Article index
- 1 – Overview
- 2 – Quadro P5000 Gallery
- 3 – Quadro P5000 GPU Data
- 4 – Benchmarks: Quadro P5000 vs GeForce GTX 1080
- 4.1 – Preparation
- 4.2 – SPECviewperf 12.1
- 4.3 – LuxMark 3.1
- 4.4 – CineBench R15.0
- 4.5 – FurMark 1.19
- 4.6 – Unigine Superposition
- 4.7 – GeeXLab: heavy pixel shader test
- 4.8 – GeeXLab: two sided lighting test
- 5 – Burn-in Test
- 6 – Conclusion
1 – Overview
I received a PNY Quadro P5000 graphics accelerator three days ago. But for few days only… So during this short lapse of time, I decided to compare its graphics performances with the GeForce GTX 1080 ones. Why the GTX 1080? Because both graphics cards are based on the same Pascal GP104 GPU and the P5000 cost more than twice the price of the GTX 1080…
Here are the main features of this PNY Quadro P5000:
- GPU: Pascal GP104, base clock: 1607MHz, boost clock: 1733MHz, TSMC 16nm
- CUDA cores: 2560
- Texture units: 160
- ROPs: 64
- Memory: 16GB GDDR5X @ 4513MHz, 256-bit memory interface
- TDP: 180W
- FP32: 8.9 TFLOPS
- FP64: 1/32 FP32
2 – Quadro P5000 Gallery
The bundle: the Quadro P5000, a CDROM with drivers / utilities, a quick user’s guide, a power connector, three DisplayPort to DVI adapters and an audio stereo connector.
Now the Quadro P5000:
3 – Quadro P5000 GPU Data
The Quadro P5000 has the same clock speed than the GeForce GTX 180 reference model. The max power target is 100% TDP, the overclocking is clearly not an option (the P5000 has only one 8-pin power connector).
Update (2017.05.15 @ 20:30 pm)
The GTX 1080 (or GTX 1070/1060) exposes 391 OpenGL extensions while 398 are present on the Quadro P5000. Here is the list of those 7 additional GL extensions available on the P5000:
- GL_ARB_cl_event - GL_NVX_shared_sync_object - GL_NV_deep_texture3D - GL_NVX_sysmem_buffer - WGL_NV_gpu_affinity - WGL_NV_swap_group - WGL_NV_video_capture
Next page: the benchmarks!
The Quadro has seven more OpenGL extensions than the GeForce.
Whats the extra mumbo-jumbo there?
Shame on me, I didn’t look at the GL extensions. I’m fixing this bug…
List of the 7 mysterious extensions:
http://www.geeks3d.com/20170515/test-nvidia-quadro-p5000-vs-geforce-gtx-1080/#20170515_01
Nice. Can you test also the CUDA Performace?
Just want to notice, what some OGL fixed pipeline functions like
glLightModeli(GL_LIGHT_MODEL_TWO_SIDE, GL_TRUE);
in test “GeeXLab: two sided lighting test”
is software-limited on not quadro cards. (10-40 times performance loss)
To get around this we can use shaders.
Discussion and tests on nvidia/quadro and radeon you can find here http://www.gamedev.ru/code/forum/?id=192325
Is there a way to use the GTX 1080 drivers on the Quadro P5000? Thanks!
No. You think they’re stupid to let you do that? And also there are hardware and firmware differences.
And there’s another reason they block you from doing that. Quadro are used for 10-bit content editing. Geforce is locked out and can only output 8-bit.
Your score is messed up. You messed up with installing the drivers or settings because my Quadro P600 gets 107FPS in cinebench.
It was a great test with very interesting results. There are a lot of confusions and misunderstandings about Quadro graphic cards which this test was a light on those.
One thing grabbed my attention in SPECviewperf test. I noticed that all applications that using polygons (3ds Max, Maya and Showcase) are having almost the same results between P5000 and GTX 1080. However, in the applications that using NURBS surfaces by default and having a “tessellation” process in the background (CATIA, SolidWorks, NX, Creo), P5000 performs a lot better than GTX.
I’m using Autodesk Alias and SolidWorks and I experienced this lack of performance on GeForce cards before. Also, SolidWorks has a feature called Realview which it doesn’t work on GeForce graphic cards.
Thank you for this test. You did an awesome job.
hi !
a lot of others things are intentionally *removed* from geforce vs quadro drivers.
(appart from ECC support, both cards are 100% identical hardware )
– 10bits support : Nvidia is changing is mind there, geforce does have “partial” 10bit display (for example in full screen HDR games). but – for now – only quadro can display a windowed 10bit openGL frame
– famous error 42 when doing GPU passthrough in a virtual machine. only GRID and quadro card can be virtualized… so nice to have to pay 5x the price to enable virtualization!!
– TCC mode (telsa compute cluster) : a specific mode that allow a card to NOT use the windows WDDM driver (that consume a bunch of VRAM) + no need to hook a display to use the card + TCC mode is needed by high end CUDA software. all Titan card can be switched between WDDM and TCC mode… so why need to buy a 5x more expensive Tesla card, that is slower 🙂
– many OpenGL optimization are simply removed from the geforce driver
and *of course* you can not :
– install a geforce card with a quadro driver, and vice versa
– mix a geforce with a quadro card in the same system and use both drivers
– this is a intentional mess…
– edit : not sure yet what i’ve changed in the newest titanXP drivers, but competition from AMD vega FE is kicking in the ass of this bullshit politic of Nvidia :). currently only affecting openGL optimization
https://www.reddit.com/r/nvidia/comments/6qtf9b/titan_x_pascal_benches_with_the_new_38512_driver/