ThisOldTech wrote on 2021-05-19, 19:32:

It does appear that I’ve got hardware acceleration in DirectX within W95

I would interpret it as having DirectX installed while having no compliant video driver.

I ran several tests and created a video showing the various software titles I tested in Windows NT and W95...
https://www.youtube.com/watch?v=xONCBNipb8k

From what I could tell -- vs the default S3 Trio64v+ or zero drivers installed... there definitely was SOME acceleration.
Tested on P1-133 & P1-200 machines

Original Quake is not an OpenGL example, you have to apply GLQuake to get the accelerated version.

In your GLInfo display it shows the renderer as 'Generic GDI'... I would have taken that to mean it's using software OpenGL.

megatron-uk wrote on 2021-05-25, 11:38:

In your GLInfo display it shows the renderer as 'Generic GDI'... I would have taken that to mean it's using software OpenGL.

I suspect you may be right about that. I googled around but I found people who said some non-standard cards detect that way and others concuring the same as you.
I don't pretend to be an expert on 3D acceleration or software - so I just don't know but you may be right.

I couldn't find any proof that Win NT or W95 actually offered any support for the Artist Graphics, but what I can say is that compared to the basic S3 TrioV64+ that was in both the Compaq and the Dell, it was definitely better than those... but I suspect AG never actually provided drivers that would unlock it's true capabilities outside of AutoCAD 🙁

I sweeped through their bigger windows driver packages again, no sign of 3d support. I will assume Artist simply gave up on that unless somebody will show otherwise. Bummer, I have Netvision 2200.

Putas wrote on 2021-05-26, 11:27:

I sweeped through their bigger windows driver packages again, no sign of 3d support. I will assume Artist simply gave up on that unless somebody will show otherwise. Bummer, I have Netvision 2200.

Yep I think your right. It's still interesting that it *SEEMS* to perform well in games with built-in APIs or even VESA supporting games. It's definitely better than any Trident or Cirrus Logic card, but if they'd put good effort into gaming drivers, I suspect the 3GA GPU would have been able to compare to an early version of the ATI Rage series.

Silly me had hopes when I first started reading about the card of finding a secret 3D accelerator nobody had heard of that was a killer retro gaming card for old games.
Not like we need more to pick from -- there's plenty floating around the market... It's just a neat story if anything. I love the idea that there was likely some rare/strange hardware out there that few have ever really played with.

I still have a WinSprint 1000i to play with - guaranteed not a 3D accelerator per-se... and definitely doubt there was gaming drivers.. but it still should do high res in Win3.11 and some VESA support.... and if one felt like it... AutoCAD implementations. Not sure I even know what type of 72pin sim it would recognize or if there's any point to adding it when it's only for CAD type storage.

This is interesting. I would be more careful about saying that there is no 3D support in the drivers available. Sadly, the only laptop I have now on hand is an M1 Mac, so I cannot easily extract the .exe files for NT4 drivers. However, the NT3.5 driver (which is just zip) seems to have 3D functions in the main video driver DLL.

HW_INIT_STATE@8_HW_DEFAULT_STATE@4_RxCreateContext@4_HW_CREATE_RC@8_HW_CONTEXT_CHECK_MODE@8_RxDeleteResource@4_RxGetInfo@4_HW_INFO_CHECK_MODE@8_HW_GLOBAL_INFO@12_HW_SURFACE_INFO@12_HW_SPAN_CAPS@12_HW_LINE_CAPS@12_HW_INTLINE_CAPS@12_HW_TRIANGLE_CAPS@12_HW_QUAD_CAPS@12_RxNullFunc@4_RxNullPrimitive@8_RxFlush@4_HW_FLUSH@4_RxReadRect@4_RxIntersectRect@12_HW_START_READ_RECT@12_HW_READ_RECT@20_HW_DONE_READ_RECT@8_RxWriteRect@4_HW_START_WRITE_RECT@12_ag_set_buff2@8_HW_WRITE_RECT@20_HW_DONE_WRITE_RECT@12_RxFillRect@4_HW_START_FILL_RECT@12_HW_FILL_RECT@16_HW_DONE_FILL_RECT@8_RxSetState@4_HW_LINE_PATTERN@12_HW_SOLID_COLOR@12_HW_MAP_COLOR@8_HW_FILL_COLOR@12_HW_FILL_Z@12_HW_PLANE_MASK@12_HW_Z_WRITE_ENABLE@12_HW_SEND_MODEREG@4_HW_ACTIVE_BUFFER@12_HW_ROP2@12_HW_CULL_MODE@12_HW_LAST_PIXEL@12_HW_Z_FUNC@12_HW_SHADE_MODE@12_HW_Z_ENABLE@12_HW_VERTEX_TYPE@12_HW_MASK_START@12_HW_BLEND_ENABLE@12_HW_ALPHA_TEST_ENABLE@12_HW_SCISSORS_ENABLE@8_HW_DITHER_ENABLE@12_HW_SCISSORS_RECT@12_HW_SPAN_TYPE@12_HW_SPAN_DIRECTION@12_HW_VERTEX_COLOR_TYPE@12_HW_SPAN_COLOR_TYPE@12_HW_TEX_MAG@12_HW_TEX_MIN@12_HW_SRC_BLEND@12_HW_DST_BLEND@12_HW_ALPHA_REF@12_HW_ALPHA_FUNC@12_HW_TEX_MAP_BLEND@12_HW_TEXTURE_PERSPECTIVE@12_HW_TEX_TRANSP_ENABLE@12_HW_TEX_TRANSP_COLOR@12_HW_DITHER_ORIGIN@12_HW_STIPPLE@12_HW_TEXTURE@12_RxSwapBuffers@4_HW_START_SWAP_BUFFERS@12_HW_SWAP_RECT@12_HW_DONE_SWAP_BUFFERS@8_RxDrawPrim@4_RxEnableBuffers@4_HW_ENABLE_BUFFERS@12_RxMapMem@4_RxQueryTextureMemory@4_HW_TEX_MEM_STATS@12_RxTextureHeap@4_HW_CREATE_TEXTURE_HEAP@8_HW_CLEAR_TEXTURE_HEAP@8_HW_FREE_TEXTURE_HEAP@4_RxAllocTexture@4_RxLoadTexture@4_HW_LOAD_TEXTURE@12_GetPtrFromHandle@8_FreeMapMemObj@4_CreateMapMemObj@8_DestroyMapMemObj@8_NewRxRc@4_HW_INIT_DEFAULT_RC@8_FreeRCObj@4_HW_DESTROY_RC@4_CreateRcObj@24_DestroyRCObj@12_HW_FREE_TEXTURE@8_DestroyRxWindow@4_DestroyRC@8_FreeTexHeapObj@4_CreateTexHeapObj@8_DestroyTexHeapObj@4_FreeTexObj@4_CreateTexObj@16_HW_ALLOC_TEXTURE@16_DestroyTexObj@4_GetScissorClip@4_GetClipLists@8_WndObjChangeProc@8_NewRxWindowTrack@8_HW_INIT_DEFAULT_WNDOBJ@4_RxPolyDrawSpan@(...)

I will take a look into NT4 drivers later. Even though OpenGL software says that this is generic GDI (like the software renderer), that does not always mean that the display driver is not using any 3D acceleration. Early 3D accelerator vendors prefered OpenGL MCD (over ICD) when writing drivers for NT4. It was easier and all functions unsupported by HW were implemented at least in the software pipeline (without any additional work). I explained this on my website: https://retro.swarm.cz/nt4-opengl-mini-client … ti-rage-ii-pro/

Anyway, to differentiate MCD from full software, you need to properly ask the OpenGL driver (PFD_GENERIC_ACCELERATED). This info is reported in my GPUbench: http://swarm.cz/gpubench/ in the gpubench.log file after any test is run.

You can try running the gpubench by starting the _All-Tests-f640low.bat. Just set the resolution first to 640x480 (if there is not enough memory for additional buffers (back, Z) OpenGL MCD driver will silently skip the 3D acceleration. Another issue can be color depth. Matrox Millennium uses OpenGL MCD in Windows NT4.0 and only certain color depths are accelerated.

If you run the tests, I would like to see the results in gpubench_output-f640low.csv (in addition to the gpubench.log). If you want to be sure, run the same benchmark in Windows 95 (where there is no 3D acceleration for sure) and you will see if the speed is different.

Very interesting card I don't remember having heard or read about in the past. Beside 3D acceleration that I hope could at least work, would be interesting to see it compared to others 2D cards of the PCI period with usual benchmark tools in some graph. 😉

I've checked the NT4 driver (both versions) and there is nothing related to 3D acceleration. Just the DirectDraw acceleration is present there. So it seems that only the NT 3.5 driver has some 3D capability of the chip leveraged to the operating system.

diagon_swarm wrote on 2022-10-11, 08:32:

I've checked the NT4 driver (both versions) and there is nothing related to 3D acceleration. Just the DirectDraw acceleration is present there. So it seems that only the NT 3.5 driver has some 3D capability of the chip leveraged to the operating system.

Greetings!
Can you please tell me if those drivers will work on this card? i can't find exact drivers for this model. By the way this is a 2mb or 4mb card?

Thx

pitchshifter wrote on 2023-11-15, 14:33:

By the way this is a 2mb or 4mb card?

Lets see...
The datasheet ( https://rocelec.widen.net/view/pdf/fugr4dndhu … IIS184961-1.pdf ) says these are DRAM: 262 144 Words × 16 Bits (this line says a word consists of 16 bits / 2 bytes)
To get the usable size per chip you have to multiply 262 144 with 2 (to get 8 bits) = 524.288 byte (where each byte is now counted as 8 bits)
You have 4 chips (524.288 byte * 4) = 2.097.152 byte = 2 mega byte