Dolphin, the GameCube and Wii emulator - Forums

Pages: 1 2

Hi

I have an idea, test the TFOPS of the CPU/GPU and real world performance and then define the relationship between them. That will show for sure if TFLOPS is a valid measure

The theoretical TFLOP rating of my CPU is 1.2 TFLOPS. My CPU is the Ryzen 5 3500X. A main frame can do 442000 TFLOPS.

Edited.

My GPU does 3 TFLOPS, nearly.

There are 2 issues with this for this. One is the theoretical fltops represent an all core workload that is in ideal and for the most part unrealistic optimize code to get. The problem is dolphin won't use all the core of a modern system. For the most part dolphin can only use 2 or 3 core at best even if you have 12 cores in my desktop r9 5900x case. Next like Nintendo Maniac 64 said other factor like cache size, memory speed, instruction set usage, etc can make measuring Tflops useless. There are countless difference benchmark because 2 different cpu can perform different depending on the program. For example the r9 5900x(The cpu in my main system) geekbench score is around 1669 compared to 1244 for the r5 3600x. This is a different of about 34 percent. Going to anandtech bench page and looking at the dolphin benchmark . the r9 time is 229 compare to the r5 263 is only about 14 percent. Here the page https://www.anandtech.com/bench/product/2674?vs=2610 . I use the r5 3600x since anandtech didn't have the r5 3500x in the bench page. Notice in the page that the difference between the 2 cpu can vary greatly depending on the program.

Gpu is just as useless if not more to compare tflops. Drivers, graphic api, game optimization are examples of variables that can influence performance. This says nothing of memory type and speed in the case of integrated graphic or some graphic cards(gt 1030 had both ddr 4 and gddr 5 option for example).

It's interesting to note on the anandtech page, the 3600X out performs 5900x in a couple of tests, but seeing as there was about 80 benchmarks, this puts it in perspective: your cores increase the TFLOPS 'power' of your CPU.

Here's how to calculate the speed of your CPU, in TFLOPS.

1. For a CPU, multiply the number of sockets by the number of cores for each socket.

2. Take that and multiply the number of clock cycles per second by the number of floating-point operations per cycle. (Where do I find this info?). Should I just use geekbench?

Remember that almost all computers (desktop or laptop) used for personal use will have a CPU with just one socket.

Edit: the X500 mini computer while being portable, the Vega 8 gpu doesn't really cut it on the modern games https://www.notebookcheck.net/AMD-Radeon...180.0.html

(09-03-2022, 10:11 PM)davediamond Wrote: [ -> ]the Vega 8 gpu doesn't really cut it on the modern games

That's kind of what I was alluding to when I said that the Steam Deck's iGPU is substantially more powerful.

Thing is though, for emulation workloads, you want more focus on the CPU cores than the GPU but, for native non-emulation gaming, you want more focus on the GPU than the CPU (and console hardware themselves take this "less focus on the CPU" thing to an even greater extreme than native PC gaming tends to).

...until raytracing gets factored in, then it seems like you want pretty equal focus on both the GPU and CPU (though the next two gens of GPU architectures from both AMD and Nvidia are rumored to start absolutely doubling-down on raytracing performance, in which case it's very possible that raytracing in the future could become more of a CPU bottleneck just like emulation tends to be, especially as resolutions and refresh rates are beginning to approach diminishing returns outside of VR).

(09-03-2022, 07:03 AM)Nintendo Maniac 64 Wrote: [ -> ]Hi, I went for the 256GB version.

Just keep in mind that you'll be waiting a bit for one, though it seems not nearly as long as previously. Don't suppose it'd be rude of me to ask which model you went for?

I do not because TFLOPS tend to not be comparable between different architectures, especially on the GPU side of things. That being said, GPU performance in emulation shouldn't really be a special case like it is with CPU performance, so one should be able to just look at normal gaming benchmarks to get an idea of emulation GPU performance from the usual places (exclusive ubershaders may be an exception to this but, in general, there really shouldn't be a practical reason to ever use exclusive ubershaders... but I guess it could make for an interesting GPU-compute benchmark?).

And for CPUs it's usually easier to just look at the Dolphin 5.0 benchmark results (whether in my signature or via Anandtech's bench) and, if a specific CPU isn't present, then to use a CPU of identical architecture and then do some relatively simple math to figure out what performance it would be at the clockrate used by your processor (e.g. a 2.5GHz 6500T will take nearly exactly twice as long as a 5.0GHz 10900K since they use the same core architecture save for cache size which impacts Intel's performance only a bit until you get to non-Atom-based Celeron CPUs which have such a major cache reduction that performance can be drastically reduced).

The one niggle is that AMD CPUs seem to gain much more from cache capacity in Dolphin than Intel CPUs, so AMD's G-series processors such as the 5600G perform considerably worse clock-for-clock than their non-G equivalents like the Ryzen 5600... at least up to a point since the V-cache models with absurd amounts of cache like the 5800X3D only seem to perform a teeny bit better clock-for-clock than the non-V-cache 5800X.

Is there any way I can get around the moderator delay when I post? It's quite annoying!

It likely because you a new user. Probably intended to prevent abuse like spam. I don't remember if it time or post based though.
For the most part you're better off ignoring tflop and look at benchmark related to you program/games yoi want to run. That generally the way to go. For example the i3 12300 can run dolphin faster then my r9 5900 according to this https://www.anandtech.com/bench/product/2897?vs=2674, but in other program is completely outperformed by that same cpu.

I have an Optiplex 3040M with an i5-6500T processor that I use as a Linux small server. When I'm checking out some Linux software, I occasionally run Dolphin on it, and the games I've played at 1x and 2x play just great.
Whether you have a certain game (or games) in mind, I can see if I have it in my collection and play it for you.

Pages: 1 2

Dolphin, the GameCube and Wii emulator - Forums

davediamond

davediamond

themaster123

davediamond

Nintendo Maniac 64

davediamond

davediamond

themaster123

MichealRoss