finding the bottleneck
Posted: Sun May 28, 2017 2:07 pm
I have this second desktop computer, but its processor and mainboard are quite old (only pcie 1.1 and sata II). however i still want to use it for magic. Unfortunately the second computer performs at about 18fps, while my first desktop performs with about 60fps in the same scene.
I'm litteraly taking my magic-ssd (containing magic and my scene and its assets) out of my first and put it into my second desktop. so the magic version is the same, and also the scene i'm testing this. Further I'm also moving my gpu between the computers. so I assume the performance drop can neither be the ssd nor the gpu.
the scene contains mostly HAP movie clips and a few jpg-sequences. no 3d stuff and no generative stuff like polygons or glsl's.
on my second desktop, while running the scene, neither the cpu or the gpu are completely maxed out. i also tried to lower the resolution by half and to save the scene and its parts in the graphics memory, which weirdly resulted in no significant fps change at all. i also plugged in an older card (gtx 660 instead of my gtx 980), which also resulted in no fps change. meaning that at least the 980 cant rise to its full potential.
so these tests remain inconclusive. the only noticable hardware difference of my two computers are, that the second computer has only pcie 1.1 (while the first has 2.0) and only sata II (while the first sata III). but i think the random-read speed of my ssd doesnt go over the sata II limit anyway. and as i'm using HAP movie files, i assume i have less bandwidth going through the pcie than when i'd use regular clips which would go uncompressed from the cpu to the gpu (or at least that's how i understood the hap thingy being decompressed on the gpu). meaning i doubt that pcie 2vs3 should be the problem.
the assets of this scene have a combined size of almost 1 GB. pci-e 1.1 should still be capable of 4GB/s, so i doubt that's the problem.
what I havent tried yet is using a ram disk, however that would only prove that my ssd is the bottleneck, which it cant be as it's the same device on both computers.
right now i'm concluding that the bottleneck is either the cpu (even t hough not maxed out), the mainboard and/or the ram. however i cant really prove that. maybe there are some unknowns, like new gpu technologies or new cpu technologies that the old hardware doesnt support?
i'd be really interested in knowing your thoughts on this, if you maybe have any more insights or ideas.
the second computer is up to date with all the latest drivers from the manufacturers. i was thorough. the computer is also running stable, temperature- and otherwise.
\\edit: made some changes to the pci-e numbers.
I'm litteraly taking my magic-ssd (containing magic and my scene and its assets) out of my first and put it into my second desktop. so the magic version is the same, and also the scene i'm testing this. Further I'm also moving my gpu between the computers. so I assume the performance drop can neither be the ssd nor the gpu.
the scene contains mostly HAP movie clips and a few jpg-sequences. no 3d stuff and no generative stuff like polygons or glsl's.
on my second desktop, while running the scene, neither the cpu or the gpu are completely maxed out. i also tried to lower the resolution by half and to save the scene and its parts in the graphics memory, which weirdly resulted in no significant fps change at all. i also plugged in an older card (gtx 660 instead of my gtx 980), which also resulted in no fps change. meaning that at least the 980 cant rise to its full potential.
so these tests remain inconclusive. the only noticable hardware difference of my two computers are, that the second computer has only pcie 1.1 (while the first has 2.0) and only sata II (while the first sata III). but i think the random-read speed of my ssd doesnt go over the sata II limit anyway. and as i'm using HAP movie files, i assume i have less bandwidth going through the pcie than when i'd use regular clips which would go uncompressed from the cpu to the gpu (or at least that's how i understood the hap thingy being decompressed on the gpu). meaning i doubt that pcie 2vs3 should be the problem.
the assets of this scene have a combined size of almost 1 GB. pci-e 1.1 should still be capable of 4GB/s, so i doubt that's the problem.
what I havent tried yet is using a ram disk, however that would only prove that my ssd is the bottleneck, which it cant be as it's the same device on both computers.
right now i'm concluding that the bottleneck is either the cpu (even t hough not maxed out), the mainboard and/or the ram. however i cant really prove that. maybe there are some unknowns, like new gpu technologies or new cpu technologies that the old hardware doesnt support?
i'd be really interested in knowing your thoughts on this, if you maybe have any more insights or ideas.
the second computer is up to date with all the latest drivers from the manufacturers. i was thorough. the computer is also running stable, temperature- and otherwise.
\\edit: made some changes to the pci-e numbers.
