When it comes to Intel, how they play this will be interesting. Software Solution will be A LOT slower at SSE Instructions. Potentially making their New CPUs worse for Legacy Gaming. What’s more the Industry typically doesn’t support new approaches quickly.
Intel is looking into a new x86 uArch
Compute, for example, took literally a DECADE before it started being used over Shaders. Even then the only reason it’s starting to be heavily used now is because both Vulkan and DirectX 12 essentially force this support. In many cases without the Developers even realizing it.
Even 2016 Games are still primarily using SSE Instructions. So this change in Intel if they merely drop support could be as monumental a shift as Multi-Core and/ or 64-Bit. Which again it’s taken about a decade to be properly supported. Not 6 years as Intel is likely planning on here.
The other option is to mimic AMDs approach of Virtualised Architecture. This, however, means overall performance would drop on everything. And AMD is VERY Far ahead in maturing this approach. Ryzen finally closing the performance gap, showcases this exceptionally well. And AMD has been at this for the past 8 years now.
Intel can probably achieve similar results. But you’ll be looking at the better part of a Decade. During which time they’ll swap places in terms of Performance.
The Final Alternative could be “Specialized” Operation Extension. As in, the Instructions that can’t easily be mapped retain specialized Silicon. So like a cut-down SSE Core. With Common Tasks, shifted to the VMX and FMA4.
This wouldn’t give them the silicon gains they’d have hoped for and could potentially have some teething problems that might affect stability. But overall they wouldn’t really lose performance or compatibility.
It would, however, make the Architecture a mess to expand. And additional changes would result in larger silicon.
Source: Tweak Town