GPU-Z 2.72.0 Released: Adds Recognition of NVIDIA RTX Spark, RTX 4090 Ti ES, and New AMD Multi-GPU Hot Spot Temperature Monitoring
IT之家 November 20 news: TechPowerUp today released the GPU-Z 2.72.0 version update, which adds support for the NVIDIA RTX Spark processor, the GB203-based GeForce RTX 5070, and the unreleased GeForce RTX 4090 Ti engineering sample, while also expanding recognition capabilities for multiple AMD and Intel graphics products.

On the NVIDIA product side, GPU-Z 2.72.0 adds support for RTX Spark, but some memory information still cannot be read for now. In addition, the software adds support for the GB203-core GeForce RTX 5070 and the GeForce RTX 4090 Ti engineering sample (ES).
On the AMD side, the new version adds support for the BC-250 mining card and adds clock frequency and temperature monitoring for some AMD integrated GPUs. At the same time, the software can now read GPU hot spot temperatures for AMD R9 200/300, Fury, and RX 400/500 series graphics cards, covering GPU architectures including Polaris, Lexa, Hawaii, Fiji, Bonaire, and Tonga.
On the Intel side, GPU-Z 2.72.0 adds support for the Raptor Lake-H Refresh platform and Iris Pro P555 (E3-1558L v5).

Sensor functionality is also one of the focuses of this update. When users hover the mouse over sensor values, they can view related statistical information through newly added tooltips. In addition, the minimum, maximum, and average calculations on the Sensors page now automatically exclude invalid readings, preventing abnormal data from affecting statistical results.
For devices equipped with integrated graphics, the new version adjusts the way memory is displayed, no longer showing "N/A" but instead labeling the capacity of shared system memory as iGPU, for example "iGPU (64 GB)". On the Advanced information page, the software also adds display support for unknown NVIDIA power rails, with the relevant data presented in the form of "Rail ### power draw"; this feature is disabled by default.
For graphics cards whose GPU model cannot be recognized (IT之家 note: including some counterfeit graphics cards), GPU-Z can now read the DirectX API version and feature level support directly from the current system, instead of uniformly displaying "Unknown".
Other improvements include: adding CUDA capability information on the Advanced information page, adding capability recognition for CUDA 6.2, 7.x, 8.x, 9.0, 10.x, and 12.x and multiple CUDA properties; expanding the DXVA GUID names for AMD and NVIDIA graphics cards, and fixing the 8K video support detection issue.
In addition, the new version fixes the issue of OpenGL detection failing in some cases, adjusts the names of the hot spot temperature sensors for AMD and NVIDIA graphics cards to uniformly display as "Hot Spot"; if the NVIDIA BIOS cannot be read at all, the Advanced information page no longer displays a corrupted BIOS information block. GPU-Z 2.72.0 also fixes multiple spelling and detail errors on the advanced CUDA information page and makes other minor improvements.
- "Suspected NVIDIA GeForce RTX 4090 Ti graphics card 'non-FE version' engineering sample appears online"
- "Giant: NVIDIA RTX 4090 Ti graphics card exposed again, four-slot design"
- "NVIDIA reportedly to launch GeForce RTX 5070 graphics card equipped with GB203 GPU"
Related Articles
- AI inference MLPerf 6.1 released: AMD sets a new record with 512 MI355X, NVIDIA Vera Rubin appears for the first time
- About 5~12% gaming performance improvement: Resizable BAR option to enter NVIDIA NVIDIA App
- AMD executive: A key advantage of the company over NVIDIA is being more open
- Performance advantage higher than expected: AMD says Venice processors are about 20% faster than NVIDIA Vera
- AMD is about to launch a new generation of AI infrastructure, confronting NVIDIA head-on
- Morgan Stanley estimates AMD EPYC Venice shipments of 6.75 million units in 2027, 17% more than NVIDIA Vera processors