Gallium3D LLVMpipe Benchmarks From Intel Haswell
Being published this afternoon are benchmarks of the Gallium3D LLVMpipe software driver compared to Intel HD 4600 graphics on Mesa 9.2 Git when using an Intel Core i7 4770K. While this Intel "Haswell" CPU is faster than previous generations, it's still obviously best not relying upon LLVMpipe.
LLVMpipe is used commonly now as a software fallback in cases where there is no physical GPU / hardware driver available or working. LLVMpipe is also used by graphics driver developers as a separate code-path when working on Gallium3D or validating regressions/problems along a vendor-neutral driver.
With the Core i7 4770K CPU having HD 4600 graphics, obviously that's the best to use, but for kicks we decided to run some LLVMpipe benchmarks on the CPU.
The testing is quite straightforward and the results in full can be found in 1306102-UT-INTELHASW34.
To no real surprise, using the Haswell HD 4600 graphics is many times faster than using the LLVMpipe driver on the Intel i7 4770K CPU.
At least LLVMpipe tends to be fast enough on modern processors for handling a composited Linux desktop in fallback scenarios...
Now view the rest on OpenBenchmarking.org.
LLVMpipe is used commonly now as a software fallback in cases where there is no physical GPU / hardware driver available or working. LLVMpipe is also used by graphics driver developers as a separate code-path when working on Gallium3D or validating regressions/problems along a vendor-neutral driver.
With the Core i7 4770K CPU having HD 4600 graphics, obviously that's the best to use, but for kicks we decided to run some LLVMpipe benchmarks on the CPU.
The testing is quite straightforward and the results in full can be found in 1306102-UT-INTELHASW34.
To no real surprise, using the Haswell HD 4600 graphics is many times faster than using the LLVMpipe driver on the Intel i7 4770K CPU.
At least LLVMpipe tends to be fast enough on modern processors for handling a composited Linux desktop in fallback scenarios...
Now view the rest on OpenBenchmarking.org.
12 Comments