Orochi

Orochi is a library loading HIP and CUDA APIs dynamically, allowing the user to switch APIs at runtime. Therefore you don't need to compile two separate implementations for each API. This allows you to compile and maintain a single binary that can run on both AMD and NVIDIA GPUs. Unlike HIP, which uses hipamd or CUDA at compile-time, Orochi will dynamically load the corresponding HIP/CUDA shared libraries depending on your platform. In other words, it combines the functionality offered by HIPEW and CUEW into a single library

---

Video: GPU Performance Portability Using Standard C++ with SYCL

---

Intel opened it's cpu opencl runtime

---

IWOCL 2023 Program

---

HypSycl/OpenSycl has renamed as AdaptiveCpp

AdaptiveCpp is the independent, community-driven modern platform for C++-based heterogeneous programming models targeting CPUs and GPUs from all major vendors. AdaptiveCpp lets applications adapt themselves to all the hardware found in the system. This includes use cases where a single binary needs to be able to target all supported hardware, or utilize hardware from different vendors simultaneously.

It currently supports the following programming models:

SYCL: At its core is a SYCL implementation that supports many use cases and approaches of implementing SYCL.
C++ standard parallelism: Additionally, AdaptiveCpp features experimental support for offloading C++ algorithms from the parallel STL. See here for details on which algorithms can be offloaded. AdaptiveCpp is currently the only solution that can offload C++ standard parallelism constructs to GPUs from Intel, NVIDIA and AMD -- even from a single binary.

AdaptiveCpp has been repeatedly shown to deliver very competitive performance compared to other SYCL implementations or proprietary solutions like CUDA.

---

AdaptiveCpp has been repeatedly shown to deliver very competitive performance compared to other SYCL implementations or proprietary solutions like CUDA.

The new version increases performance!
And...

No targets specification needed anymore! AdaptiveCpp now by default compiles with --acpp-targets=generic. This means that a simple compiler invocation such as acpp -o test -O3 test.cpp will create a binary that can run on Intel, NVIDIA and AMD GPUs. AdaptiveCpp 24.02 is the world's only SYCL compiler that does not require specifying compilation targets to generate a binary that can run "everywhere".

---

EuroHack24

EuroHack is a unique opportunity for current or prospective users groups of large hybrid CPU-GPU systems to either (1) port their (potentially) scalable application to GPU accelerators, (2) optimize an existing GPU-enabled application, on a state-of-the-art GPU system, or (3) optimize for the multicore. Focus should be in any case the parallelism of the application. The goal is that the development teams leave at the end of the week with applications executing faster, or at least with a clear roadmap of how to get there.

---

Not a good day for the "openess" of possibility to code on gpu: Zluda taken down

---

Two good news:
Zluda is back online
Amd is trying to run C/C++ code DIRECTLY on GPU

---

Nvidia 5090 seems a little monster in gpgpu (cost and consumption are not for everyone, by the way)
It's a pity that the works on gpu's app seems blocked...

---

Nvidia 5090 seems a little monster in gpgpu (cost and consumption are not for everyone, by the way)

Yeah, it's a compute/productivity/AI/ML card pure & simple.
You can use it for gaming (and that's what the x090 series of cards used to be for), but it's gaming performance isn't all that much better than the previous RTX 4090.
However, for compute work- it's way ahead. For some types of compute work, it's way, way, way ahead (and for the price they're charging for it, it would have to be to make it worthwhile for people to buy it).



With optimisation to take advantage of the RTX 5090s new & improved capabilities, it will be even more impressive.

---

Grant
Darwin NT

However, for compute work- it's way ahead. For some types of compute work, it's way, way, way ahead (and for the price they're charging for it, it would have to be to make it worthwhile for people to buy it).

2000$ is NOT cheap.
But it's less (much less) than a H100/H200 gpu accelerator

---