Yah, best case would be streaming data into RAM in some predictive way so you get lower latency access to the data from RAM. Accessing RAM is already avoided as much as possible, so accessing an nvme directly would be even worse.
Playstation 5 [PS5] [Release November 12 2020]
- Thread starter BRiT
- Start date
Yah, best case would be streaming data into RAM in some predictive way so you get lower latency access to the data from RAM. Accessing RAM is already avoided as much as possible, so accessing an nvme directly would be even worse.
I think some of the confusion maybe comes from thinking about power you'd read from a gpu in a graph like in rivatuner, but wouldn't show you power consumption within a frame. AMD smartshift is operating within frame times, so you have variability on maybe the ms level or faster. If you were to graph with rivatuner it's probably not even sampling that fast.
Edit: I think a good question to ask would be how fast smartshift can actually adjust power delivery. Is it on the ms, us or ns scale?
Edit: I think a good question to ask would be how fast smartshift can actually adjust power delivery. Is it on the ms, us or ns scale?
Last edited:
It will be streaming into RAM. The hw decompressor works in RAM too, at least that's what the patent says.
I would be pretty impressed if any render pipeline will use over 40% VALU.
You don’t need 98% ALU constant to cause voltages to spike. It doesn’t work like that anyway.
all you need is a few areas of code that light up all the cores all doing heavy operations even for a split second here and there to compile a frame. Multiply that by a high frame rate and you’ve got higher power consumption and high heat.
Games are developed with all the hardware in mind. I don’t ask a laptop cpu to crunch a 50 GB training data for data science. You wouldn’t ask a Jaguar to load 1000-2000x objects in view.
the more the hardware can do the more load we will give it. What if we were all wrong about SSDs and everyone is right that I/O has been the largest bottleneck for more complexity even for this generation? Well that problem is solved. Now the data is there ready to crunch.
@sebbbi puts it succinctly: programming is all about taking data from somewhere; doing something with it; and then putting it somewhere.
if you can’t get data fast enough to saturate your ALU then your sitting idle. If you don’t have the bandwidth; you’re sitting idle.
if these issues are relieved then expect to see everything else go up.
It comes down to how much power the gpu has to draw before it needs to start "borrowing" power from the CPU to maintain it's core clock. We don't know any of these numbers, or under what conditions the gpu will hit those limits. Then if you think per frame, over a 16ms time span, how quickly can smartshift adjust those power needs. I'm assuming it's on the us scale, so as gpu workloads fluctuate per frame, it can adjust power accordingly. Beyond that, I really don't think any of this is a big deal outside general curiosity. If they can lower power draw by 27% by lowering clocks by 10%, you're probably looking at clock adjustments that are in the low single digits percentages as the gpu needs more or less power. The average will vary depending on the game. The end result will be something near the published specs.
Last edited:
Theoretical.
Most of the problems in PS4 Pro "high heat" scenarios came from simpler bandwidth oriented code. Like the mentioned "HZD map screen".
That’s just the CPU lighting up in addition to a heavy loaded GPU.
Nope. That's just a "happy path" GPU.
Simple platformers had similar issues AFAIR.
From the horse mouth:
"Developers don't need to optimise in any way; if necessary, the frequency will adjust to whatever actions the CPU and GPU are performing," Mark Cerny counters. "I think you're asking what happens if there is a piece of code intentionally written so that every transistor (or the maximum number of transistors possible) in the CPU and GPU flip on every cycle. That's a pretty abstract question, games aren't anywhere near that amount of power consumption. In fact, if such a piece of code were to run on existing consoles, the power consumption would be well out of the intended operating range and it's even possible that the console would go into thermal shutdown. PS5 would handle such an unrealistic piece of code more gracefully."
The biggest unknown for me is the power cap. We can assume the physical power cap is anywhere between 125W and 200W. The subjective statements are indicating it will be able to maintain close to peak so it gives an idea of what they're aiming for.
Knowing the power cap they settled with will allow us to re-estimate the BOM, so we can predict the retail price before they actually tell us.
I also agree with you we need to know more about what the outlier consumption are in real games. We also previously wanted to know what the outlier power consumption was at mid-gen, but it looks like DF received a 170W xb1x while some other journalists had a 200W. Hopefully there will be a more representative test this time around.
If you wonder what specific load made him say 2/3 was "hard to attain" with the previous engineering methodology, the context is more important because there's no specific load. He said it was about attempting to guess the worst case load in the future, and add safety margins in the cooling solution and PSU. It's an unknown, a guesswork. That does hint they're not going for a 300W giant V (otherwise it would be an very expensive fixed clock), but the comment was only about engineering methodology, not actual power figures.
He did mention AVX256 as a power-hungry outlier. And he said it was about deciding where to invest the BOM (more silicon or more cooling?), calling it a balancing act. High clocks, or large silicon, or both, are easy if the engineer doesn't care about getting the absolute highest performance from a fixed BOM, just throw money at the problem. The biggest engineering headache is knowing you used every last drop of that BOM.
I’m not talking bout PS5. I’m just saying the frame rate when unlocked will be limited by the first bottleneck. Whether that is GPU or CPU or I/O. If you don’t lock frames it will just keep processing stuff faster and faster. With less down time. That is how it works.
And this is simplistic workload just being run at a very high frequency.
This a case the rasterizer is at maximum efficiency. Simple geometry equals to maximum rasterizer efficiency. This is the reason Mark Cerny said simple geometry push further the GPU than complex geometry. This is as simple as that but many people seems to forget how rasterizing work.
Last edited:
Riddlewire
Regular
"PS5 would handle such an unrealistic piece of code more gracefully."
Thanks for clearing that up with such precision, Mark.
Thanks for clearing that up with such precision, Mark.
Dead Rising 2 on PC had this problem for me. Loading screens would exceed 1000 FPS and my GPU would hit 105c, but that game has a built in framerate limiter that fixed this. I'm not sure f that ever got fixed without the limit, haven't played it in a while.
True, even if consoles where 8TF or even lower, with a 1GB/s SSD, and something half decent of a CPU, the machines wouldn't be seen as anything bad anyway.
@Dictator Thanks for your input
It had something to do with profiling.
I don't understand this quote ... The APU should have a max rated power draw, and the cooler should be designed to handle that max power at a particular ambient temp and assuming adequate space has been given for airflow. So as long as your environment is within those ranges, if the APU drew it's max power, which should be an unrealistic number, the console should not overheat. If my console ever overheated in an ideal environment I'd be pissed. "It's the software's fault," would not be an acceptable answer.
That's all statistics. You know. When a fair coin flips head 100 times in a row, it's still pretty possible.
I suspect you're telling me that for PS4 Pro the "max power draw" was not considered. Or Sony were just incompetent?
That's all statistics. You know. When a fair coin flips head 100 times in a row, it's still pretty possible.
I suspect you're telling me that for PS4 Pro the "max power draw" was not considered. Or Sony were just incompetent?
I have no experience with the PS4 Pro, so I don't know what the situation is. As long is it doesn't overheat and shut off, then they did it right. Fan noise is more subjective in terms of what people will tolerate.
My PC GPU has a power limit and the clock is set so it'll never hit the limit. If you overclock you can monitor the power limit and see the clock adjust if you hit the limit. This is something the hardware or firmware should handle. This is not something software developers should have to consider. The whole point of this stuff is to increase utilization and push it as hard as possible so we can get better games.
Similar threads
- Locked
- Deleted member 7537
- Games Industry
- Sticky