AMD Ryzen AI 9 HX 370: 100+ Benchmarks Validate Zen 5's Captivating Power Efficiency & Performance

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

  • KDE_FOREVER
    replied
    Originally posted by sophisticles View Post

    I think Intel removed HT on these chips so that they could have an easy way to increase performance with the chips that replace LL.

    Thanks to architectural improvements they have produced increased efficiency and performance that will be appealing without the need for HT.

    With future versions where they need to add performance relatively easily, they can just add HT back and instant performance improvement.
    you started with "I think" and that is a lie right there

    Leave a comment:


  • KDE_FOREVER
    replied
    Originally posted by sophisticles View Post
    Yawn!

    As already pointed out, the performance is not that great when you consider 12C/24T for the new chips vs 8C/16T for the older chips, gee, I wonder which one is going to be faster?

    The power consumption is a nice improvement, very impressive, though I expect Lunar Lake to take these chips to school in that department.

    Overall, it looks like a hard pass is in order for these chips from Linux users perspective, at least until Linux support is up to snuff.


    squeal more maggot.

    Amd wipes the floor with intel

    Intel announced firing of almost 18000 people​

    Leave a comment:


  • robclark
    replied
    Originally posted by smitty3268 View Post

    To save people having to manually look up the scores to compare:

    7zip compression:
    qualcomm x1e-78: 86,318
    amd 365: 86,059
    amd 370: 90,275
    amd 7840HS: 81,522

    quantlib size xxs:
    qualcomm x1e-78: 6.28025
    amd 365: 5.38639
    amd 370: 6.01127
    amd 7840HS: 6.25762​
    I suspect this result might be power limited on the new chips, given the higher zen 4 score. I believe Qualcomm allows their chips to use quite a bit of power on multi-threaded code.
    so I fumbled my way around building up a test suite, but several didn't have aarch64 (looks like PTS is trying to download an x86 binary?), and TesorFlow installed but failed to run (looks like I'm missing some py module, I'll have to try again tonight). PTS bailed out without uploading results when TensorFlow failed to run, or maybe I did something wrong. But here are the results of what did run:
    Code:
    GraphicsMagick 1.3.43:
        pts/graphics-magick-2.2.0 [Operation: HWB Color Space]
        Test 2 of 4
        Estimated Trial Run Count:    3                              
        Estimated Test Run-Time:      4 Minutes                      
        Estimated Time To Completion: 1 Hour, 14 Minutes [04:00 UTC]
            Started Run 1 @ 02:47:26
            Started Run 2 @ 02:48:30
            Started Run 3 @ 02:49:35
    
        Operation: HWB Color Space:
            196
            202
            200
    
        Average: 199 Iterations Per Minute
        Deviation: 1.53%
    
        Comparison of 524 OpenBenchmarking.org samples since 27 March; median result: 245 Iterations Per Minute. Box plot of samples:
        [    |-------------------------------------*########*!##*#*#---*-------------------------------------------------------------------*-----|    ]
                                                   ^ This Result (22nd Percentile): 199
                                         INTEL XEON PLATINUM 8592: 294 ^                                            AMD Ryzen 9 5950X: 609 ^
                                   Intel Xeon Platinum 8490H: 269 ^
                              2 x INTEL XEON PLATINUM 8592: 262 ^
                          2 x Intel Xeon Platinum 8468: 244 ^
    
    Timed LLVM Compilation 16.0:
        pts/build-llvm-1.5.0 [Build System: Ninja]
        Test 3 of 4
        Estimated Trial Run Count:    3                              
        Estimated Test Run-Time:      58 Minutes                    
        Estimated Time To Completion: 1 Hour, 11 Minutes [04:01 UTC]
            Running Pre-Test Script @ 02:50:45
            Started Run 1 @ 02:51:02
            Running Interim Test Script @ 03:02:06
            Started Run 2 @ 03:02:28
            Running Interim Test Script @ 03:13:50
            Started Run 3 @ 03:14:11
            Running Post-Test Script @ 03:25:29
    
        Build System: Ninja:
            661.87
            680.207
            675.523
    
        Average: 672.533 Seconds
        Deviation: 1.42%
    
        Comparison of 1,518 OpenBenchmarking.org samples since 22 March 2023; median result: 326 Seconds. Box plot of samples:
        [                                          *            |----------------------------*--*---------------*-----------------*--*-###*#####!##*|*]
                                                                                                       This Result (27th Percentile): 673 ^
                         Intel Core i5-6260U: 5340 ^                  Intel Core i7-8565U: 2966 ^                  2 x INTEL XEON PLATINUM 8592: 101 ^
                                                                                     Intel Core i7-1185G7: 2093 ^   Intel Xeon Platinum 8471N: 209 ^
                                                                  Intel Core i7-1065G7: 3077 ^            AMD Ryzen 7 PRO 7840U: 940 ^
                                                                                                          AMD Ryzen 7 6800U: 1098 ^
    
    ​
    idk what the state of GraphicsMagick is as far as neon support, it looks like it could use some optimization.

    Leave a comment:


  • smitty3268
    replied
    Originally posted by robclark View Post

    I figured out how to upload the other couple benchmark runs I did:
    To save people having to manually look up the scores to compare:

    7zip compression:
    qualcomm x1e-78: 86,318
    amd 365: 86,059
    amd 370: 90,275
    amd 7840HS: 81,522

    quantlib size xxs:
    qualcomm x1e-78: 6.28025
    amd 365: 5.38639
    amd 370: 6.01127
    amd 7840HS: 6.25762​
    I suspect this result might be power limited on the new chips, given the higher zen 4 score. I believe Qualcomm allows their chips to use quite a bit of power on multi-threaded code.
    Last edited by smitty3268; 30 July 2024, 03:29 AM.

    Leave a comment:


  • mahirzukic
    replied
    Originally posted by AdrianBc View Post



    For a personal computer that you use for reading and writing documents, for Internet browsing, for watching movies and other similar activities, it makes perfect sense to use a monolithic CPU with integrated GPU, e.g. by using a NUC-like small computer, which may have an idle power consumption of 5 W or even less.

    On the other hand, if you use a computer for solving any problem that requires a lot of computations, a non-monolithic CPU, like a desktop Ryzen, will provide a speed several times higher than any low-power CPU and even its energy efficiency can be greater, despite the higher idle power consumption, because when all cores are 100% busy for a long time, the idle power consumption will form only a small part of the total power consumption and the desktop CPUs can have better energy efficiency than the laptop CPUs for the part that exceeds the idle power consumption, e.g. due to bigger cache memories or due to wider SIMD execution units.

    This is actually how I work. I use an Intel NUC as my PC, but I also have a few servers, including with Ryzen CPUs. Whenever I need to execute a program for which my PC would be too weak, I launch it on one or more servers, by using Wake-on-LAN. When the task is completed, the servers are shut down. Thus the average electrical power consumption is much smaller than that of a big desktop, while the peak performance is much greater than that of a laptop or SFF computer.

    There is a reason for the existence of many kinds of CPUs, because each of them is the best for certain applications.



    Yeah, but the difference of 5w to 23w from the wall is negligible.

    Leave a comment:


  • zaps166
    replied
    Maybe Tuxedo will start selling their 14" laptops with this CPU instead Ryzen 7 8845HS? It'd be a nice improvement (also for GPU).

    Leave a comment:


  • AdrianBc
    replied
    Originally posted by Firnefex View Post

    I don't know the details, but if your description is exactly how 7000 behaves, then my explanation might be off, but the reason against it stays the same. Therefore I never want to buy nonmonolithic CPUs.


    For a personal computer that you use for reading and writing documents, for Internet browsing, for watching movies and other similar activities, it makes perfect sense to use a monolithic CPU with integrated GPU, e.g. by using a NUC-like small computer, which may have an idle power consumption of 5 W or even less.

    On the other hand, if you use a computer for solving any problem that requires a lot of computations, a non-monolithic CPU, like a desktop Ryzen, will provide a speed several times higher than any low-power CPU and even its energy efficiency can be greater, despite the higher idle power consumption, because when all cores are 100% busy for a long time, the idle power consumption will form only a small part of the total power consumption and the desktop CPUs can have better energy efficiency than the laptop CPUs for the part that exceeds the idle power consumption, e.g. due to bigger cache memories or due to wider SIMD execution units.

    This is actually how I work. I use an Intel NUC as my PC, but I also have a few servers, including with Ryzen CPUs. Whenever I need to execute a program for which my PC would be too weak, I launch it on one or more servers, by using Wake-on-LAN. When the task is completed, the servers are shut down. Thus the average electrical power consumption is much smaller than that of a big desktop, while the peak performance is much greater than that of a laptop or SFF computer.

    There is a reason for the existence of many kinds of CPUs, because each of them is the best for certain applications.




    Last edited by AdrianBc; 29 July 2024, 12:16 PM.

    Leave a comment:


  • Anux
    replied
    Originally posted by Firnefex View Post
    but the reason against it stays the same. Therefore I never want to buy nonmonolithic CPUs.
    Of course, if you somehow need lowest idle but need highest performance and don't care about max power then Intel is your only choice. Else you can go with AMD's APUs (monolithic chips).
    My 5700g idles at ~ 23W on the wall. I prefer silent and efficient PCs.

    Leave a comment:


  • Michael
    replied
    Originally posted by Coreyfro View Post
    It seems you have all the information necessary for performance per watt but didn't take the necessary step to calculate it per benchmark. I think that is the most useful information given this is a laptop CPU. Would have loved to have seen Apple CPU's where possible, too. Last generation, AMD Flagship CPU's just beat out Apples Flagship's by a hair in power/performance in some benchmarks. This is where the action is!

    Also. Your popups are.making it impossible to write comments. Please consider making it so they don't pop up over text forms. I blame any typos on this! ;-)
    ​​​
    See the OB link at end of article if wanting to see the perf-per-Watt for each and every benchmark... it's there.

    Unfortunately no review samples / budget for always testing Apple.

    Leave a comment:


  • Coreyfro
    replied
    It seems you have all the information necessary for performance per watt but didn't take the necessary step to calculate it per benchmark. I think that is the most useful information given this is a laptop CPU. Would have loved to have seen Apple CPU's where possible, too. Last generation, AMD Flagship CPU's just beat out Apples Flagship's by a hair in power/performance in some benchmarks. This is where the action is!

    Also. Your popups are.making it impossible to write comments. Please consider making it so they don't pop up over text forms. I blame any typos on this! ;-)
    ​​​

    Leave a comment:

Working...
X