2018 Hardware Thread

Nvidia makes GPU overclocking a lot smarter with “Scanner� | Ars Technica

Dude said:

Nvidia makes GPU overclocking a lot smarter with “Scanner� | Ars Technica

Unwinder mentioned that awhile ago: Latest MSI Afterburner Betas Updates - Page 13 - Windows 10 Forums

After the last NDA & drivers are dropped, he'll post the newest Afterburner beta.

Also:

I spent whole weekend playing with NVIDIA Scanner integration in MSI Afterburner so I have some interesting observations to share with you. All the facts I’ll post are rather generic things (still useful for reviewers), which don’t seem to be covered by NDA, so I hope that green rays of love from California won’t burn me alive

- NVIDIA’s Scanner API is purely 64-bit only. Which means that in order to support it we would also either need to start releasing both 32-bit and 64-bit branches of MSI Afterburner or make MSI Afterburner pure 64-bit application only (and this way drop support for 32-bit OS’es and leave huge part of user base without updates, which makes me absolutely unhappy because I do my all to keep backward compatibility and still provide support for ancient Window XP OS and 10 years old GPUs starting from GeForce 6 or newer and RADEON HD 2000). As a compromise solution, our implementation is an independent 64-bit MSI OC Scanner application bundled with MSI Afterburner and communicating with it via MACM shared memory interface. Such modular architecture will allow us to update scanner independently of MSI Afterburner in future as well as let us to provide OC scanner component’s source code to GPU vendors for their internal tests.
- It was declared by NV that scanner API is using embedded NVIDIA tests for stressing GPU and testing overclocking stability. It is closed proprietary technology, so it is not clear now how much hardware dependent is the implementation, it is quite possible that it will work exclusively on RTX 20x0 series only.
- OC Scanner will provide you two main modes: test and scan. In test mode your manually configured voltage/frequency curve is being stress-tested, result is returned as GPU stability confidence level in % (0% - unstable, 100% - stable). I’ve never seen 100% result even on stock clocks BTW, it always seem to max on 90% on all cards we tried so probably it is just a precaution from NV side, no need to worry if you see <100% result. In scan mode scanner’s proprietary algorithms are trying to detect the maximum stable overclocking achieved by your GPU, result is returned as average core clock overclocking in MHz. Modified overclocked non-linear voltage/frequency curve is also sent to MSI Afterburner by scanner in the end of scanning process. Test mode takes approximately 5 minutes to complete, scan mode takes 15-20 minutes to complete.
- It is very convenient to keep voltage/frequency curve editor window open while the scanner is running. V/F curve editor is showing you current GPU clock and voltage levels with horizontal/vertical dotted lines directly on the curve in real time (2070MHz / 975mV on the previous screenshot). This way you can visually control the process of scanning, so you always know which point of voltage/frequency curve is currently being tested or overclocked. In scan mode it doesn’t try to overclock each point independently, as it would take much more time than just 20 minutes. Instead of doing so, it is detecting working range of voltages on the first stage, split this voltage range on a few fixed parts (4 key points on the screenshot above) then try to find the maximum possible overclock for each of those key points by fixing the voltage in each point and slowly increasing the clock until instability is detected. You can see this process on the previous screenshot as well on jigsaw core clock graph. Each tooth of this jigsaw is the process of finding the maximum possible clock for each of 4 key points. Once the maximum stable overclocking is detected in each key point, the scanner is building whole curve by linear interpolation between the points and sending result to MSI Afterburner.

RTSS 6.7.0 beta 1 | Page 67 | guru3D Forums

FINAL FANTASY XV WINDOWS EDITION | SQUARE ENIX

It looks like first benchmarks might be showing up.
Kicks ass when using just rasterization it looks like.

Probably better for benchmarks that use the faster(14 Gbps) 11GB GDDR6 memory.

Cliff S said:

It looks like first benchmarks might be showing up.
Kicks ass when using just rasterization it looks like.

Probably better for benchmarks that use the faster(14 Gbps) 11GB GDDR6 memory.

Not only that but the whole shader pipeline is completely different. It's way more efficient, even with less cores. All leaks i've seen so far are exactly on par with what my expectations were around the time the cards were announced.

Once DLSS gets proper 3D-engine support the results will be impressive. The RTX2080Ti is a true 4K number crunching monster. I would not be surprised if nVidia launched one more model that is even more powerful. There's still room for a few more shaders and more VRAM.

Rog Ryujin 360 full secs are finally available at ASUS. Nothing new about a release date or price.

ROG RYUJIN 360 | Cooling | ASUS USA

Fan Air Flow: 121.8 CFM
Fan Noise: 29.7 dB(A)

Embedded Fan Air Flow: 19.41 CFM
Embedded Fan Noise: 31 dB(A)