Laptop Forums and Notebook Computer Discussion - View Single Post

mangos47 · 07-20-2011, 05:29 AM

Benchmark results of my first eGPU setup. Thanks for the elite eGPU team here at NBR for making such a fantastic solution for laptop gamers.

My setup hardware:

Laptop: Sager NP8662 (Clevo M860tu), P8700@2.53GHz, 4GB DDR3, dedicated GPU (dGPU) Nvidia GTX 260m 1GB. This laptop is based on PM45 chipset with ICH9M south bridge so 6 pcie ports other than the pciex16 used by dGPU. On default port 1 is WWAN card, port 2 is expresscard, port 3 is WLAN/wifi card, port 4 is not physically connected, port 5 is card reader + 1394 port, port 6 is Realtek Gigabit Lan.

PSU: Andyson F500M, 500W

external GPU (eGPU): PE4H v2.4, EC2C on port 2, PM3N on port 1. Gainward GTX460 1GB. ASUS GTX580 1.5GB (DirectCuII).

External monitor: 1920x1080

eGPU configuration:

software environment: Windows 7 Ultimate x64, Nvidia driver 275.33 notebook version. I only edited nvam.inf of my current laptop driver to add in the description for GTX460 and manually install it. There’s a warning about unverified publisher which I just ignore.

x1/x1e mode: using either PM3N in WWAN slot (port 1) or EC2C in expresscard slot (port 2) alone to link up PE4H. This setup works as hot plug and play (EC2C) without the need of setup1.x program. When using PM3N on port 1 I have the option to set the port as x2 mode though only a single lane is utilized here. This is known as x1e mode (correct me if I’m wrong) but since my eGPU is GTX460 so it doesn’t have any advantage over usual x1 mode like ATI/older Nvidia cards do.

x2 mode: using both PM3N in WWAN slot (port 1) and EC2C in expresscard slot (port 2). This requires the setup 1.x program to configure port 1 as x2 mode and disable port 2 though both port/slot are physically connected to PE4H. Thanks Khenglish for the help on properly set up x2 mode in setup1.x program (a couple threads above).

So here are the benchmark results I have so far. All benchmark are done on external monitor either at program default resolution or 1920x1080.
red color bold number shows how much improvement of x2 over x1 link.

1. 3dmark06 (Pro 1.2), 1280x1024, default setting

a. dGPU (GTX 260m): 9601, SM2.0:4501, HDR/SM3.0: 4481.
b. eGPU (GTX 460) x1: 5185, SM2.0:2124, HDR/SM3.0:2021
c. eGPU (GTX 460) x2: 8115, SM2.0: 3599, HDR/SM3.0: 3647, 56%
d. eGPU (GTX 580) x1: 5461, SM2.0: 2271, HDR/SM3.0: 2123.
e. eGPU (GTX 580) x2: 8732, SM2.0: 3848, HDR/SM3.0: 3994, 59%

2. 3dmarkVantage, 1280x1024, default setting

a. dGPU (GTX 260m): P5526, GPU4402
b. eGPU (GTX 460) x1: P8988, GPU7646
c. eGPU (GTX 460) x2: P11420, GPU9754, 27%
d. eGPU (GTX 580) x1: P12740, GPU11147
e. eGPU (GTX 580) x2: P15457, GPU14129, 26%

3. 3dmark2011, performance preset, (720p / 1080p)

a. dGPU (GTX 260m): not supported
b. eGPU (GTX 460) x1: P2110 / X977
c. eGPU (GTX 460) x2: P2408 / X1045, 14% / 7%
d. eGPU (GTX 580) x1: P2695 / X1569
e. eGPU (GTX 580) x2: P3274 / X1740, 21% / 10%

4. Warhammer 40k DoW2 Chaos Rising, 1920x1080, ultra setting (average/ max/ min, in fps)

a. dGPU (GTX 260m): 22/ 51/ 7
b. eGPU (GTX 460) x1: 22/55/8
c. eGPU (GTX 460) x2: 30/70/11, 36%
c. eGPU (GTX 580) x1: 25/65/12,
c. eGPU (GTX 580) x2: 35/81/14, 40%

5. Resident Evil 5 benchmark, dx10.0 version, 1920x1080, 4xAA, high for everything, motion blur off, variable scene.

a. dGPU (GTX 260m): 39.9fps
b. eGPU (GTX 460) x1: 51.7fps
c. eGPU (GTX 460) x2: 62.5fps, 20%
d. eGPU (GTX 580) x1: 61.1fps
e. eGPU (GTX 580) x2: 73.4fps, 20%

6. Devil May Cry 4 benchmark, dx10, 1920x1080, AA:C16xQ, super high, variable, scene 1-4 fps:

a. dGPU (GTX 260m): 51.7 | 34.3 | 55.1 | 39.2
b. eGPU (GTX 460) x1: 69.9 | 45.7 | 97.7 | 49.6
c. eGPU (GTX 460) x2: 93.2 | 62.6 | 112.8 | 64.3, about 30%
d. eGPU (GTX 580) x1: 88.6 | 56.1 | 117.1 | 65.1
e. eGPU (GTX 580) x2: 136.7 | 91.2 | 166.92 | 93.4, 40%~60%

7. Crysis I, GPU test, 1920x1080, 4xAA, very high, average|min|max fps:

a. dGPU (GTX 260m): 9.7|7.2|10.7
b. eGPU (GTX 460) x1: 11.6|6.4|15.2
c. eGPU (GTX 460) x2: 15.9|6.0|19.5, 37%
d. eGPU (GTX 580) x1: 14.4|7.8|20.4
e. eGPU (GTX 580) x2: 22.1|7.9|28.8, 53%

8. Crysis II, 1920x1080, high res texture, average|min|max fps:

DX9, high graphics setting (worst)
a. dGPU (GTX 260m): 30.6 | 20.3 | 42.1
b. eGPU (GTX 460) x1: 28.2 | 18.8 | 39.2
c. eGPU (GTX 460) x2: 35.9 | 21.9 | 54.6, 27%
d. eGPU (GTX 580) x1: 33.4|21.8|46.6
e. eGPU (GTX 580) x2: 48.6|28.4|63.2, 45%
-----------------------------------------
DX9, ultra graphics setting (best)
a. dGPU (GTX 260m): 11.9 | 11.7 | 18.4
b. eGPU (GTX 460) x1: 12.2 | 9.00 | 18.1
c. eGPU (GTX 460) x2: 18.7 | 14.4 | 24.7, 53%
d. eGPU (GTX 580) x1: 16.1|12.6|22.1
e. eGPU (GTX 580) x2: 24.8|17.1|34.6, 54%
-----------------------------------------
DX11, high graphics setting (worst)
a. dGPU (GTX 260m): [font=Calibri]not supported
b. eGPU (GTX 460) x1: 23.9 | 14.3 | 31.1
c. eGPU (GTX 460) x2: 29.2 | 22.6 | 31.7, 22%
d. eGPU (GTX 580) x1: 31.1|18.9|47.8
e. eGPU (GTX 580) x2: 35.2|20.1|57.1, 13%
-----------------------------------------
DX11, ultra graphics setting (best)
a. dGPU (GTX 260m): not supported
b. eGPU (GTX 460) x1: 10.4 | 7.60 | 15.0
c. eGPU (GTX 460) x2: 13.7 | 10.2 | 19.7, 31%
d. eGPU (GTX 580) x1: 12.7|10.9|19.8
e. eGPU (GTX 580) x2: 19.4|14.4|28.2, 52%

9. Metro 2033, 1920x1080, AAA, average|min|max fps:

DX10, low, 4xAF
a. dGPU (GTX 260m): 20.0 | 4.2 | 69.2
b. eGPU (GTX 460) x1: 17.3 | 1.9 | 77.6
c. eGPU (GTX 460) x2: 24.3 | 4.7 | 109.6, 40%
d. eGPU (GTX 580) x1: 20.0|5.2|94.7
e. eGPU (GTX 580) x2: 26.3|3.1|123.5, 31%
-----------------------------------------
DX10, high, 16xAF
a. dGPU (GTX 260m): 11.0 | 2.4 | 35.8
b. eGPU (GTX 460) x1: 12.0 | 3.1 | 37.2
c. eGPU (GTX 460) x2: 16.3 | 4.9 | 62.3, 35%
d. eGPU (GTX 580) x1: 15.0|3.1|55.4
e. eGPU (GTX 580) x2: 22.6|4.1|94.3, 50%
-----------------------------------------
DX11, high, 16xAF
a. dGPU (GTX 260m): [font=Calibri]not supported
b. eGPU (GTX 460) x1: 9.7 | 3.0 | 38.4
c. eGPU (GTX 460) x2: 12.3 | 3.9 | 45.3, 27%
d. eGPU (GTX 580) x1: 13.0|4.0|51.5
e. eGPU (GTX 580) x2: 19.3|2.4|49.7, 48%

10. Furmark 1.8.2, 1920x1080, 60s benchmark (0x/4x/16x, as in MSAA)

a. dGPU (GTX 260m): 2102/1223/359
b. eGPU (GTX 460) x1: 3785/3255/1080
c. eGPU (GTX 460) x2: 3916/3355/1092, 3%
d. eGPU (GTX 580) x1: 6414/5451/1829
e. eGPU (GTX 580) x2: 6802/5816/1909, 6%

11 CUDA bandwidth test (host-to-device/device-to-host/device-to-device, in MB/s)

a. dGPU (GTX 260m): 1328/1092.9/41683.1
b. eGPU (GTX 460) x1e: 156.8/189.5/60167.0
c. eGPU (GTX 460) x2: 300.7/356.9/60058.8, x2

12 OpenCL bandwidth test (host-to-device/device-to-host/device-to-device, in MB/s)

a. dGPU (GTX 260m): 1391.2/1090.8/23775.9
b. eGPU (GTX 460) x1: 156.9/190.9/49745.1
c. eGPU (GTX 460) x2: 302.2/3361.1/52010.2, x2

Overall I’m quite satisfied with the x2 performance of this eGPU setup. x1 is not so impressive as compared to laptop’s internal GTX 260m but still boosts performance in most cases. If you are using a non-gaming laptop then x1 can still mean a lot to you. Now I’m thinking my next laptop will probably be a thin & light SB CPU based with intel HD/HD 3000 so that I can have Optimus enabled for even better performance (Optimus > non-Optimus x2 if I understand it correctly).

07-20-2011, 05:29 AM	#4531 (permalink)
mangos47 Notebook Consultant Join Date: Feb 2010 Posts: 186 Rep Power: 6	re: DIY eGPU experiences Benchmark results of my first eGPU setup. Thanks for the elite eGPU team here at NBR for making such a fantastic solution for laptop gamers. My setup hardware: Laptop: Sager NP8662 (Clevo M860tu), P8700@2.53GHz, 4GB DDR3, dedicated GPU (dGPU) Nvidia GTX 260m 1GB. This laptop is based on PM45 chipset with ICH9M south bridge so 6 pcie ports other than the pciex16 used by dGPU. On default port 1 is WWAN card, port 2 is expresscard, port 3 is WLAN/wifi card, port 4 is not physically connected, port 5 is card reader + 1394 port, port 6 is Realtek Gigabit Lan. PSU: Andyson F500M, 500W external GPU (eGPU): PE4H v2.4, EC2C on port 2, PM3N on port 1. Gainward GTX460 1GB. ASUS GTX580 1.5GB (DirectCuII). External monitor: 1920x1080 eGPU configuration: software environment: Windows 7 Ultimate x64, Nvidia driver 275.33 notebook version. I only edited nvam.inf of my current laptop driver to add in the description for GTX460 and manually install it. There’s a warning about unverified publisher which I just ignore. x1/x1e mode: using either PM3N in WWAN slot (port 1) or EC2C in expresscard slot (port 2) alone to link up PE4H. This setup works as hot plug and play (EC2C) without the need of setup1.x program. When using PM3N on port 1 I have the option to set the port as x2 mode though only a single lane is utilized here. This is known as x1e mode (correct me if I’m wrong) but since my eGPU is GTX460 so it doesn’t have any advantage over usual x1 mode like ATI/older Nvidia cards do. x2 mode: using both PM3N in WWAN slot (port 1) and EC2C in expresscard slot (port 2). This requires the setup 1.x program to configure port 1 as x2 mode and disable port 2 though both port/slot are physically connected to PE4H. Thanks Khenglish for the help on properly set up x2 mode in setup1.x program (a couple threads above). So here are the benchmark results I have so far. All benchmark are done on external monitor either at program default resolution or 1920x1080. red color bold number shows how much improvement of x2 over x1 link. 1. 3dmark06 (Pro 1.2), 1280x1024, default setting a. dGPU (GTX 260m): 9601, SM2.0:4501, HDR/SM3.0: 4481. b. eGPU (GTX 460) x1: 5185, SM2.0:2124, HDR/SM3.0:2021 c. eGPU (GTX 460) x2: 8115, SM2.0: 3599, HDR/SM3.0: 3647, 56% d. eGPU (GTX 580) x1: 5461, SM2.0: 2271, HDR/SM3.0: 2123. e. eGPU (GTX 580) x2: 8732, SM2.0: 3848, HDR/SM3.0: 3994, 59% 2. 3dmarkVantage, 1280x1024, default setting a. dGPU (GTX 260m): P5526, GPU4402 b. eGPU (GTX 460) x1: P8988, GPU7646 c. eGPU (GTX 460) x2: P11420, GPU9754, 27% d. eGPU (GTX 580) x1: P12740, GPU11147 e. eGPU (GTX 580) x2: P15457, GPU14129, 26% 3. 3dmark2011, performance preset, (720p / 1080p) a. dGPU (GTX 260m): not supported b. eGPU (GTX 460) x1: P2110 / X977 c. eGPU (GTX 460) x2: P2408 / X1045, 14% / 7% d. eGPU (GTX 580) x1: P2695 / X1569 e. eGPU (GTX 580) x2: P3274 / X1740, 21% / 10% 4. Warhammer 40k DoW2 Chaos Rising, 1920x1080, ultra setting (average/ max/ min, in fps) a. dGPU (GTX 260m): 22/ 51/ 7 b. eGPU (GTX 460) x1: 22/55/8 c. eGPU (GTX 460) x2: 30/70/11, 36% c. eGPU (GTX 580) x1: 25/65/12, c. eGPU (GTX 580) x2: 35/81/14, 40% 5. Resident Evil 5 benchmark, dx10.0 version, 1920x1080, 4xAA, high for everything, motion blur off, variable scene. a. dGPU (GTX 260m): 39.9fps b. eGPU (GTX 460) x1: 51.7fps c. eGPU (GTX 460) x2: 62.5fps, 20% d. eGPU (GTX 580) x1: 61.1fps e. eGPU (GTX 580) x2: 73.4fps, 20% 6. Devil May Cry 4 benchmark, dx10, 1920x1080, AA:C16xQ, super high, variable, scene 1-4 fps: a. dGPU (GTX 260m): 51.7 \| 34.3 \| 55.1 \| 39.2 b. eGPU (GTX 460) x1: 69.9 \| 45.7 \| 97.7 \| 49.6 c. eGPU (GTX 460) x2: 93.2 \| 62.6 \| 112.8 \| 64.3, about 30% d. eGPU (GTX 580) x1: 88.6 \| 56.1 \| 117.1 \| 65.1 e. eGPU (GTX 580) x2: 136.7 \| 91.2 \| 166.92 \| 93.4, 40%~60% 7. Crysis I, GPU test, 1920x1080, 4xAA, very high, average\|min\|max fps: a. dGPU (GTX 260m): 9.7\|7.2\|10.7 b. eGPU (GTX 460) x1: 11.6\|6.4\|15.2 c. eGPU (GTX 460) x2: 15.9\|6.0\|19.5, 37% d. eGPU (GTX 580) x1: 14.4\|7.8\|20.4 e. eGPU (GTX 580) x2: 22.1\|7.9\|28.8, 53% 8. Crysis II, 1920x1080, high res texture, average\|min\|max fps: DX9, high graphics setting (worst) a. dGPU (GTX 260m): 30.6 \| 20.3 \| 42.1 b. eGPU (GTX 460) x1: 28.2 \| 18.8 \| 39.2 c. eGPU (GTX 460) x2: 35.9 \| 21.9 \| 54.6, 27% d. eGPU (GTX 580) x1: 33.4\|21.8\|46.6 e. eGPU (GTX 580) x2: 48.6\|28.4\|63.2, 45% ----------------------------------------- DX9, ultra graphics setting (best) a. dGPU (GTX 260m): 11.9 \| 11.7 \| 18.4 b. eGPU (GTX 460) x1: 12.2 \| 9.00 \| 18.1 c. eGPU (GTX 460) x2: 18.7 \| 14.4 \| 24.7, 53% d. eGPU (GTX 580) x1: 16.1\|12.6\|22.1 e. eGPU (GTX 580) x2: 24.8\|17.1\|34.6, 54% ----------------------------------------- DX11, high graphics setting (worst) a. dGPU (GTX 260m): [font=Calibri]not supported b. eGPU (GTX 460) x1: 23.9 \| 14.3 \| 31.1 c. eGPU (GTX 460) x2: 29.2 \| 22.6 \| 31.7, 22% d. eGPU (GTX 580) x1: 31.1\|18.9\|47.8 e. eGPU (GTX 580) x2: 35.2\|20.1\|57.1, 13% ----------------------------------------- DX11, ultra graphics setting (best) a. dGPU (GTX 260m): not supported b. eGPU (GTX 460) x1: 10.4 \| 7.60 \| 15.0 c. eGPU (GTX 460) x2: 13.7 \| 10.2 \| 19.7, 31% d. eGPU (GTX 580) x1: 12.7\|10.9\|19.8 e. eGPU (GTX 580) x2: 19.4\|14.4\|28.2, 52% 9. Metro 2033, 1920x1080, AAA, average\|min\|max fps: DX10, low, 4xAF a. dGPU (GTX 260m): 20.0 \| 4.2 \| 69.2 b. eGPU (GTX 460) x1: 17.3 \| 1.9 \| 77.6 c. eGPU (GTX 460) x2: 24.3 \| 4.7 \| 109.6, 40% d. eGPU (GTX 580) x1: 20.0\|5.2\|94.7 e. eGPU (GTX 580) x2: 26.3\|3.1\|123.5, 31% ----------------------------------------- DX10, high, 16xAF a. dGPU (GTX 260m): 11.0 \| 2.4 \| 35.8 b. eGPU (GTX 460) x1: 12.0 \| 3.1 \| 37.2 c. eGPU (GTX 460) x2: 16.3 \| 4.9 \| 62.3, 35% d. eGPU (GTX 580) x1: 15.0\|3.1\|55.4 e. eGPU (GTX 580) x2: 22.6\|4.1\|94.3, 50% ----------------------------------------- DX11, high, 16xAF a. dGPU (GTX 260m): [font=Calibri]not supported b. eGPU (GTX 460) x1: 9.7 \| 3.0 \| 38.4 c. eGPU (GTX 460) x2: 12.3 \| 3.9 \| 45.3, 27% d. eGPU (GTX 580) x1: 13.0\|4.0\|51.5 e. eGPU (GTX 580) x2: 19.3\|2.4\|49.7, 48% 10. Furmark 1.8.2, 1920x1080, 60s benchmark (0x/4x/16x, as in MSAA) a. dGPU (GTX 260m): 2102/1223/359 b. eGPU (GTX 460) x1: 3785/3255/1080 c. eGPU (GTX 460) x2: 3916/3355/1092, 3% d. eGPU (GTX 580) x1: 6414/5451/1829 e. eGPU (GTX 580) x2: 6802/5816/1909, 6% 11 CUDA bandwidth test (host-to-device/device-to-host/device-to-device, in MB/s) a. dGPU (GTX 260m): 1328/1092.9/41683.1 b. eGPU (GTX 460) x1e: 156.8/189.5/60167.0 c. eGPU (GTX 460) x2: 300.7/356.9/60058.8, x2 12 OpenCL bandwidth test (host-to-device/device-to-host/device-to-device, in MB/s) a. dGPU (GTX 260m): 1391.2/1090.8/23775.9 b. eGPU (GTX 460) x1: 156.9/190.9/49745.1 c. eGPU (GTX 460) x2: 302.2/3361.1/52010.2, x2 Overall I’m quite satisfied with the x2 performance of this eGPU setup. x1 is not so impressive as compared to laptop’s internal GTX 260m but still boosts performance in most cases. If you are using a non-gaming laptop then x1 can still mean a lot to you. Now I’m thinking my next laptop will probably be a thin & light SB CPU based with intel HD/HD 3000 so that I can have Optimus enabled for even better performance (Optimus > non-Optimus x2 if I understand it correctly). Last edited by mangos47 : 09-09-2011 at 04:52 AM.