Announcement

Collapse
No announcement yet.

NVIDIA Tegra K1 Compared To AMD AM1 APUs

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #61
    Originally posted by deppman View Post
    AMD representatives first claimed that the R9 290x card was 250W "in typical gaming scenarios" as reported by Anandtech
    It also beats all the recent AMD AM1 25W TDP chips while pushing 2.5x the number of pixels (1920x1080 vs 1024x768)!
    As i said i have Athlon 5350 and compared it in many situations for me, it never draws that much power - nope ... AM1 and processors, they are just marked at 25W TDP level that is per design, and that does not mean processors will use that much power, they may (maybe if you overclock and do up voltage manually or something) but they not ;D... actualy during intensive gaming Athlon 5350 uses max 16 Watts and that also overclocked by 550MHz at 2600 MHz . Actually i runned it that much overclocked on Windows... and that with great stress testing i can do = prime+playing a game at the same time so 16W max .

    As per resoultions you say 1920x1080 vs 1024x768... there is also not much a difference when resoultion goes up if you have hyperz enabled with radeon driver... maybe it is 5% to max 10% fps drop difference . Of course blobs are another story
    Last edited by dungeon; 11 May 2014, 08:53 PM.

    Comment


    • #62
      xononic high and ultra scores

      On "low" detail settings, the TK1 runs 83FPS average compared to 63FPS on the J1900 @ 2.9GHz, or about 32% faster. While the J1900 claims 10W TDP @ 2.0 GHz, I'm guessing we are seeing more than that for a 33% boost in clock (13.3W?). The TK1 shows < 5W for SOC + RAM. It also beats all the recent AMD AM1 25W TDP chips while pushing 2.5x the number of pixels (1920x1080 vs 1024x768)! Assuming these tests are comparable to the Phoronix setup, that's very impressive!
      I had to shut down Chromium and KDE to get the memory needed, but here are the benchmarks for high and ultra settings. These were found using the-big-benchmark timed demos and xinit. These are nearly double the J1900 results.

      High: 17/42/60 - low/avg/high
      Ultra: 06/29/47 - low/avg/high

      High detail is very playable. Ultra, not so much. Maybe at 720p. I think another 1GB would provide some much needed headroom, and might improve these scores. I'm done benchmarking at least for the day.

      Since some folks don't think that dropping to 720p (40% of the pixels) makes much difference, I might post those results for Mom. But that's it for tonight.

      Comment


      • #63
        Originally posted by deppman View Post

        Since some folks don't think that dropping to 720p (40% of the pixels) makes much difference, I might post those results for Mom. But that's it for tonight.
        When i benchmarking Xonotic at default (normal) i do not see much difference using radeon driver with hyperz enabled... 1080p vs 720p it is only 8% fps results difference .

        Comment


        • #64
          Actually ... you're right! (Crow served here)

          Originally posted by dungeon View Post
          When i benchmarking Xonotic at default (normal) i do not see much difference using radeon driver with hyperz enabled... 1080p vs 720p it is only 8% fps results difference .
          Well, I was wrong. I dropped the resolution to 1366x720 (or something like that) because my display or the Tegra driver couldn't handle 1024x768. And ... nothing. Frame rates remained almost exactly the same. Which I found odd since most GL benchmarks seem to really be sensitive to resolution. So there, I ate some crow :P

          I also pushed all the settings to the max - dynamic lighting, shadows, relief maps, etc. One would expect this to be the Phoronix "Ultimate" setting. The result was 19FPS average - which is over double that shown for the J1900. That makes sense: NV's hardware and 4.4 OGL driver should show better for more sophisticated features. So in that sense, I guess it doesn't disappoint, and it's not that much slower than "Ultra" settings. But ... both Ultra and Ultimate are unplayable. "High" still looks great though, and plays quite well; "Normal" is buttery smooth. I might experiment to see if some of the more "advanced" features come at a lower cost than normal. One example is relief maps - a.k.a. tessellation. NV supports HW acceleration in OGL 4.4; its an open question if Xonotic can take advantage of it.

          I don't much like these types of games but was very interested in seeing the power draw and performance for a typical demanding use case. The source, assets, and compiled files take up 8.7GB on the SD card!

          Comment


          • #65
            Originally posted by deppman View Post
            Tests are in!

            Charlie needs to write a retraction. His pseudo-analysis looks to be off by around 250%, or 20-30W.

            While these results are far from exhaustive, here are some highlights and observations:
            • I have yet to observe > 12W total system draw under any workload.
            • I confirmed the fan's power draw by unplugging it and noting the difference in amps.
            • Nvidia's numbers found in their technical brief (page 13) appear accurate to conservative.
            • Nvidia's point about drawing comparisons to mobile are valid. Notice we are using a bevy of desktop-level ports and peripherals (e.g. HDMI out, GbE, RAM, etc) instead of low power mobile alternatives.
            • The base system does indeed appear to draw about 2W with the processor running at 0.66W at idle.
            • As load ramps, the increased power draw comes from many components, not just the SOC.
            • In conclusion, this looks like a fine tablet or steambox chip.


            Code:
            FULL SYSTEM JETSON TK1 DC POWER ANALYSIS
              Includes
                * Audio in/out Active
                * Gigabit Ethernet
                * USB 3 driving Keyboard, Mouse, Logitech C210 Webcam
                * HDMI out at 1920x1080
                * Cooling Fan (no heat sink)
                * Installed 64GB SD card
            
            BASE MEASUREMENTS
              Power-in Voltage : 12.15V
              Amperage and wattage
                Idle KDE Desktop : 0.22A ( 2.67W)
                Fan Amp          : 0.07A ( 0.85W)
                System Amp*      : 0.17A ( 2.06W) *NV measurement
            
                Idle Less Fan    : 0.15A ( 1.82W)
                Idle Less System : 0.05A ( 0.61W)
            
            TEST1 : glmark2 -s 1920x1080 --off-screen
              Score : 282 (Intel Celeron [email protected] = 151)
              Power Measurements:
                Base             : 0.22A ( 2.67W)
                Peak             : 0.62A ( 7.53W)
                Observed Avg.    : 0.35A ( 4.25W)
            
                Avg. Less Fan    : 0.28A ( 3.40W)
                Avg. Less Sys    : 0.18A ( 2.19W)
            
            TEST2 : CUDA Smoke particle demo
              Power Measurements:
                Base             : 0.62A ( 7.53W)
                Peak             : 0.91A (11.06W)
                Observed Avg.    : 0.88A (10.69W)
            
                Avg. Less Fan    : 0.81A ( 9.85W)     
                Avg. Less Sys    : 0.71A ( 8.26W)
            
            TEST3 : VLC streaming 720p video from NAS GbE
              Power Measurements:
                Base             : 0.29A ( 3.52W)
                Peak             : 0.41A ( 4.98W)
                Observed Avg.    : 0.34A ( 4.13W)
            
                Avg. Less Fan    : 0.27A ( 3.28W)     
                Avg. Less Sys    : 0.17A ( 2.01W)
            Sorry, I forgot about this. Michaels recent p state article reminded me.
            First, thanks for doing this.
            The numbers look good, if still too high for a tablet. Have you tried unloading all of the drivers you don't need? The gige driver, in particular, can consume way too much power doing nothing on my laptop (for some reason autosuspend doesn't guarantee it staying off when not in use). In addition, there are the audio codec, which aren't needed for some benchmarks, sata connector, USB, etc.
            I'm curious about the CPU performance compared to that celeron. Is expect it to very lower, say half, for non highly threaded loads, but I'm not sure of the subsequent power draw.

            Comment


            • #66
              Originally posted by liam View Post
              Sorry, I forgot about this. Michaels recent p state article reminded me.
              First, thanks for doing this.
              The numbers look good, if still too high for a tablet. Have you tried unloading all of the drivers you don't need? The gige driver, in particular, can consume way too much power doing nothing on my laptop (for some reason autosuspend doesn't guarantee it staying off when not in use). In addition, there are the audio codec, which aren't needed for some benchmarks, sata connector, USB, etc.
              I'm curious about the CPU performance compared to that celeron. Is expect it to very lower, say half, for non highly threaded loads, but I'm not sure of the subsequent power draw.
              Hey, you're welcome!

              I have not tried to disable the drivers, but thanks for the advice. Perhaps this weekend when tinkering I can try it out. Some of the benchmarks I made required GbE, but most didn't.

              As for the power draw, the SoC + RAM when used for most applications ranged from 0.6W (idle) to 4.7W (intense 3D gaming). I don't think frequency scaling is implemented here either. So in a tablet with mixed use (browsing, gaming, reading) average draw might be 2W or so. That seems perfectly acceptable for a tablet, but I don't claim to be an EE.

              The only time the SoC + RAM averaged more than 5W was when it was "lighting up" all CUDA cores with the smoke particle test. Unless folks want to use their tablets for folding or other massively parallel tasks, I don't see that as a typical workload

              I don't know about you, but I really want the new Xiaomi tablet. For $270 for 64GB, it's almost an impulse buy. Hmmm, maybe I *will* try folding...

              Comment


              • #67
                A small be aware, if you didn't have a look at 1st or even didn't think to seem... This is contrasting closed-source Nvida as opposed to open up source Radeon-- not really Switch. And so overall performance about the AMD part is not going to end up being 'full components potential.

                Comment


                • #68
                  Is there Catalyst support for AM1 chips?

                  Originally posted by nengtyas View Post
                  A small be aware, if you didn't have a look at 1st or even didn't think to seem... This is contrasting closed-source Nvida as opposed to open up source Radeon-- not really Switch. And so overall performance about the AMD part is not going to end up being 'full components potential.
                  It there a stable, reliable driver available for Linux that does take advantage of the full component potential? In typical AMD fashion, it doesn't appear there is. Therefore this appears to be a comparison of best available configurations.

                  Typical AMD - decent hardware with substandard drivers. The problem exists on Windows too, but it is magnified on Linux.

                  Comment


                  • #69
                    Originally posted by deppman View Post
                    It there a stable, reliable driver available for Linux that does take advantage of the full component potential? In typical AMD fashion, it doesn't appear there is. Therefore this appears to be a comparison of best available configurations.

                    Typical AMD - decent hardware with substandard drivers. The problem exists on Windows too, but it is magnified on Linux.
                    Catalyst driver for Athlon 5350 was there on release date for both Windows and Linux . On Linux there was also support in opensource radeon driver on AM1 release date. Athlon 5350 is just socketed Kabini for Desktop variant nothing more , chip is introduced year before as embeded/ultra-mobile variant A6-5200 APU so initial support for those Kabinis was added 10 months before AM1 launch in opensource driver.

                    There was really no and not any major problem for fglrx nor opensource radeon driver to supports AM1 platform on launch date .
                    Last edited by dungeon; 23 June 2014, 09:21 PM.

                    Comment

                    Working...
                    X