Alleged 'A12' benchmark for 2018 iPhone with 4GB RAM pops up

13

Comments

  • Reply 41 of 77
    tmaytmay Posts: 3,545member
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    You took that out of context:

    "When we’re looking at competitor devices we see only the the iPhone X able to compete with the last generation Snapdragon 835 devices – however with a catch. The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle. Unfortunately this also applies to current and last generation Exynos and Kirin SoCs as both shed great amount of performance after only a few minutes. I’ve addressed this issue and made a great rant about it in our review of the Kirin 970. For this reason going forward AnandTech is going to distinguish between Peak and Sustained scores across all 3D benchmarks. This however needs to be tested on commercial devices as the QRD platform isn’t a thermally representative phone for the SoC, so until that happens, we’ll have to just estimate based on power consumption where the Snapdragon 845 ends up.

    This article was a preview on the Snapdragon 845, not a device test.

    Show me a device with a Snapdragon 845, in production, that does better than the A11 and doesn't throttle, and I'll give you a "like".

     
    I was never talking about the 845.

    The Snapdragon 835 does better and doesn't throttle nearly as much.

    For the Pixel 2 using OpenGL ES 3.1 API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2907
    Sustained: 2864

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4390
    Sustained: 3591
    Okay, but what about all of the other tests that the A11 scored better than the 835?
    We don't have sustained performance tests for most other tests, and for the ones that we do, the A11 throttles hard.

    Hence, we circle back to my original post. Geekbench uses pauses to prevent thermal throttling, so all we see is the peak performance, and not the sustained performance.

    Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle.
    So, you're confident that a couple of synthetic tests of a single aspect of an SOC, the GPU, are a reliable means to predicting the performance of a SOC in general use? 

    Yes, the physics test is CPU bound, the graphics test is GPU bound. Regardless of the results compared to other SoCs, it's very easy to see the difference between peak and sustained performance on the A11.

    That, and we also know each one of the A11's big cores uses 3~3.5 W at load. Having all 6 cores at load (2 big and 4 small) would simply not be sustainable in an iPhone. If the A12 brought that TDP down, with potential adjustments to architecture and the move to a 7nm process, we'll likely see better sustained performance.
    You keep speaking of this, but you haven't actually showed this "difference between peak and sustained performance".  I'm not doubting you, though I presume that you have links?

    "If the A12 brought the TDP down":

    Why would Apple want a lower TDP if for certain workloads, maximum performance, even with throttling, was the desired result? In fact, Apple specifically spoke of this at WWDC, wrt a very rapid ramp of the CPU (that's an instantaneously high TDP!) followed by a more gradual reduction in clock rate and then a shift to the low power cores. I might be incorrect on those details, but I specifically remember the slide.

    Here's an example of a "peak performance" scenario, with all it's bits and pieces. For this hybrid, let's just deem braking an analog of throttling, and boost, that's the battery kicking in those electric motors...


    edited July 2018
  • Reply 42 of 77
    mcdavemcdave Posts: 1,114member
    bsimpsen said:
    At 8-11% faster for the CPU, I think this would be the smallest generational improvement in the A-family since its introduction. Eventually Apple will run up against the wall or simply not need much additional horsepower, but I was expecting bigger gains, particularly in light of the rumored ditching of x86 in the next couple years.
    Or just stop the pissing contest with Android vendors and build fit-for-purposes APUs.  It always made sense to let the iPad Pro take the lead.  Why does a phone need more CPU performance?  I’d rather see more custom silicon units (though Geekbench iOS should keep up).

    I’m liking the GPU figures; +40% compute. If they’ve kept the core count the same that’s awesome news.  I just wish they would put the iPad Pro power budget back up to 12 or even 16-cores; that way they could compete/beat Kaby-Lake G/ Vega M.

    Disappointing scores (would like the to have overtaken i7-8550u) but may be a good sign of a change of emphasis.
    tmay
  • Reply 43 of 77
    KITAKITA Posts: 163member
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    You took that out of context:

    "When we’re looking at competitor devices we see only the the iPhone X able to compete with the last generation Snapdragon 835 devices – however with a catch. The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle. Unfortunately this also applies to current and last generation Exynos and Kirin SoCs as both shed great amount of performance after only a few minutes. I’ve addressed this issue and made a great rant about it in our review of the Kirin 970. For this reason going forward AnandTech is going to distinguish between Peak and Sustained scores across all 3D benchmarks. This however needs to be tested on commercial devices as the QRD platform isn’t a thermally representative phone for the SoC, so until that happens, we’ll have to just estimate based on power consumption where the Snapdragon 845 ends up.

    This article was a preview on the Snapdragon 845, not a device test.

    Show me a device with a Snapdragon 845, in production, that does better than the A11 and doesn't throttle, and I'll give you a "like".

     
    I was never talking about the 845.

    The Snapdragon 835 does better and doesn't throttle nearly as much.

    For the Pixel 2 using OpenGL ES 3.1 API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2907
    Sustained: 2864

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4390
    Sustained: 3591
    Okay, but what about all of the other tests that the A11 scored better than the 835?
    We don't have sustained performance tests for most other tests, and for the ones that we do, the A11 throttles hard.

    Hence, we circle back to my original post. Geekbench uses pauses to prevent thermal throttling, so all we see is the peak performance, and not the sustained performance.

    Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle.
    So, you're confident that a couple of synthetic tests of a single aspect of an SOC, the GPU, are a reliable means to predicting the performance of a SOC in general use? 

    Yes, the physics test is CPU bound, the graphics test is GPU bound. Regardless of the results compared to other SoCs, it's very easy to see the difference between peak and sustained performance on the A11.

    That, and we also know each one of the A11's big cores uses 3~3.5 W at load. Having all 6 cores at load (2 big and 4 small) would simply not be sustainable in an iPhone. If the A12 brought that TDP down, with potential adjustments to architecture and the move to a 7nm process, we'll likely see better sustained performance.
    You keep speaking of this, but you haven't actually showed this "difference between peak and sustained performance".  I'm not doubting you, though I presume that you have links?

    "If the A12 brought the TDP down":

    Why would Apple want a lower TDP if for certain workloads, maximum performance, even with throttling, was the desired result? In fact, Apple specifically spoke of this at WWDC, wrt a very rapid ramp of the CPU (that's an instantaneously high TDP!) followed by a more gradual reduction in clock rate and then a shift to the low power cores. I might be incorrect on those details, but I specifically remember the slide.
    I everything I've been talking about is from Anandtech.





    Anandtech

    If these iPhone 11,2 numbers are accurate, Apple might not be pushing the peak performance this iteration and want to ensure sustained performance and efficiency take priority. It is a mobile device after all. Of course, there are still so many factors that weigh in on all of this. As was mentioned earlier, there's a lot more to a mobile SoC that just the CPU and GPU. There's the DSP, ISP, etc., and now a larger push into dedicated ML hardware (although, this can still fall under other components as was the case for Qualcomm's Hexagon DSP).
  • Reply 44 of 77
    tmaytmay Posts: 3,545member
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    You took that out of context:

    "When we’re looking at competitor devices we see only the the iPhone X able to compete with the last generation Snapdragon 835 devices – however with a catch. The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle. Unfortunately this also applies to current and last generation Exynos and Kirin SoCs as both shed great amount of performance after only a few minutes. I’ve addressed this issue and made a great rant about it in our review of the Kirin 970. For this reason going forward AnandTech is going to distinguish between Peak and Sustained scores across all 3D benchmarks. This however needs to be tested on commercial devices as the QRD platform isn’t a thermally representative phone for the SoC, so until that happens, we’ll have to just estimate based on power consumption where the Snapdragon 845 ends up.

    This article was a preview on the Snapdragon 845, not a device test.

    Show me a device with a Snapdragon 845, in production, that does better than the A11 and doesn't throttle, and I'll give you a "like".

     
    I was never talking about the 845.

    The Snapdragon 835 does better and doesn't throttle nearly as much.

    For the Pixel 2 using OpenGL ES 3.1 API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2907
    Sustained: 2864

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4390
    Sustained: 3591
    Okay, but what about all of the other tests that the A11 scored better than the 835?
    We don't have sustained performance tests for most other tests, and for the ones that we do, the A11 throttles hard.

    Hence, we circle back to my original post. Geekbench uses pauses to prevent thermal throttling, so all we see is the peak performance, and not the sustained performance.

    Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle.
    So, you're confident that a couple of synthetic tests of a single aspect of an SOC, the GPU, are a reliable means to predicting the performance of a SOC in general use? 

    Yes, the physics test is CPU bound, the graphics test is GPU bound. Regardless of the results compared to other SoCs, it's very easy to see the difference between peak and sustained performance on the A11.

    That, and we also know each one of the A11's big cores uses 3~3.5 W at load. Having all 6 cores at load (2 big and 4 small) would simply not be sustainable in an iPhone. If the A12 brought that TDP down, with potential adjustments to architecture and the move to a 7nm process, we'll likely see better sustained performance.
    You keep speaking of this, but you haven't actually showed this "difference between peak and sustained performance".  I'm not doubting you, though I presume that you have links?

    "If the A12 brought the TDP down":

    Why would Apple want a lower TDP if for certain workloads, maximum performance, even with throttling, was the desired result? In fact, Apple specifically spoke of this at WWDC, wrt a very rapid ramp of the CPU (that's an instantaneously high TDP!) followed by a more gradual reduction in clock rate and then a shift to the low power cores. I might be incorrect on those details, but I specifically remember the slide.
    I everything I've been talking about is from Anandtech.





    Anandtech

    If these iPhone 11,2 numbers are accurate, Apple might not be pushing the peak performance this iteration and want to ensure sustained performance and efficiency take priority. It is a mobile device after all. Of course, there are still so many factors that weigh in on all of this. As was mentioned earlier, there's a lot more to a mobile SoC that just the CPU and GPU. There's the DSP, ISP, etc., and now a larger push into dedicated ML hardware (although, this can still fall under other components as was the case for Qualcomm's Hexagon DSP).
    So, your argument is that Apple needs to redesign its A series to throttle less for these two sustained benchmarks?

    Folly, pure and simple.
  • Reply 45 of 77
    KITAKITA Posts: 163member
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    You took that out of context:

    "When we’re looking at competitor devices we see only the the iPhone X able to compete with the last generation Snapdragon 835 devices – however with a catch. The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle. Unfortunately this also applies to current and last generation Exynos and Kirin SoCs as both shed great amount of performance after only a few minutes. I’ve addressed this issue and made a great rant about it in our review of the Kirin 970. For this reason going forward AnandTech is going to distinguish between Peak and Sustained scores across all 3D benchmarks. This however needs to be tested on commercial devices as the QRD platform isn’t a thermally representative phone for the SoC, so until that happens, we’ll have to just estimate based on power consumption where the Snapdragon 845 ends up.

    This article was a preview on the Snapdragon 845, not a device test.

    Show me a device with a Snapdragon 845, in production, that does better than the A11 and doesn't throttle, and I'll give you a "like".

     
    I was never talking about the 845.

    The Snapdragon 835 does better and doesn't throttle nearly as much.

    For the Pixel 2 using OpenGL ES 3.1 API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2907
    Sustained: 2864

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4390
    Sustained: 3591
    Okay, but what about all of the other tests that the A11 scored better than the 835?
    We don't have sustained performance tests for most other tests, and for the ones that we do, the A11 throttles hard.

    Hence, we circle back to my original post. Geekbench uses pauses to prevent thermal throttling, so all we see is the peak performance, and not the sustained performance.

    Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle.
    So, you're confident that a couple of synthetic tests of a single aspect of an SOC, the GPU, are a reliable means to predicting the performance of a SOC in general use? 

    Yes, the physics test is CPU bound, the graphics test is GPU bound. Regardless of the results compared to other SoCs, it's very easy to see the difference between peak and sustained performance on the A11.

    That, and we also know each one of the A11's big cores uses 3~3.5 W at load. Having all 6 cores at load (2 big and 4 small) would simply not be sustainable in an iPhone. If the A12 brought that TDP down, with potential adjustments to architecture and the move to a 7nm process, we'll likely see better sustained performance.
    You keep speaking of this, but you haven't actually showed this "difference between peak and sustained performance".  I'm not doubting you, though I presume that you have links?

    "If the A12 brought the TDP down":

    Why would Apple want a lower TDP if for certain workloads, maximum performance, even with throttling, was the desired result? In fact, Apple specifically spoke of this at WWDC, wrt a very rapid ramp of the CPU (that's an instantaneously high TDP!) followed by a more gradual reduction in clock rate and then a shift to the low power cores. I might be incorrect on those details, but I specifically remember the slide.
    I everything I've been talking about is from Anandtech.





    Anandtech

    If these iPhone 11,2 numbers are accurate, Apple might not be pushing the peak performance this iteration and want to ensure sustained performance and efficiency take priority. It is a mobile device after all. Of course, there are still so many factors that weigh in on all of this. As was mentioned earlier, there's a lot more to a mobile SoC that just the CPU and GPU. There's the DSP, ISP, etc., and now a larger push into dedicated ML hardware (although, this can still fall under other components as was the case for Qualcomm's Hexagon DSP).
    So, your argument is that Apple needs to redesign its A series to throttle less for these two sustained benchmarks?

    Folly, pure and simple.
    Argument? If that's what you see, then folly indeed. I've been only posting facts about the A11, it's ability to sustain load and its known TDP values for the large cores. I haven't suggested that "Apple needs to redesign" it's SoC. Everything is speculation based on the current facts and the apparent benchmark values we've been provided with in Appleinsider's post.
    muthuk_vanalingam
  • Reply 46 of 77
    tmaytmay Posts: 3,545member
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    You took that out of context:

    "When we’re looking at competitor devices we see only the the iPhone X able to compete with the last generation Snapdragon 835 devices – however with a catch. The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle. Unfortunately this also applies to current and last generation Exynos and Kirin SoCs as both shed great amount of performance after only a few minutes. I’ve addressed this issue and made a great rant about it in our review of the Kirin 970. For this reason going forward AnandTech is going to distinguish between Peak and Sustained scores across all 3D benchmarks. This however needs to be tested on commercial devices as the QRD platform isn’t a thermally representative phone for the SoC, so until that happens, we’ll have to just estimate based on power consumption where the Snapdragon 845 ends up.

    This article was a preview on the Snapdragon 845, not a device test.

    Show me a device with a Snapdragon 845, in production, that does better than the A11 and doesn't throttle, and I'll give you a "like".

     
    I was never talking about the 845.

    The Snapdragon 835 does better and doesn't throttle nearly as much.

    For the Pixel 2 using OpenGL ES 3.1 API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2907
    Sustained: 2864

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4390
    Sustained: 3591
    Okay, but what about all of the other tests that the A11 scored better than the 835?
    We don't have sustained performance tests for most other tests, and for the ones that we do, the A11 throttles hard.

    Hence, we circle back to my original post. Geekbench uses pauses to prevent thermal throttling, so all we see is the peak performance, and not the sustained performance.

    Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle.
    So, you're confident that a couple of synthetic tests of a single aspect of an SOC, the GPU, are a reliable means to predicting the performance of a SOC in general use? 

    Yes, the physics test is CPU bound, the graphics test is GPU bound. Regardless of the results compared to other SoCs, it's very easy to see the difference between peak and sustained performance on the A11.

    That, and we also know each one of the A11's big cores uses 3~3.5 W at load. Having all 6 cores at load (2 big and 4 small) would simply not be sustainable in an iPhone. If the A12 brought that TDP down, with potential adjustments to architecture and the move to a 7nm process, we'll likely see better sustained performance.
    You keep speaking of this, but you haven't actually showed this "difference between peak and sustained performance".  I'm not doubting you, though I presume that you have links?

    "If the A12 brought the TDP down":

    Why would Apple want a lower TDP if for certain workloads, maximum performance, even with throttling, was the desired result? In fact, Apple specifically spoke of this at WWDC, wrt a very rapid ramp of the CPU (that's an instantaneously high TDP!) followed by a more gradual reduction in clock rate and then a shift to the low power cores. I might be incorrect on those details, but I specifically remember the slide.
    I everything I've been talking about is from Anandtech.





    Anandtech

    If these iPhone 11,2 numbers are accurate, Apple might not be pushing the peak performance this iteration and want to ensure sustained performance and efficiency take priority. It is a mobile device after all. Of course, there are still so many factors that weigh in on all of this. As was mentioned earlier, there's a lot more to a mobile SoC that just the CPU and GPU. There's the DSP, ISP, etc., and now a larger push into dedicated ML hardware (although, this can still fall under other components as was the case for Qualcomm's Hexagon DSP).
    So, your argument is that Apple needs to redesign its A series to throttle less for these two sustained benchmarks?

    Folly, pure and simple.
    Argument? If that's what you see, then folly indeed. I've been only posting facts about the A11, it's ability to sustain load and its known TDP values for the large cores. I haven't suggested that "Apple needs to redesign" it's SoC. Everything is speculation based on the current facts and the apparent benchmark values we've been provided with in Appleinsider's post.
    You do know that you relied on a Geekbench score, which doesn't test for sustained performance, to support your view:

    "Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle."

    There's everything to suggest that there will be an improvement in peak performance, but nothing to indicate that Apple will do much to reduce throttling,  I'd be as likely to be correct that it will continue to be a large mismatch, as it is now, because, in most workloads, it just isn't an issue.
    edited July 2018
  • Reply 47 of 77
    tipootipoo Posts: 1,027member

    Year over year I kept expecting Apples single threaded performance gains to slow down, and they kept surprising me thus far. This looks like the year where it's finally too hard to push forward on existing nodes, or else they focused on similar performance with better battery life, if this is real.

    They were getting right up to Intel execution windows, and scaling up past there gets exponentially harder, it's why we've had meager single threaded gains out of Intel for years.


    The GPU is still improved by 40% though, I could almost hear them saying something like "And this year, it's all about the GPU for AR" on the inevitable A12 slide. 

  • Reply 48 of 77
    chasmchasm Posts: 1,263member
    Standard disclaimer: benchmarks are incredibly easy to fake, and anyone who knows the motherboard code of existing phones could guess what the next one would be, so that's no proof either. This far out, the chances of this being fake are very high.
  • Reply 49 of 77
    ericthehalfbeeericthehalfbee Posts: 4,019member
    KITA said:

    Argument? If that's what you see, then folly indeed. I've been only posting facts about the A11, it's ability to sustain load and its known TDP values for the large cores. I haven't suggested that "Apple needs to redesign" it's SoC. Everything is speculation based on the current facts and the apparent benchmark values we've been provided with in Appleinsider's post.



    You're only posting facts?

    Bullshit.

    You're carefully selecting a few key tests where the A11 doesn't perform as good, and condemning the entire SoC because of it. It's why you keep linking those same tests over and over and over as if repeating yourself makes it true.

    Here's a few other slides from AnandTech.


    And here's a great one just for you. This shows the throttling of various devices and indeed the 835 does throttle very little. However, you're missing the point. These are smartphones, not laptops or desktops. Sustained performance isn't nearly as critical as burst performance and race-to-sleep. An A11 or 845 throttled has the same performance as an 835. An A11 or 845 running under normal circumstances will mop the floor with an 835. And this is what's really important. In the worst case scenario an A11 could slow down to the same speed as an 835. For the rest of the time it's waaaaaaaaay ahead. So you can stop with your stupid posts about throttling as if they're somehow relevant.




    And one last one. This is on the S9 equipped with the 845. Not only is it throttled, but you're warned about overheating. I don't think I've ever received such a warning on an iPhone, and I benchmark/stress test mine regularly.


    tmayradarthekatmelgross
  • Reply 50 of 77
    melgrossmelgross Posts: 31,471member
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    Very selective quote that isn’t in context.
  • Reply 51 of 77
    KITAKITA Posts: 163member
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    tmay said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    You took that out of context:

    "When we’re looking at competitor devices we see only the the iPhone X able to compete with the last generation Snapdragon 835 devices – however with a catch. The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle. Unfortunately this also applies to current and last generation Exynos and Kirin SoCs as both shed great amount of performance after only a few minutes. I’ve addressed this issue and made a great rant about it in our review of the Kirin 970. For this reason going forward AnandTech is going to distinguish between Peak and Sustained scores across all 3D benchmarks. This however needs to be tested on commercial devices as the QRD platform isn’t a thermally representative phone for the SoC, so until that happens, we’ll have to just estimate based on power consumption where the Snapdragon 845 ends up.

    This article was a preview on the Snapdragon 845, not a device test.

    Show me a device with a Snapdragon 845, in production, that does better than the A11 and doesn't throttle, and I'll give you a "like".

     
    I was never talking about the 845.

    The Snapdragon 835 does better and doesn't throttle nearly as much.

    For the Pixel 2 using OpenGL ES 3.1 API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2907
    Sustained: 2864

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4390
    Sustained: 3591
    Okay, but what about all of the other tests that the A11 scored better than the 835?
    We don't have sustained performance tests for most other tests, and for the ones that we do, the A11 throttles hard.

    Hence, we circle back to my original post. Geekbench uses pauses to prevent thermal throttling, so all we see is the peak performance, and not the sustained performance.

    Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle.
    So, you're confident that a couple of synthetic tests of a single aspect of an SOC, the GPU, are a reliable means to predicting the performance of a SOC in general use? 

    Yes, the physics test is CPU bound, the graphics test is GPU bound. Regardless of the results compared to other SoCs, it's very easy to see the difference between peak and sustained performance on the A11.

    That, and we also know each one of the A11's big cores uses 3~3.5 W at load. Having all 6 cores at load (2 big and 4 small) would simply not be sustainable in an iPhone. If the A12 brought that TDP down, with potential adjustments to architecture and the move to a 7nm process, we'll likely see better sustained performance.
    You keep speaking of this, but you haven't actually showed this "difference between peak and sustained performance".  I'm not doubting you, though I presume that you have links?

    "If the A12 brought the TDP down":

    Why would Apple want a lower TDP if for certain workloads, maximum performance, even with throttling, was the desired result? In fact, Apple specifically spoke of this at WWDC, wrt a very rapid ramp of the CPU (that's an instantaneously high TDP!) followed by a more gradual reduction in clock rate and then a shift to the low power cores. I might be incorrect on those details, but I specifically remember the slide.
    I everything I've been talking about is from Anandtech.





    Anandtech

    If these iPhone 11,2 numbers are accurate, Apple might not be pushing the peak performance this iteration and want to ensure sustained performance and efficiency take priority. It is a mobile device after all. Of course, there are still so many factors that weigh in on all of this. As was mentioned earlier, there's a lot more to a mobile SoC that just the CPU and GPU. There's the DSP, ISP, etc., and now a larger push into dedicated ML hardware (although, this can still fall under other components as was the case for Qualcomm's Hexagon DSP).
    So, your argument is that Apple needs to redesign its A series to throttle less for these two sustained benchmarks?

    Folly, pure and simple.
    Argument? If that's what you see, then folly indeed. I've been only posting facts about the A11, it's ability to sustain load and its known TDP values for the large cores. I haven't suggested that "Apple needs to redesign" it's SoC. Everything is speculation based on the current facts and the apparent benchmark values we've been provided with in Appleinsider's post.
    You do know that you relied on a Geekbench score, which doesn't test for sustained performance, to support your view:

    "Melgross was claiming that the current A12 scores in the iPhone 11,2 aren't very impressive, but contrary to that, the A12 scores might better represent sustained performance as it's on a new 7nm process. As such, it's likely a considerable improvement over the A11, an SoC that was known to throttle."

    There's everything to suggest that there will be an improvement in peak performance, but nothing to indicate that Apple will do much to reduce throttling,  I'd be as likely to be correct that it will continue to be a large mismatch, as it is now, because, in most workloads, it just isn't an issue.
    As I've said plenty of times now, I relied on Geekbench for peak scores.

    The iPhone 11,2 scores that we have would suggest that peak performance has hardly improved, but it does not mean that sustained performance hasn't.

    Most workloads don't involve some of the new features and experiences coming to mobile. For example, using AR on an iPhone will get it to warm up pretty quickly. In order to create a better experience, sustained performance is critical. Otherwise the user would only be able to use an extensive application for short durations. This isn't just about supporting the status quo, but ensuring that the hardware will support new developments and applications.
    edited July 2018 muthuk_vanalingam
  • Reply 52 of 77
    KITAKITA Posts: 163member
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.
    For the iPhone X using Metal API:

    3DMark Sling Shot 3.1 Extreme Unlimited - Physics (CPU bound):

    Peak: 2523
    Sustained: 1895

    3DMark Sling Shot 3.1 Extreme Unlimited - Graphics (GPU bound):

    Peak: 4428
    Sustained: 2884

    The Snapdragon 835 throttled much less than the A11.
    You should look at the reviews of phones on aRstechnica and anandtech, instead of choosing one spec, which is the one in which Apple doesn’t seem to care much about, which is physics. You need an overal test for this.
    That is from Anandtech.

    This is also from Anandtech:

    The A11 is severely thermally constrained and is only able to achieve these scores when the devices are cold. Indeed as seen from the smaller score of the iPhone 8, the SoC isn’t able to sustain maximum performance for even one benchmark run before having to throttle.

    - Andrei F.

    Very selective quote that isn’t in context.
    It's certainly in context and it's straight to the point. The A11, without a doubt, will throttle considerably under sustained loads.
  • Reply 53 of 77
    KITAKITA Posts: 163member
    KITA said:

    Argument? If that's what you see, then folly indeed. I've been only posting facts about the A11, it's ability to sustain load and its known TDP values for the large cores. I haven't suggested that "Apple needs to redesign" it's SoC. Everything is speculation based on the current facts and the apparent benchmark values we've been provided with in Appleinsider's post.



    You're only posting facts?

    Bullshit.

    You're carefully selecting a few key tests where the A11 doesn't perform as good, and condemning the entire SoC because of it. It's why you keep linking those same tests over and over and over as if repeating yourself makes it true.

    Here's a few other slides from AnandTech.


    And here's a great one just for you. This shows the throttling of various devices and indeed the 835 does throttle very little. However, you're missing the point. These are smartphones, not laptops or desktops. Sustained performance isn't nearly as critical as burst performance and race-to-sleep. An A11 or 845 throttled has the same performance as an 835. An A11 or 845 running under normal circumstances will mop the floor with an 835. And this is what's really important. In the worst case scenario an A11 could slow down to the same speed as an 835. For the rest of the time it's waaaaaaaaay ahead. So you can stop with your stupid posts about throttling as if they're somehow relevant.




    And one last one. This is on the S9 equipped with the 845. Not only is it throttled, but you're warned about overheating. I don't think I've ever received such a warning on an iPhone, and I benchmark/stress test mine regularly.


    Condemning the entire SoC? Such nonsense. I'm discussing how it is known to throttle considerably under sustained loads. 

    I don't know why you're showing those other tests, I'm very clearly talking about sustained performance. We're all aware of iOS + the A11's performance in web based tasks.

    The Snapdragon 845 is on the same process as the Snapdragon 835. To put it simply, Qualcomm will be making major changes come Q1 2019 to their next flagship as it will be using a brand new architecture and make the move to a 7nm process. The 845, in a normal chassis, can't sustain that performance. That being said, some of the gaming oriented smartphones that use it, such as the ASUS ROG Phone or Xiaomi Black Shark, will not throttle as much. One last thing to all of this, although they haven't done yet, Anandtech will be moving onto a benchmark that can utilize Vulkan, and not just older Android APIs.

    Andrei, from Anandtech, pointed out on a number of occasions some major issues with Samsung's software on the S9 (on both the Exynos and Snapdragon models). He made a couple of articles about the software, some of which has apparently been addressed in the consumer release, although I'm sure you'll never read them anyways.

    Beyond gaming, or other niche tools, sustained performance is very useful for things such as AR applications. As I was saying to another user, it's necessary to move beyond the status quo and allow for new experiences to be developed. 

    Although I've done enough to entertain you, once again, you're just supplying useless conjecture, most of it has nothing to do with what I've been talking about. 
    muthuk_vanalingamavon b7
  • Reply 54 of 77
    mcdavemcdave Posts: 1,114member
    KITA said:

    Argument? If that's what you see, then folly indeed. I've been only posting facts about the A11, it's ability to sustain load and its known TDP values for the large cores. I haven't suggested that "Apple needs to redesign" it's SoC. Everything is speculation based on the current facts and the apparent benchmark values we've been provided with in Appleinsider's post.



    You're only posting facts?

    Bullshit.

    You're carefully selecting a few key tests where the A11 doesn't perform as good, and condemning the entire SoC because of it. It's why you keep linking those same tests over and over and over as if repeating yourself makes it true.

    Here's a few other slides from AnandTech.


    And here's a great one just for you. This shows the throttling of various devices and indeed the 835 does throttle very little. However, you're missing the point. These are smartphones, not laptops or desktops. Sustained performance isn't nearly as critical as burst performance and race-to-sleep. An A11 or 845 throttled has the same performance as an 835. An A11 or 845 running under normal circumstances will mop the floor with an 835. And this is what's really important. In the worst case scenario an A11 could slow down to the same speed as an 835. For the rest of the time it's waaaaaaaaay ahead. So you can stop with your stupid posts about throttling as if they're somehow relevant.




    And one last one. This is on the S9 equipped with the 845. Not only is it throttled, but you're warned about overheating. I don't think I've ever received such a warning on an iPhone, and I benchmark/stress test mine regularly.


    Yep, it’s the law.  Apple trolls will always wheel out the 3D Mark Physics benchmarks, whilst ignoring GFXBench peak vs sustained data - misleading by omission.

    Sorry KITA, you picked the wrong forum - best stick to Reddit readers who know no better.
  • Reply 55 of 77
    sphericspheric Posts: 1,730member
    ireland said:
    sflocal said:
    ireland said:
    Battery life gains are everything people want, at this stage. iPad Pro would be 15+ hours on light use would be ideal.
    People who buy the iPad Pro aren't "light" users for sure... 
    Ok, but it was an example. "Two days I used my iPad Pro light and got 15 hours, and yesterday I used it quite heavy and got over 9 hours". That'd be awesome.
    That’s more or less what it feels like using my 10.5“ Pro.
  • Reply 56 of 77
    Koll3manKoll3man Posts: 29member
    melgross said:
    No, it did not. All chips, particularly mobile chips, throttle under load, but the A11 throttles significantly less than any competing chip.

    You wish. Actually the A11 throttles faster and more than the Snapdragon 835 or 845.
    Anandtech made it very clear. Also the performance core on the A11 consume almost 3X more power in full load than S845's performance core.

    edited July 2018
  • Reply 57 of 77
    Koll3manKoll3man Posts: 29member
    And here's a great one just for you. This shows the throttling of various devices and indeed the 835 does throttle very little. However, you're missing the point. These are smartphones, not laptops or desktops. Sustained performance isn't nearly as critical as burst performance and race-to-sleep.

    Wasn't the A11 great because of it's performance in video editing apps? Where it actually needs sustained performance.
    I know that since the A10 and A11 video editing apps have become very important for some iphone fans.

    Burst performance and race-to-sleep is not the end of everything. Right now where I live temperatures are quite high(summer) so phones get hot and can throttle much faster. Isn't this a relevant scenario where improving the efficiency and sustained performance would make a big difference?
    edited July 2018 muthuk_vanalingam
  • Reply 58 of 77
    tmaytmay Posts: 3,545member
    Koll3man said:
    And here's a great one just for you. This shows the throttling of various devices and indeed the 835 does throttle very little. However, you're missing the point. These are smartphones, not laptops or desktops. Sustained performance isn't nearly as critical as burst performance and race-to-sleep.

    Wasn't the A11 great because of it's performance in video editing apps? Where it actually needs sustained performance.
    I know that since the A10 and A11 video editing apps have become very important for some iphone fans.

    Burst performance and race-to-sleep is not the end of everything. Right now where I live temperatures are quite high(summer) so phones get hot and can throttle much faster. Isn't this a relevant scenario where improving the efficiency and sustained performance would make a big difference?
    I'm outraged that there aren't any synthetic benchmarks that simulate video editing on my smartphone in Baghdad in the summer!




    edited July 2018
  • Reply 59 of 77
    blastdoorblastdoor Posts: 1,893member
    KITA said:
    melgross said:
    If true, not an impressive boost from last year, particularly since they say it’s got six cores. I hope this is off.
    The A11 was known to throttle considerably under sustained load.

    Geekbench 4 has pauses built in to avoid thermal throttling.

    These numbers might actually be quite impressive if they're closer to the actual sustained performance of the A12.
    I think this is right. So what we might have with the A12 is:

    more GPU cores
    more sustainable thermals
    more headroom for higher clock speeds in an iPad

    This all lines up with my prediction that the next iPad Pro will have an A12 with a higher clock than the iPhone.  It makes sense -- 6 active cores at a high clock speed is a lot of power. If Apple can make one chip for both the iPhone and iPad, just differing clock speed, it saves on design cost 
    KITA
  • Reply 60 of 77
    Avieshek said:
    Why does Apple ignore the L3 cache? Or even the lack thereof with L4 in an in-house chip.
    More levels does not mean automatically that it's going to be faster, it just means that the designer chose to introduce more tiers because creating a single, large and super-fast cache is not feasible.

    Generally caches are tiered in multiple levels because each higher level is implemented using less-responsive SRAM cells which in turn take longer to access but are smaller on the chip, meaning that L3 or even L4 would be much, much slower than L2 to access but there would be plenty of it. At least thats the differentiation between layers.

    If Apple can afford to make a very responsive Cache (which takes lots of SRAM) like L2 cache and make it this big, that's already really good, in fact the only design I have seen no separate L3 tier because Apple can afford it. It makes cache implementation more easy and can also avoid some multi-processing issues, I think. The A12 pendant chip looks like its the last chip from their design generation which avoids using L3 caches since not many cores are present and the cache does not need to be as big as e.g. a notebook would require.

    Of course I can be mistaken, but that's how it makes sense to me.
    edited July 2018
Sign In or Register to comment.