3 ms·

> and also eliminates a lot of the need to copy data from RAM to VRAM, as both the GPU and the CPU can access the same memory. In return, this allows the GPU to

> and also eliminates a lot of the need to copy data from RAM to VRAM, as both the GPU and the CPU can access the same memory. In return, this allows the GPU to match the performance of discrete GPUs that have a lot more cores. Plenty of PC hardware reviewers have done sensitivity analysis experiments to see how discrete GPU performance is affected by running with a slower or narrower PCIe link. The consensus is usually that GPUs connected by PCIe have more than sufficient bandwidth, and cutting it in half only affects gaming framerates by a few percent. Tighter coupling between CPU and GPU can plausibly have a bigger impact for some GPU compute workloads, but for traditional 3D graphics it doesn't help performance much.

Retric 3y ago
You can generally ignore FPS numbers as irrelevant outside of extremes, what people notice is stutter and that’s where PCIe link bandwidth matters.
- wtallis 3y ago
  I think it's more accurate to say that PCIe link bandwidth matters when your VRAM capacity starts to become insufficient, a problem that Apple sidesteps. The key advantage of Apple's unified memory strategy isn't fast communication between CPU and GPU, but in granting the GPU access to a high-capacity memory pool (with some side benefits of giving the CPU access to GPU-like memory bandwidth).
- dtx1 3y ago
  Measured as one percent lows and again, most GPUs under most Conditions simply don't saturate PCIe enough for that to make a difference. Der8auer and GamersNexus on youtube have plenty of examples.
  - Retric 3y ago
    1% lows in a given benchmark isn’t really measuring stutter in practice, what you want is percentage of frames that don’t finish in time. If once a minute you miss 4 frames in a row that’s noticeable even if everything else is rock solid. The thing is people adjust their resolution/settings to reach acceptable FPS, thus it’s fairly GPU independent. What matters is rendering volatility as assets are loaded etc.
    - sudosysgen 3y ago
      That metric is irrelevant now that we have adaptive refresh rates. You don't really "miss" a frame anymore. 1% low is now the correct metric.
      - Retric 3y ago
        If the pipeline stalls for 70ms it’s a problem even if in theory you don’t “miss” a frame you still aren’t getting new frames until rendering finishes. All adaptive refresh rates do is avoiding unwanted delays after the frame finishes.
        sudosysgen 3y ago
        That's not how it works. A 70ms pipeline stall means a ~90ms frame which means that your 1% framerate low is going to be reduced. Adaptive refresh rate means that only frame rate matters. You can't miss a frame, you can only have a frame take longer, and that's measured perfectly well by 1% or 0.1% low frame rates. Also I don't see why an M series CPU is going to be any better at feeding the (asynchrous, multiple frames in flight) GPU rendering pipeline than a modern x86 CPU, when both are just as fast. macOS is also pretty bad at realtime scheduling for heavy workloads compared to modern Linux and Windows.
        Retric 3y ago
        Yes occasional 90ms stall reduces 1% frame rates, but less than how annoying it is. A 10 minute test at 120 FPS is 10 * 60 * 120 ~= 72,000 frames. Your 1% lows is an average of ~720 frames but so what if of 300 them are at 60 FPS you’re going to have trouble noticing. While if you’re almost rock solid at 120FPS a dozen 70ms stalls it’s really obvious that something is wrong but the calculated 1% low’s may actually look better. This is especially true as those stalls generally correlate with interning things happening. The M series CPU comparison people are talking about isn’t average FPS or even 1% lows where dedicated graphics cars have an advantage, the comparison is what happens when things go very wrong. And it’s in that very specific case when they may have an advantage.
astrange 3y ago
The only reason it wouldn't matter is that software is designed for it to not matter. It enables software to be designed differently ie by not having to ever copy things to VRAM.
- wtallis 3y ago
  That's true to an extent, but when discussing the current performance of GPU hardware the context is necessarily software that actually exists rather than hypothetical software that could exist in the future. The high CPU-to-GPU bandwidth is not currently enabling Apple's GPUs to punch above their weight class (GFLOPS or whatever) except in scenarios where competing GPUs simply don't have enough VRAM. In a future where applications are written with Apple's architecture in mind, any advantage this provides would pretty much boil down to using more VRAM because VRAM capacity isn't as limited anymore.

Retric 3y ago

You can generally ignore FPS numbers as irrelevant outside of extremes, what people notice is stutter and that’s where PCIe link bandwidth matters.

wtallis 3y ago
I think it's more accurate to say that PCIe link bandwidth matters when your VRAM capacity starts to become insufficient, a problem that Apple sidesteps. The key advantage of Apple's unified memory strategy isn't fast communication between CPU and GPU, but in granting the GPU access to a high-capacity memory pool (with some side benefits of giving the CPU access to GPU-like memory bandwidth).
dtx1 3y ago
Measured as one percent lows and again, most GPUs under most Conditions simply don't saturate PCIe enough for that to make a difference. Der8auer and GamersNexus on youtube have plenty of examples.
- Retric 3y ago
  1% lows in a given benchmark isn’t really measuring stutter in practice, what you want is percentage of frames that don’t finish in time. If once a minute you miss 4 frames in a row that’s noticeable even if everything else is rock solid. The thing is people adjust their resolution/settings to reach acceptable FPS, thus it’s fairly GPU independent. What matters is rendering volatility as assets are loaded etc.
  - sudosysgen 3y ago
    That metric is irrelevant now that we have adaptive refresh rates. You don't really "miss" a frame anymore. 1% low is now the correct metric.
    - Retric 3y ago
      If the pipeline stalls for 70ms it’s a problem even if in theory you don’t “miss” a frame you still aren’t getting new frames until rendering finishes. All adaptive refresh rates do is avoiding unwanted delays after the frame finishes.
      - sudosysgen 3y ago
        That's not how it works. A 70ms pipeline stall means a ~90ms frame which means that your 1% framerate low is going to be reduced. Adaptive refresh rate means that only frame rate matters. You can't miss a frame, you can only have a frame take longer, and that's measured perfectly well by 1% or 0.1% low frame rates. Also I don't see why an M series CPU is going to be any better at feeding the (asynchrous, multiple frames in flight) GPU rendering pipeline than a modern x86 CPU, when both are just as fast. macOS is also pretty bad at realtime scheduling for heavy workloads compared to modern Linux and Windows.
        Retric 3y ago
        Yes occasional 90ms stall reduces 1% frame rates, but less than how annoying it is. A 10 minute test at 120 FPS is 10 * 60 * 120 ~= 72,000 frames. Your 1% lows is an average of ~720 frames but so what if of 300 them are at 60 FPS you’re going to have trouble noticing. While if you’re almost rock solid at 120FPS a dozen 70ms stalls it’s really obvious that something is wrong but the calculated 1% low’s may actually look better. This is especially true as those stalls generally correlate with interning things happening. The M series CPU comparison people are talking about isn’t average FPS or even 1% lows where dedicated graphics cars have an advantage, the comparison is what happens when things go very wrong. And it’s in that very specific case when they may have an advantage.

astrange 3y ago

The only reason it wouldn't matter is that software is designed for it to not matter. It enables software to be designed differently ie by not having to ever copy things to VRAM.

wtallis 3y ago
That's true to an extent, but when discussing the current performance of GPU hardware the context is necessarily software that actually exists rather than hypothetical software that could exist in the future. The high CPU-to-GPU bandwidth is not currently enabling Apple's GPUs to punch above their weight class (GFLOPS or whatever) except in scenarios where competing GPUs simply don't have enough VRAM. In a future where applications are written with Apple's architecture in mind, any advantage this provides would pretty much boil down to using more VRAM because VRAM capacity isn't as limited anymore.