back in Dec it was reported AMD was working on the RDNA 3+ refresh as the true fix to the frequency problems, but in Jan or Feb it was reported that they canceled those efforts as they weren't fixable and they would instead be working on Phoenix Point and RDNA 4. Who knows anymore, in either event, be it RTX 4000 or RX 7000 I am sitting this round out, too sketchy on all sides, I'll just watch as it unfolds.