2 ms·

I specifically address higher end GPU segment, e.g. current RTX2080/RTX2080S/RTX2080Ti. If you want/need top-tier performance in consumer GPU market, in recent

by nirv 7y ago

I specifically address higher end GPU segment, e.g. current RTX2080/RTX2080S/RTX2080Ti. If you want/need top-tier performance in consumer GPU market, in recent years AMD had generally nothing to offer in two years window. I afraid that this trend may continue with the AMD's "Big Navi" and NVIDIA's new 30-series. I hope Radeon will finally catch-up this year. AMD Radeon is an amazing choice in low/mid GPU market segment though.

aneutron 7y ago
I'm really curious about the kind of workload you're hitting that GPU with. Because from what I gather, you change your GPU every two cycles ? Do you use it for a specific professional application, or do you simply want the best performance ? Not intended to be snarky just purely curious.
- nirv 7y ago
  Ha, good question for a casual self-reflection. To be honest, I certainly do want the reasonably best performance: I find pleasure in the possession of technologies, which allow me to avoid being distracted in my work and leisure activities. But I always justify the practical outcome and amount of money I'm able to invest (e.g. RTX 2080 Ti is a no go). As for types of workload, in fact, nothing too fancy: - Gaming. I expect playing current AAA-titles on 4K@75-100fps. And, oh boy, I'm an avid Linux gamer (Valve Proton/DXVK)! This is probably the most GPU performance consuming activity in the long term; - GPGPU/ML/Hashcat/Pyrit/mining. Some ML hobby/educational projects, occasional hash-cracking, mining and other fun activities. From time to time I stumble (on Github or elsewhere) at weird, but interesting experimental CUDA/OpenCL-accelerated solutions for a variety of tasks. Hey, even my terminal emulator is GPU-accelerated nowadays[1]; - Video transcoding. Encoding 4K+ video content in the background, while keeping up with my general activity. Occasionally lending my GPU to a (professional video guy) friend for CUDA-accelerated DaVinci Resolve for A/V post-production. I certainly don't upgrade GPU every two years, more like 3-5 years for every decently improved generation. As for Nvidia, going from 16nm EV (Pascal) to (supposedly) 7nm EUV[2], might yield in up to +40-60% performance comparing to current gen Turing; so I'm pretty hyped to upgrade from the already aging GTX1080Ti with no ray-tracing. Or perhaps AMD will be finally able to pull out some ground-breaking GPU performance boost with their so-called "Big Navi", based on the new RDNA2 architecture. But I'm afraid that it will be a competitor for the current gen of Nvidia graphic cards, and not future ones. In any case, 2020 promises to be interesting for GPU/APU market. As usual, I will look at the specifications, benchmarks and price tags — and make my consumer choice. [1] https://github.com/alacritty/alacritty https://github.com/alacritty/alacritty [2] https://www.sammobile.com/news/win-samsung-nvidia-7nm-process-next-gen-gpu/ https://www.sammobile.com/news/win-samsung-nvidia-7nm-proces...

by nirv 7y ago

aneutron 7y ago

I'm really curious about the kind of workload you're hitting that GPU with. Because from what I gather, you change your GPU every two cycles ? Do you use it for a specific professional application, or do you simply want the best performance ? Not intended to be snarky just purely curious.

nirv 7y ago
Ha, good question for a casual self-reflection. To be honest, I certainly do want the reasonably best performance: I find pleasure in the possession of technologies, which allow me to avoid being distracted in my work and leisure activities. But I always justify the practical outcome and amount of money I'm able to invest (e.g. RTX 2080 Ti is a no go). As for types of workload, in fact, nothing too fancy: - Gaming. I expect playing current AAA-titles on 4K@75-100fps. And, oh boy, I'm an avid Linux gamer (Valve Proton/DXVK)! This is probably the most GPU performance consuming activity in the long term; - GPGPU/ML/Hashcat/Pyrit/mining. Some ML hobby/educational projects, occasional hash-cracking, mining and other fun activities. From time to time I stumble (on Github or elsewhere) at weird, but interesting experimental CUDA/OpenCL-accelerated solutions for a variety of tasks. Hey, even my terminal emulator is GPU-accelerated nowadays[1]; - Video transcoding. Encoding 4K+ video content in the background, while keeping up with my general activity. Occasionally lending my GPU to a (professional video guy) friend for CUDA-accelerated DaVinci Resolve for A/V post-production. I certainly don't upgrade GPU every two years, more like 3-5 years for every decently improved generation. As for Nvidia, going from 16nm EV (Pascal) to (supposedly) 7nm EUV[2], might yield in up to +40-60% performance comparing to current gen Turing; so I'm pretty hyped to upgrade from the already aging GTX1080Ti with no ray-tracing. Or perhaps AMD will be finally able to pull out some ground-breaking GPU performance boost with their so-called "Big Navi", based on the new RDNA2 architecture. But I'm afraid that it will be a competitor for the current gen of Nvidia graphic cards, and not future ones. In any case, 2020 promises to be interesting for GPU/APU market. As usual, I will look at the specifications, benchmarks and price tags — and make my consumer choice. [1] https://github.com/alacritty/alacritty https://github.com/alacritty/alacritty [2] https://www.sammobile.com/news/win-samsung-nvidia-7nm-process-next-gen-gpu/ https://www.sammobile.com/news/win-samsung-nvidia-7nm-proces...