Feat/intel support 11056799649423722682

[DealsBeam]

No description provided.

[try228]

Now it looks like that:

Almost done. But please note that Intel iGPUs use Intel Ring Bus(put it in bus interface).

[try228]

Just a few more things, like displaying logo or other checks.

[try228]

Please fix the "," instead of "." in pixel and texture filtrate speed - it catches the eye immediately. On my AMD card it is "." consistently.
Additionally, could you populate the remaining "N/A" fields with the actual data? For instance, for Transistors, BIOS Version, and other Memory fields, you could either mirror the "System Shared" value or implement a more robust detection for RAM frequency. Also, regarding Resizable BAR, please implement a check for integrated graphics; since these lack a dedicated PCIe bus, ReBAR support is technically not applicable, and it should reflect that accordingly.

Finally, could you add support for Intel OneAPI (and OneAPI Level Zero)? As it is a direct competitor to NVIDIA CUDA and AMD ROCm/HIP, it would be a very useful addition. Also, please specify which packages from the Arch repository are required for OneAPI so i could test it.

[try228]

OneAPI UI: The checkbox is correctly marked as checked, but the "oneAPI" label is greyed out on my AMD GPU. This visual inconsistency is confusing because the greyed-out style typically implies that the option is not available on this GPU. Could you align the label's style with the checkbox's active state?
Bus Interface: I've noticed an intermittent issue where the Bus Interface field occasionally defaults to "N/A" instead of displaying the actual data.

Missing Data: The following fields remain unpopulated ("N/A"): Memory specifications, Transistor count, and BIOS version.
Wrong lookup link: The "Lookup" button directs to the TechPowerUp page for the Intel UHD P750 (Xeon graphics) instead of the Intel UHD Graphics 750. It seems the link generation logic is using a loose string match and picking up the wrong model.

Other implemented features look correct. Great work on these improvements.

P.S. I also forgot to mention SYCL, which serves as an analogue to HIP. Please consider adding support for it as well.

[lseurttyuu]

Hey @try228, thanks for stepping in to test this on actual Intel silicon!

By the way, the oneAPI addition is justified, however the current shape of it in the UI has to change. I believe for Intel GPUs, oneAPI and SYCL check-marks should display INSTEAD of ROCm + HIP. (Not sure about SYCL, I'll have to read more about it). Even 5 things is too many, not mentioning 6.

This way CUDA is displayed always (as a long-time industry standard), ROCm + HIP displays for NV and AMD GPUs, oneAPI + SYCL displays only for Intel GPUs.

Or even this way:
a) Nvidia GPU: CUDA / ROCm / oneAPI
b) AMD GPU: CUDA / ROCm / HIP
c) Intel GPU: CUDA / oneAPI / SYCL

Or abandoning HIP & SYCL for the sake of simplicity, eg. just leave it CUDA / ROCm / oneAPI for all GPUs. I don't know yet, I'll decide later on this, when Intel support is at a later stage.

@DealsBeam (or whoever is operating the agent), once @try228 confirms the readouts are accurate and stable, let me know and I will do a full architectural review. Please ensure that the Intel sensor polling doesn't block the UI thread (it needs to follow the same async Task pattern established in the NVIDIA update) and that it properly integrates with the MVVM factory pattern.

[try228]

It looks like there has been a regression with the latest changes. Several fields that were previously working correctly have reverted to "N/A" or changed unexpectedly:
Fillrates: Pixel and Texture Fillrate have reverted to "N/A".
Logo: Logo now reverted to standard for unknown GPUs.
No more oneAPI tick: The oneAPI check now gone too, showing nothing.

Could you please check what caused these fields to break? It seems like some of the previous logic might have been overwritten or accidentally removed.

By the way, the oneAPI addition is justified, however the current shape of it in the UI has to change. I believe for Intel GPUs, oneAPI and SYCL check-marks should display INSTEAD of ROCm + HIP. (Not sure about SYCL, I'll have to read more about it). Even 5 things is too many, not mentioning 6.

This way CUDA is displayed always (as a long-time industry standard), ROCm + HIP displays for NV and AMD GPUs, oneAPI + SYCL displays only for Intel GPUs.

Or even this way: a) Nvidia GPU: CUDA / ROCm / oneAPI b) AMD GPU: CUDA / ROCm / HIP c) Intel GPU: CUDA / oneAPI / SYCL

@lseurttyuu Regarding the oneAPI/AMD support point: just a quick technical clarification. AMD technically does support oneAPI, but it's not a "native" feature—it relies on specific runtimes and acts as a translation layer on top of ROCm (requires Codeplay, HIP). In fact, ROCm/ZLUDA perform a very similar "translation" role for CUDA on AMD hardware. This reinforces the argument that these are often compatibility layers rather than native implementations.

That said, I completely agree that cluttering the UI with 5-6 checkboxes is bad UX. Instead of hardcoding which GPUs see which options, maybe the best approach would be to dynamically show only the checkboxes for the compute APIs actually detected/installed on the system? This way, users with AMD would naturally see ROCm/HIP (and maybe oneAPI if they have the plugins), and Intel users would see oneAPI/SYCL, keeping the UI clean without arbitrary restrictions.

I'll continue checking the stability and accuracy of the sensor readouts as you requested.

Not sure about SYCL, I'll have to read more about it

Regarding your note on SYCL: to clarify, SYCL is indeed the functional analogue to HIP in the Intel ecosystem. It provides the same type of cross-platform abstraction that HIP provides for AMD/NVIDIA, which is why it makes sense to pair them in the UI.

[try228]

still broken. i dont even know why.