locked
RemoteFX Unofficial graphics cards and graphics score results for SP1 RC RRS feed

 > 

  • General discussion

  • Discussion
    Sign in to vote
    0
    Sign in to vote

    Officially you should be using one of the approved Nvidia or AMD workstation cards.  However many of you may be testing at home or working for a small business where the cost of the workstation graphics cards are prohibitive.  Thus to make it easier for other testers and future small businesses...

    1) What graphics card have you used on the server

    2) What graphics scores have you gotten when connecting in to your Hyper-V R2 SP1 RC hosted Windows 7 virtual machine from a SP1 RC client (my 2008 R2 server and Windows 7 PC scores were close)

    Here is what I have tested this last weekend:

    ATI 3870 - Futuremark 3dmark06 score of approximately 4400; Furmark failed (remoteFX has limited openGL)

    add your results below:

    • Edited by dkyeager Thursday, December 2, 2010 2:55 PM
    Monday, November 8, 2010 7:28 PM

All replies

  • Discussion
    Sign in to vote
    0
    Sign in to vote

    Host is dual socket AMD Opteron 8347 HE (quad-core 1.9 GHz) with 20 GB (4x4gb + 4x1gb) DDR2 Registered ECC RAM.

    VM1 is Windows 7 Ultimate x64 with 2 virtual CPUs and dynamic memory (min 2048: max: 4096) and RemoteFX with 2x 1920x1200 monitors

    Both with SP1 RC

     

    1) ATI Radeon HD 5770 1GB (850 MHz GPU / 1200 MHz Memory)

    2)

      Host VM1
    3DMarks Score: 8447 3100
    SM2.0 Score: 3464 1504
    HDR/SM3.0 Score: 3465 1565
    CPU Score: 2964 594
    GT1 - Return To Proxycon: 27.72 (FPS) 17.59 (FPS)
    GT2 - Firefly Forest: 30.00 (FPS) 7.48 (FPS)
    CPU1 - Red Valley: 1.08 (FPS) 0.25 (FPS)
    CPU2 - Red Valley: 1.30 (FPS) 0.23 (FPS)
    HDR1 - Canyon Flight: 48.08 (FPS) 21.61 (FPS)
    HDR2 - Deep Freeze: 21.22 (FPS) 9.68 (FPS)

    These are all lower than I would have expected (even on the host), so I may have some other issue going on here.

    Tuesday, November 9, 2010 5:37 AM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    The operating system for the host does make a difference.  Here is the HD 3870 once again under Windows Vista 64-bit versus 2008 server and VM1 and then I added yours on the right:

    Vista       2008 VM1 VM1/2008 2008 VM1 VM1/2008
    12027 9708 4378 45% Overall 3dmark06 score 8447 3100 37%
    4784 4188 2058 49% SM2 3464 1504 43%
    5283 3835 1949 51% SM3 3465 1565 45%
    3847 3289 1022 31% CPU 2964 594 20%
    Graphics
    37.78 33.72 23.68 70% Return to proxycon 27.72 17.59 63%
    41.96 36.08 11.23 31% Firefly forest 30 7.48 25%
    CPU
    1.28 1.1 0.42 38% Red Valley 1 1.08 0.25 23%
    1.86 1.57 0.4 25% Red Valley 2 1.3 0.23 18%
    HDR
    48.86 42.71 26.29 62% Canyon flight 48.08 21.61 45%
    56.79 33.99 12.7 37% Deep Freeze 21.22 9.68 46%

    Futuremark does say that results do no transfer accross operating systems, but Vista and Windows 7 are quite close.  Thus definately a hit - up to two thirds whan calculated from the maximum desktop OS to the VM1.  High variation on the ratios between the different graphics and HDR tests.

    Random thoughts:

    ·         I assume we are getting the best results with the large resolution we both have chosen, however I may test with much smaller resolution. (I did test that you can mix the resolutions for the VMs on the card (ie 1920 x1200 plus 1280x1024 plus 1024x768). 

    ·         RemoteFX might be holding back some for multiple users or maybe some debug symbols still left the software (unlikely for a RC).

    ·         Could also increase the CPU count for the VM just to see what happens (I would prefer to  run with two, but I may want to up to three given our CPU ratios and  that we are both set for 2 of 4 CPUs, the RemoteFX driver overhead most likely used the same processors rather than being more multithreaded and running on the other processors). 

    ·         Ratios might also improve with a second card with no 2008 use (I do have one if it does not have to match the os card, but it is nvidia which could cause some problems).  I am also using amd's drivers on the 2008 side.

    Thursday, November 11, 2010 7:43 AM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    I don't have the numbers in front of me right now, but I also ran the tests with 1 virtual CPU and 4 virtual CPUs.  It did change the numbers to some extent.  If I remember right, the CPU score went up with 4 cores, but the GPU scores went down.  I'll try and check the actual numbers again tonight when I am in front of my test machine.

    I also tested to see if there was a difference between static memory and dynamic memory.  I didn't see much of a difference one way or the other.

    I should be getting some newer CPUs later today (dual 6-core 1.8 GHz), as well as some additional memory within the next couple of days (will have 8x 4 GB = 32 GB).  I will run the numbers again to see if the newer CPU architecture (45nm 6-core Istanbul vs 64nm quad-core Barcelona) makes a difference.

    If you don't mind me asking... what cpu and ram type are you using?

    Thursday, November 11, 2010 5:57 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    Nice Thread. Rfx peformance would also depend on the following parameters. So it would be good to mention these as well:

    1. GPU driver version (you can share what worked best for you)

    2. Network throughput (VM -> Client)

    3. Client <->VM latency

    This posting is provided "AS IS" with no warranties, and confers no rights. Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.
    Thursday, November 11, 2010 7:12 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    Nice Thread. Rfx peformance would also depend on the following parameters. So it would be good to mention these as well:

    1. GPU driver version (you can share what worked best for you)

    2. Network throughput (VM -> Client)

    3. Client <->VM latency

    This posting is provided "AS IS" with no warranties, and confers no rights. Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    Sorry for delay, was out of town.

    1.) current gpu driver version is 8.712 from earlier this year on a 3870.

    2.) stats taken during same time period while running 3dmark06:

                                                          Maximum                       Average

    server bytes/sec:                                   5,281                             115
    server packets/sec:                                    16                                1
    server processor time:                                52%                             18%

    vm1 bytes/sec:                                2,446,936                       757,466
    vm1 packets/sec:                                   1,074                             335           
    vm1 processor time:                                  100%                            65%

    workstation bytes/sec:                      4,852,212                       577,317
    workstation packets/sec:                         5,124                             613
    workstation processor time:                        100%                            38%

    (I realize they don't add up.  Will attempt again at a later date.  The workstation is a pentium 4 3.0ghz with 133 memory.  The server is a 2.9ghz phemon x4.  VM1 restricted to 2 cores.)  Other VMs were paused, other PCs were on but not in use.  network speed is 1gb, 10gb, 100mb respectively with jumbo packets enabled.)

    3) latency is 1ms, as reported by ping.

    Wednesday, November 17, 2010 9:53 PM
  • Discussion
    Sign in to vote
    1
    Sign in to vote

    I don't have the numbers in front of me right now, but I also ran the tests with 1 virtual CPU and 4 virtual CPUs.  It did change the numbers to some extent.  If I remember right, the CPU score went up with 4 cores, but the GPU scores went down.  I'll try and check the actual numbers again tonight when I am in front of my test machine.

    I also tested to see if there was a difference between static memory and dynamic memory.  I didn't see much of a difference one way or the other.

    I should be getting some newer CPUs later today (dual 6-core 1.8 GHz), as well as some additional memory within the next couple of days (will have 8x 4 GB = 32 GB).  I will run the numbers again to see if the newer CPU architecture (45nm 6-core Istanbul vs 64nm quad-core Barcelona) makes a difference.

    If you don't mind me asking... what cpu and ram type are you using?

    It is possible that I have encountered a stability problem with RemoteFX when the dynamic ram is operating below the minimum requirements for the operating system.  I would need to do more testing to confirm this (ie, no oc, different video card permutations).

    The system for this test server is an Asus M3A79-T Deluxe with a 790Fx chipset, Phenom 9850 black edition clocked rock stable at 2.9Ghz (never goes down, no real application problems, no real errors in logs), and 4-2Gb DDR2 1066 memory operating at 800mhz, a 1000watt power supply, raid 10.    I could put in an X6 and 16gb RAM in the future.   This system is more like what a small business would use as a server.

    My first goal is to complete the testing previously mentioned, then I do have other graphics cards to test.  This may take me a while (I just returned from being out of town on business).

    Wednesday, November 17, 2010 10:37 PM
  • Discussion
    Sign in to vote
    1
    Sign in to vote

    testing results continued...

    server has a phenom x4 @2.9ghz, 8GB  DDR2 @800, hd3870 pci-e@16, two seprate 1gb network links (one for management); vm1 is now windows 7 ultimate 64-bit 2048 ram static, dynamic disk; pc1 is a pentium 4 3.0 ghz, 1gb ram @ 133, hd 4650 agp x4, 100mb network link through 2 1gb switches and a 100mb hub, windows 7 professional 32-bit; vm2 is windows 7 enterprise 32-bit 2048 static ram, pc2 pentium 4 3.4 ghz hyperthreaded, 3gb dual channel ddr400 memory, 4670 AGP @ 8x, 1gb network link through 2 1gb switches, windows 7 professional 32-bit.  All have sata II raid 10 mirrored, except server which is raid 10.  All are regularly defragged.

    These results were obtained using catalyst 10.11, which was a royal pain to install compared to the 9 and early 10s.  This improved performance somewhat over earlier results.

    3dmark06 tested at different times:

                vm1/pc1   vm2/pc2

    total        4476       4000
    SM2         1992       1712
    SM3         2163       1876
    CPU         1005         991

    1             22.73       19.63
    2             10.47        8.91
    3               0.42        0.41
    4               0.39        0.38
    5             28.49       24.47
    6             14.76       13.05

    you might expect vm1/pc1 to be slower than vm2/pc2 given the faster network connection, processor, memory speed and size, and gpu.  However it could also be 32-bit being slower than 64-bit, or enterprise being slower than ultimate on the vm side or less latency on the network side so I tested vm1/pc2:

                vm1/pc2

    Total        4232
    SM2         1850 
    SM3         1998 
    CPU          1007  
    1             21.04   
    2              9.80 
    3              0.42 
    4              0.38  
    5             25.60  
    6             14.37  

    So we now know that the 64-bit ultimate runs faster than the 32-bit Enterprise on a 2GB 2 core VM.  Note: all results are the best results of several runs.  I even shutdown vm2 midway through this test with no change in results.  Thus we also now know there is no perfomance penalty for idle remotefx vms.

    Run at the same time:

                vm1/pc1   vm2/pc2

    total        2884       2662
    SM2         1233       1102
    SM3         1276       1179
    CPU           793         773

    1             13.06       12.33
    2              7.48        6.05
    3               0.34       0.33
    4               0.30       0.29
    5             14.85       13.86
    6             10.66        9.72

    If you add the results together you get 5546, which when compared to the average individual scores, says that you can get the single fxremote equivilent of 1.31 vms when extensively using 2 vms.  Is it cpu limited? The ratio there is 1.57.  It will be interesting when we hopefully look at faster GPus or the Nvidia GPUs.

    Saturday, November 20, 2010 9:57 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    I tested the vm1/pc1 with a much slower card, an ATI 9600SE at PC1.  The RemoteFX was using the 3870.
                   

     

    3870

    		            

     

    vm1/pc1

    		            

     

    		              

    total

    4365

    		            

    SM2

    2056

    		            

    SM3

    2130

    		            

    CPU

    900

    		            

     

    		              

    1

    23.34

    		            

    2

    10.93

    		            

    3

    0.34

    		            

    4

    0.38

    		            

    5

    28.13

    		            

    6

    14.47

    		            
                   

    While the stats look reasonable, the appearance was choppy, with a significant number of frames being dropped at rapidly changing screens.

    This card is last supported by the 10.2 catalyst drivers (8.593).  The 8.56.1.16 Microsoft WDDM 4/5/2009 drivers produced a D3DERR_NotAvailible error.

    Because of this performance and driver support, the low end PC video card for a PC displaying RemoteFX is likely an ATI 2400 (PCI-E x16 or AGP 4/8x but not PCI).

     

    Thursday, December 2, 2010 4:30 AM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    Next I tried an ATI HD 5770 with the ATI HD 3780, but the results did not change, thus the ambiguous wording about the video cards for RemoteFX must match really means all video cards in the computer must match.  (It was worth a try - servers typically don't need higher end video, so it would have been nice to be able to dedicate the higher video cards to only RemoteFX rather than the server and RemoteFX.)  To what exact degree the video cards must  match remains uncertain, ie 4850/4870 both use the same driver, can the memory size differ, how about the core and memory speeds? etc.

    3dmark06 tested at different times using the 5770 with 3870 left for comparison:                                                                                                             

                   3870       3870                      5770       5770   %5770/3870   %5770/3870     

                   vm1/pc1 vm2/pc2                vm1/pc1 vm2/pc2 vm1/pc1 vm2/pc2

    total        4476       4000                      4774       4390       107%      110%     

    SM2        1992       1712                      2175       1943       109%      113%     

    SM3        2163       1876                      2380       2163       110%      115%     

    CPU        1005       991                        997         964         99%        97%       

                                                                                                                           

    1             22.73      19.63                     25.02      21.94                                   

    2             10.47      8.91                       11.23      10.44                                   

    3             0.42        0.41                       0.42        0.39                                     

    4             0.39        0.38                       0.38        0.38                                     

    5             28.49      24.47                     33.75      29.41                                   

    6             14.76      13.05                     13.85      13.86                                   

                                                                                                                           

    Run at the same time:                                                                                                                   

                   3870       3870       3870                      5770       5770       5770       Total

                   vm1/pc1 vm2/pc2 Total                      vm1/pc1 vm2/pc2 Total       %

    total        2884       2662       5546                      3320       3163       6483       117%

    SM2        1233       1102       2335                      1374       1372       2746       118%

    SM3        1276       1179       2455                      1634       1448       3082       126%

    CPU        793         773         1566                      798         800         1598       102%

                                                                                                                           

    1             13.06      12.33                                    15.94      15.45                    

    2             7.48        6.05                                      6.97        7.41                      

    3             0.34        0.33                                      0.34        0.34                      

    4             0.3          0.29                                      0.3          0.3                        

    5             14.85      13.86                                    21.88      18.67                    

    6             10.66      9.72                                      10.79      10.29                    

     

     

     

     

     

     

     

     

     

    If you add the results together you get 6483, which when compared to the average individual scores (4482), says that you can get the single RemoteFX equivalent of 1.41 vms with a 1GB 5770 (compared to 1.31 vms for the 512mb 3870) when extensively using 2 vms.  Thus it is most likely better to have more than 512MB for two RemoteFX vms set at 1920 x 1200 max (but using 1280 x 1024).

     

    The 3dmark06 score difference is 3100 versus 4774 when comparing the two 5770 systems.  Serisman reported a 5770 and vm with two 1920 x1200 monitors and the server with a 1.9ghz Opteron x4 with registered RAM, and a 5770  with 850 core 1200 memory.  Mine is a 2.9ghz Phenom x4 (only two memory channels) with a 5770 at 800core 1300 memory plus it has an optimized video chipset (790FX). 

     

    So I did a quick test of increasing the monitors to dual and rerunning the 3DMark06 test - about 75 points higher.  Opposite of what I thought, which was the same or a performance drop. 

     

    Thursday, December 2, 2010 6:54 AM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    Host is dual socket AMD Opteron 8347 HE (quad-core 1.9 GHz) with 20 GB (4x4gb + 4x1gb) DDR2 Registered ECC RAM.

    VM1 is Windows 7 Ultimate x64 with 2 virtual CPUs and dynamic memory (min 2048: max: 4096) and RemoteFX with 2x 1920x1200 monitors

    Both with SP1 RC

     

    1) ATI Radeon HD 5770 1GB (850 MHz GPU / 1200 MHz Memory)

    2)

      Host VM1
    3DMarks Score: 8447 3100
    SM2.0 Score: 3464 1504
    HDR/SM3.0 Score: 3465 1565
    CPU Score: 2964 594
    GT1 - Return To Proxycon: 27.72 (FPS) 17.59 (FPS)
    GT2 - Firefly Forest: 30.00 (FPS) 7.48 (FPS)
    CPU1 - Red Valley: 1.08 (FPS) 0.25 (FPS)
    CPU2 - Red Valley: 1.30 (FPS) 0.23 (FPS)
    HDR1 - Canyon Flight: 48.08 (FPS) 21.61 (FPS)
    HDR2 - Deep Freeze: 21.22 (FPS) 9.68 (FPS)

    These are all lower than I would have expected (even on the host), so I may have some other issue going on here.


    Does the type of RemoteFX computer make a difference?  How important is processor speed?  Since I am on my last planned ATI test., I decided to match Serimans processor speed and graphics core speed and graphics memory speed.  Here are the results:

             vm2/pc2

    Total        3146
    SM2         1463 
    SM3         1367 
    CPU          773  
    1             16.52   
    2              7.86 
    3              0.30 
    4              0.31  
    5             16.63  
    6             10.72

    Thus just comparing the 3dmark06 totals indicates that with 1 ATI graphics card there is almost no difference with the same CPU speed, etc. A more careful look reveals that my cpu scrore was higher and the graphics scrores were lower.  Reruns scored lower, most likely due to other activity on the server even though it is really dedicated to this test.  Thus for computers with ATI cards CPU speed is most important, followed by graphics memory size for two RemoteFX VMs (should be grater than 512MB), followed by graphics card model.

    The rules for nvidia card configuration may differ.  If you look at folding@home, the ATI results are lower and use more cpu than nvidia with cuda. 

    Thursday, December 2, 2010 4:08 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    HELP AND I AND TRIED AND TO AND INSTALL AND HD AND 5770 AND I AND AM AND RECIEVING AND EVENT AND ERRORS

    GARDEALAN@AOL.COM AND 5615966165

    Thursday, February 10, 2011 4:22 AM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    I was out of town for a few days.  I will need more details from you to provide assistance.  My first requirement is that you be usings windows server 2008 R2 SP1 RC and have hyper-V installed with windows 7 SP1 RC and have another PC with windows 7 SP1 RC installed.  You must also list the event errors.  You will likely get a quicker response with more people resonding if you create your own thread in this forum, since this thread is more focued on RemoteFX performance.  Also be awara that SP1 has gone RTM, which will result in improved official support.

    Thanks

    Monday, February 14, 2011 4:07 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote
    I did test this with a Nvidia 8800GT and 8800GT x 2, but one card went bad.  The results were slightly slower than the ATI for one card. With 2 8800Gts, it did appear to be using both cards when two PCs were used.  I report this without detailed results since SP1 has gone RTM just to let others known it does work
    Monday, February 14, 2011 4:12 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote
    We want to build a HyperV cluster that can hold our 25 Workstations.

    But use the expensive cards quadro fx 4800/5800, will cut some of ROI

    So Would it be possiable to use 2 or more of these types of graphic cards ?


    Wednesday, February 16, 2011 2:48 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    I have not tested that specific graphics card, but see no objections. They don't seem to be using CUDA, thus Nvidia did not have any advantage.  Also don't forget that directx 11 is not used.

    I just got my release to manufacturing installed last night, thus have not had time to test.  It does seem faster than the RC is my initial impression. 

    I do urge you to do or wait for more research.  Here are factors I hope you are considering:

    1) Your user mix.  25 managers who spend alot of time in meetings will have a far different impact than 25 engineers constantly pounding away using autocad inventor in 3d.

    2) Your server's mainboard PCIE video slots.  How many and what speed.  16x for all is ideal.  Some at 8x is okay.  1x is unacceptable.  SLI and crossfire are not used.  I have not tested beyond 2 (lack of identical video cards).

    3) Speed of the processor.  The sweet spot for most graphics cards that is often mentioned is 3.5ghz.  If you look at the results above in this thread you will see that this speed has a major impact.  This threshold could be less in the production version (retesting is needed).

    4) Previously, dynamic memory use appeared to have a negative impact on remoteFX, especially when set to below the OS minimum memory requirement.  This needs to be retested.

    5) Don't forget licensing.  You have remoteFX licensing and/or remote desktop licensing plus your virtual PCs.  This needs to be fully examined with the production version.  Unfortunately all documentation is not yet available (most is RC). 

    I hope this is helpful.

     

    Thursday, February 17, 2011 1:35 PM
  • Discussion
    Sign in to vote
    0
    Sign in to vote

    Hi guys.  I'll keep this short. I'm putting together a RemoteFX capable box right now.  I'm looking for a card in the 150 - 200 range thaht will give me desktop like performance for 1 or 2 VMs.  System will be as follows:

    Mobo: Asus P7P55D-E Pro
    8GB PC3 12800 Kingston Hyper X RAM
    CPU: Core i5-760 Quad

    Suggestions for RemoteFX capable video card in that range?  Any help appreciated.  Thanks!

    EDIT: I prefer Nvidia Card suggestion.  I currently own an Nvidia 8800 GTS 320mb but I have a feeling that's not going to be good enough....

     

    Thanks!

    Tuesday, February 22, 2011 2:43 AM