News: 0001517459

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

AMD's GPUOpen HIP RT 2.5 Released With Fixes, GFX1200 RDNA4 Support

([Radeon] 2 Hours Ago GPUOpen HIPRT 2.5)


AMD's GPUOpen team today released HIP RT 2.5 as the newest version of this ray-tracing library for HIP. This library in turn is used by the likes of the Blender 3D modeling software for ray-tracing acceleration on Radeon GPUs.

For today's HIP RT 2.5 release the official change-log just mentions some minor fixes/improvements:

2.5.cfa5e2a

- More flexible vector types

- Unifying math into a single header

- Collapse crash fix

- Other minor fixes

The more flexible vector types likely being the most useful change of HIP RT 2.5 besides the bug fixing.

When I was going through the HIP RT code changes from v2.4 to v2.5, there was one other item that stood out:

HIP RT in the prior release also recognized GFX1201 as an RDNA4 GPU that is supported by this ray-tracing library, but new to HIP RT 2.5 is also recognizing the GFX1200 ID. Right now though it's not clear what GPUs correlate to GFX1200 or GFX1201 in the RDNA4 family, which as of yesterday was [1]announced with the Radeon RX 9070 and RX 9070 XT for debuting in Q1 while additional RDNA4 graphics processors are expected later.

In any event for developers interested in the AMD HIP RT 2.5 library it can be downloaded from [2]GitHub .



[1] https://www.phoronix.com/review/amd-ryzen-9950x3d-rdna4-ai-max

[2] https://github.com/GPUOpen-LibrariesAndSDKs/HIPRT/releases/tag/2.5.cfa5e2a.2


phoronix

If you ever want to have a lot of fun, I recommend that you go off and program
an imbedded system. The salient characteristic of an imbedded system is that
it cannot be allowed to get into a state from which only direct intervention
will suffice to remove it. An imbedded system can't permanently trust
anything it hears from the outside world. It must sniff around, adapt,
consider, sniff around, and adapt again. I'm not talking about ordinary
modular programming carefulness here. No. Programming an imbedded system
calls for undiluted raging maniacal paranoia. For example, our ethernet front
ends need to know what network number they are on so that they can address and
route PUPs properly. How do you find out what your network number is? Easy,
you ask a gateway. Gateways are required by definition to know their correct
network numbers. Once you've got your network number, you start using it and
before you can blink you've got it wired into fifteen different sockets spread
all over creation. Now what happens when the panic-stricken operator realizes
he was running the wrong version of the gateway which was giving out the wrong
network number? Never supposed to happen. Tough. Supposing that your
software discovers that the gateway is now giving out a different network
number than before, what's it supposed to do about it? This is not discussed
in the protocol document. Never supposed to happen. Tough. I think you get
my drift.