added note for old Intel hardware pre sycl

[alosslessdev]

Older hardware used opencl

[NeoZhangJianyu]

Intel GPU can be supported by several backends, including OpenCL.
But I don't like this sentence in SYCL backend doc:

l don't know why, current OpenCL backend narrow down it's scope from support multiple GPUs to only Qualcomm Adreno GPU. It's bad news.
As OpenCL.md, only list 3 Qualcomm Adreno GPUs are verified.
If you promise to support Intel GPU, please list them in your supported/verified hardware list.
But in the latest OpenCL.md, no any promise to support Intel GPU.
Based on above info, I don't think ask user to try OpenCL backend is good suggestion to Intel GPU users.

Every backend can promote itself in it's guide/doc.
Do not cross the line with your advertising! :)

Thank you!

[savvadesogle]

@NeoZhangJianyu
Hi, i tested some models with OpenCL backend:

C:\dev\llama_opencl\llama.cpp\build\bin\Release>llama-bench -m T:\models\TheBloke\Llama-2-7B-GGUF\llama-2-7b.Q4_0.gguf -ngl 100 -fa 0,1
ggml_opencl: selected platform: 'Intel(R) OpenCL Graphics'

ggml_opencl: device: 'Intel(R) Arc(TM) A770 Graphics (OpenCL 3.0 NEO )'
ggml_opencl: OpenCL driver: 32.0.101.8136
ggml_opencl: vector subgroup broadcast support: false
ggml_opencl: device FP16 support: true
ggml_opencl: mem base addr align: 128
ggml_opencl: max mem alloc size: 4095 MB
ggml_opencl: device max workgroup size: 1024
ggml_opencl: SVM coarse grain buffer support: true
ggml_opencl: SVM fine grain buffer support: false
ggml_opencl: SVM fine grain system support: false
ggml_opencl: SVM atomics support: false
ggml_opencl: flattening quantized weights representation as struct of arrays (GGML_OPENCL_SOA_Q)
ggml_opencl: loading OpenCL kernels....................................................................
ggml_opencl: default device: 'Intel(R) Arc(TM) A770 Graphics (OpenCL 3.0 NEO )'

model	size	params	backend	ngl	fa	test	t/s
llama 7B Q4_0	3.56 GiB	6.74 B	OpenCL	100	0	pp512	260.32 + 0.91
llama 7B Q4_0	3.56 GiB	6.74 B	OpenCL	100	0	tg128	33.67 + 0.10
llama 7B Q4_0	3.56 GiB	6.74 B	OpenCL	100	1	pp512	179.66 + 0.43
llama 7B Q4_0	3.56 GiB	6.74 B	OpenCL	100	1	tg128	31.89 + 0.48

build: 4a4f7e6 (7409)

C:\dev\llama_opencl\llama.cpp\build\bin\Release>llama-bench -m T:\models\lmstudio-community\gpt-oss-20b-GGUF\gpt-oss-20b-MXFP4.gguf -ngl 100 -fa 0,1
ggml_opencl: selected platform: 'Intel(R) OpenCL Graphics'

ggml_opencl: device: 'Intel(R) Arc(TM) A770 Graphics (OpenCL 3.0 NEO )'
ggml_opencl: OpenCL driver: 32.0.101.8136
ggml_opencl: vector subgroup broadcast support: false
ggml_opencl: device FP16 support: true
ggml_opencl: mem base addr align: 128
ggml_opencl: max mem alloc size: 4095 MB
ggml_opencl: device max workgroup size: 1024
ggml_opencl: SVM coarse grain buffer support: true
ggml_opencl: SVM fine grain buffer support: false
ggml_opencl: SVM fine grain system support: false
ggml_opencl: SVM atomics support: false
ggml_opencl: flattening quantized weights representation as struct of arrays (GGML_OPENCL_SOA_Q)
ggml_opencl: loading OpenCL kernels....................................................................
ggml_opencl: default device: 'Intel(R) Arc(TM) A770 Graphics (OpenCL 3.0 NEO )'

model	size	params	backend	ngl	fa	test	t/s
gpt-oss 20B MXFP4 MoE	11.27 GiB	20.91 B	OpenCL	100	0	pp512	293.56 + 0.56
gpt-oss 20B MXFP4 MoE	11.27 GiB	20.91 B	OpenCL	100	0	tg128	41.14 + 0.34
gpt-oss 20B MXFP4 MoE	11.27 GiB	20.91 B	OpenCL	100	1	pp512	289.83 + 0.45
gpt-oss 20B MXFP4 MoE	11.27 GiB	20.91 B	OpenCL	100	1	tg128	40.53 + 0.18

build: 4a4f7e6 (7409)

[NeoZhangJianyu]

Yes, I see.

I wonder why OpenCL.md mention to focus on Qualcomm Adreno GPU.
Will OpenCL backend include Intel GPU in it's scope?

Thank you!

[NeoZhangJianyu]

I think it provide another choice to Intel GPU users.

Thank you!