hashcat (v3.20) almost half slower vs cudaHashcat v2.01

[darkoan]

C:\hashcat-3.20>hashcat64.exe -m 2500 -b
hashcat (v3.20) starting in benchmark mode...

* Device #1: Old CUDA chipset 3.0 detected, OpenCL performance is reduced.
             For ideal hashcat performance on NVIDIA GPU you need Shader Model 5.0 or higher
nvmlDeviceGetCurrPcieLinkWidth(): Not Supported

nvmlDeviceGetClockInfo(): Not Supported

nvmlDeviceGetFanSpeed(): Not Supported

nvmlDeviceGetClockInfo(): Not Supported

nvmlDeviceGetTemperatureThreshold(): Not Supported

nvmlDeviceGetTemperatureThreshold(): Not Supported

nvmlDeviceGetUtilizationRates(): Not Supported

nvmlDeviceGetPowerManagementLimit(): Not Supported

OpenCL Platform #1: NVIDIA Corporation
======================================
* Device #1: GeForce GTX 780M, 1024/4096 MB allocatable, 8MCU

OpenCL Platform #2: Intel(R) Corporation
========================================
* Device #2: Intel(R) Core(TM) i7-4700HQ CPU @ 2.40GHz, skipped

Hashtype: WPA/WPA2

Speed.Dev.#1.....:    25503 H/s (79.99ms)

Started: Sun Jan 01 12:29:49 2017
Stopped: Sun Jan 01 12:29:59 2017

_____________________________________________________________________________________________________________________


C:\cudaHashcat-2.01>cudaHashcat64.exe -m 2500 -b
cudaHashcat v2.01 starting in benchmark-mode...

Device #1: GeForce GTX 780M, 4096MB, 797Mhz, 8MCU

Hashtype: WPA/WPA2
Workload: 1024 loops, 32 accel

Speed.GPU.#1.:    44128 H/s

Started: Sun Jan 01 12:28:19 2017
Stopped: Sun Jan 01 12:28:38 2017