gpu unlock win解算力失败。倒退到3.2才报了：tensor满血+ gen2*16解锁成功。
反倒不敢相信了。算力算解锁了吗?
我怎么看前后只变了个int8，0.93->26.97, 其他基本没变化。
解锁前：
----------------.------------------------------------------------------------|
| Device ID | 0 |
| Device Name | NVIDIA CMP 40HX |
| Device Vendor | NVIDIA Corporation |
| Device Driver | 616.92 (Windows) |
| OpenCL Version | OpenCL C 3.0 |
| Compute Units | 34 at 1650 MHz (2176 cores, 7.181 TFLOPs/s) |
| Memory, Cache | 8191 MB VRAM, 1088 KB global / 48 KB local |
| Buffer Limits | 2047 MB global, 64 KB constant |
|----------------'------------------------------------------------------------|
| Info: OpenCL C code successfully compiled. |
| FP64 Compute (double, fma ) 0.259 TFLOPs/s (1/32) |
| FP32 Compute (float , fma ) 8.306 TFLOPs/s ( 1x ) |
| FP16 Compute (half2 , fma ) 16.393 TFLOPs/s ( 2x ) |
| INT64 Compute (long , a*b+c) 1.944 TIOPs/s (1/4 ) |
| INT32 Compute (int , a*b+c) 7.759 TIOPs/s ( 1x ) |
| INT16 Compute (short2, a*b+c) 6.560 TIOPs/s ( 1x ) |
| INT8 Compute (char4 , dp4a ) 0.933 TIOPs/s (1/8 ) |
| Memory Bandwidth ( coalesced read ) 397.86 GB/s |
| Memory Bandwidth ( coalesced write) 424.81 GB/s |
| Memory Bandwidth (misaligned read ) 260.84 GB/s |
| Memory Bandwidth (misaligned write) 81.25 GB/s |
| PCIe Bandwidth (send ) 2.78 GB/s |
| PCIe Bandwidth ( receive ) 2.80 GB/s |
| PCIe Bandwidth ( bidirectional) (Gen1 x16) 2.82 GB/s |
|-----------------------------------------------------------------------------|
解锁后：
|----------------.------------------------------------------------------------|
| Device ID | 0 |
| Device Name | NVIDIA CMP 40HX |
| Device Vendor | NVIDIA Corporation |
| Device Driver | 616.92 (Windows) |
| OpenCL Version | OpenCL C 3.0 |
| Compute Units | 34 at 1650 MHz (2176 cores, 7.181 TFLOPs/s) |
| Memory, Cache | 8024 MB VRAM, 1088 KB global / 48 KB local |
| Buffer Limits | 2006 MB global, 64 KB constant |
|----------------'------------------------------------------------------------|
| Info: OpenCL C code successfully compiled. |
| FP64 Compute (double, fma ) 0.259 TFLOPs/s (1/32) |
| FP32 Compute (float , fma ) 8.315 TFLOPs/s ( 1x ) |
| FP16 Compute (half2 , fma ) 16.235 TFLOPs/s ( 2x ) |
| INT64 Compute (long , a*b+c) 1.882 TIOPs/s (1/4 ) |
| INT32 Compute (int , a*b+c) 7.511 TIOPs/s ( 1x ) |
| INT16 Compute (short2, a*b+c) 6.561 TIOPs/s ( 1x ) |
| INT8 Compute (char4 , dp4a ) 26.970 TIOPs/s ( 4x ) |
| Memory Bandwidth ( coalesced read ) 396.99 GB/s |
| Memory Bandwidth ( coalesced write) 424.99 GB/s |
| Memory Bandwidth (misaligned read ) 261.32 GB/s |
| Memory Bandwidth (misaligned write) 80.71 GB/s |
| PCIe Bandwidth (send ) 4.94 GB/s |
| PCIe Bandwidth ( receive ) 4.94 GB/s |
| PCIe Bandwidth ( bidirectional) (Gen2 x16) 4.88 GB/s |
|-----------------------------------------------------------------------------|
反倒不敢相信了。算力算解锁了吗?
我怎么看前后只变了个int8，0.93->26.97, 其他基本没变化。
解锁前：
----------------.------------------------------------------------------------|
| Device ID | 0 |
| Device Name | NVIDIA CMP 40HX |
| Device Vendor | NVIDIA Corporation |
| Device Driver | 616.92 (Windows) |
| OpenCL Version | OpenCL C 3.0 |
| Compute Units | 34 at 1650 MHz (2176 cores, 7.181 TFLOPs/s) |
| Memory, Cache | 8191 MB VRAM, 1088 KB global / 48 KB local |
| Buffer Limits | 2047 MB global, 64 KB constant |
|----------------'------------------------------------------------------------|
| Info: OpenCL C code successfully compiled. |
| FP64 Compute (double, fma ) 0.259 TFLOPs/s (1/32) |
| FP32 Compute (float , fma ) 8.306 TFLOPs/s ( 1x ) |
| FP16 Compute (half2 , fma ) 16.393 TFLOPs/s ( 2x ) |
| INT64 Compute (long , a*b+c) 1.944 TIOPs/s (1/4 ) |
| INT32 Compute (int , a*b+c) 7.759 TIOPs/s ( 1x ) |
| INT16 Compute (short2, a*b+c) 6.560 TIOPs/s ( 1x ) |
| INT8 Compute (char4 , dp4a ) 0.933 TIOPs/s (1/8 ) |
| Memory Bandwidth ( coalesced read ) 397.86 GB/s |
| Memory Bandwidth ( coalesced write) 424.81 GB/s |
| Memory Bandwidth (misaligned read ) 260.84 GB/s |
| Memory Bandwidth (misaligned write) 81.25 GB/s |
| PCIe Bandwidth (send ) 2.78 GB/s |
| PCIe Bandwidth ( receive ) 2.80 GB/s |
| PCIe Bandwidth ( bidirectional) (Gen1 x16) 2.82 GB/s |
|-----------------------------------------------------------------------------|
解锁后：
|----------------.------------------------------------------------------------|
| Device ID | 0 |
| Device Name | NVIDIA CMP 40HX |
| Device Vendor | NVIDIA Corporation |
| Device Driver | 616.92 (Windows) |
| OpenCL Version | OpenCL C 3.0 |
| Compute Units | 34 at 1650 MHz (2176 cores, 7.181 TFLOPs/s) |
| Memory, Cache | 8024 MB VRAM, 1088 KB global / 48 KB local |
| Buffer Limits | 2006 MB global, 64 KB constant |
|----------------'------------------------------------------------------------|
| Info: OpenCL C code successfully compiled. |
| FP64 Compute (double, fma ) 0.259 TFLOPs/s (1/32) |
| FP32 Compute (float , fma ) 8.315 TFLOPs/s ( 1x ) |
| FP16 Compute (half2 , fma ) 16.235 TFLOPs/s ( 2x ) |
| INT64 Compute (long , a*b+c) 1.882 TIOPs/s (1/4 ) |
| INT32 Compute (int , a*b+c) 7.511 TIOPs/s ( 1x ) |
| INT16 Compute (short2, a*b+c) 6.561 TIOPs/s ( 1x ) |
| INT8 Compute (char4 , dp4a ) 26.970 TIOPs/s ( 4x ) |
| Memory Bandwidth ( coalesced read ) 396.99 GB/s |
| Memory Bandwidth ( coalesced write) 424.99 GB/s |
| Memory Bandwidth (misaligned read ) 261.32 GB/s |
| Memory Bandwidth (misaligned write) 80.71 GB/s |
| PCIe Bandwidth (send ) 4.94 GB/s |
| PCIe Bandwidth ( receive ) 4.94 GB/s |
| PCIe Bandwidth ( bidirectional) (Gen2 x16) 4.88 GB/s |
|-----------------------------------------------------------------------------|
