Threadripper + Dual-GPU Benchmarks: Premiere Finally Benefits from Multi-GPU
video description
Date: 2020-05-06
Related videos
Comments and reviews: 10
Galactase
This may not be the right video to ask, but I do some simple 1 part 3D models and recently I required a second hand razer blade stealth. Much to my chagrin the 8550u could not run autodesk inventor, i have a student license, anyway, Inventor was completely useless until I decided to see what would happened if I turned off hyperthreading, I expected to see a small performance boost since it seemed that at 15w tdp the 8550u would get stuck around 2. 2-2. 4 ghz and inventor would be unusable. I was suprised at the results I got, I only lost about 10% cpu performance, according to timespy, and cpuz because my all core turbos were significantly higher, Inventor is now usable and works fantastic. That's great for me, but it made me wonder why intel, and amd that matter enable hyperthreading on lower power chips, it almost seems like they're just trying to hit marketing check boxes and in some cases, inventor being a rather specific one, their goal to check marketing boxes and increase multicore performance by a measly 10%, seriously tanks performance in other applications. I thought this might be an interesting idea to explore in a video particularly with amd's new chips coming out that don't have hyperthreading in the 15w package, is that just for product segmentation, or did amd consciously make that choice to not check that box in those specific cases. Hyperthreading or SMT on higher tdp chips makes sense, I've have experienced that myself, but at only 15 w to work with, to stay in spec at least, the extra power dedicated to those threads seems to really tank the power allotted to the core itself, which after all is the part that is actually processing. I know this is really long, but I thought if anyone would find this interesting it would be you guys since you explore really neat stuff and are never afraid to call someone out for making a design choice that may not make a lot of sense. Especially when it is just to check boxes on marketing material. I hope you have the time to make a video like this I'm really interested to see, because after all I am a sample size of one, also hope you guys all feel better. Thanks for your time, Regards
reply
This may not be the right video to ask, but I do some simple 1 part 3D models and recently I required a second hand razer blade stealth. Much to my chagrin the 8550u could not run autodesk inventor, i have a student license, anyway, Inventor was completely useless until I decided to see what would happened if I turned off hyperthreading, I expected to see a small performance boost since it seemed that at 15w tdp the 8550u would get stuck around 2. 2-2. 4 ghz and inventor would be unusable. I was suprised at the results I got, I only lost about 10% cpu performance, according to timespy, and cpuz because my all core turbos were significantly higher, Inventor is now usable and works fantastic. That's great for me, but it made me wonder why intel, and amd that matter enable hyperthreading on lower power chips, it almost seems like they're just trying to hit marketing check boxes and in some cases, inventor being a rather specific one, their goal to check marketing boxes and increase multicore performance by a measly 10%, seriously tanks performance in other applications. I thought this might be an interesting idea to explore in a video particularly with amd's new chips coming out that don't have hyperthreading in the 15w package, is that just for product segmentation, or did amd consciously make that choice to not check that box in those specific cases. Hyperthreading or SMT on higher tdp chips makes sense, I've have experienced that myself, but at only 15 w to work with, to stay in spec at least, the extra power dedicated to those threads seems to really tank the power allotted to the core itself, which after all is the part that is actually processing. I know this is really long, but I thought if anyone would find this interesting it would be you guys since you explore really neat stuff and are never afraid to call someone out for making a design choice that may not make a lot of sense. Especially when it is just to check boxes on marketing material. I hope you have the time to make a video like this I'm really interested to see, because after all I am a sample size of one, also hope you guys all feel better. Thanks for your time, Regards
reply
OneCharmingQuark
If correcting your video after a long export is a problem, then it might be a good idea to look into prerendering solutions. I'm not familiar with the latest version of Premiere, but software like Resolve or FCPX can be set to automatically create high quality ProRes or DNXHR files in the background when you're not actively interacting with the timeline. The implementation in Final Cut Pro is especially good. These generated media files will have great playback, scrubbing, and fast-forwarding in the timeline after they have been rendered. When you do your final video export you have the option to simply transcode these generated media files without the need to re-render graphs, transitions and color grading. If you discover an error in the final product you can change the relevant clip in the timeline while the rest of your files remain rendered. It seems like this could save you a lot of time. It might be difficult to get a powerful Mac to do testing, but I would strongly recommend looking into FCPX, since the overall generated media implementation is still much better than even Resolve.
reply
If correcting your video after a long export is a problem, then it might be a good idea to look into prerendering solutions. I'm not familiar with the latest version of Premiere, but software like Resolve or FCPX can be set to automatically create high quality ProRes or DNXHR files in the background when you're not actively interacting with the timeline. The implementation in Final Cut Pro is especially good. These generated media files will have great playback, scrubbing, and fast-forwarding in the timeline after they have been rendered. When you do your final video export you have the option to simply transcode these generated media files without the need to re-render graphs, transitions and color grading. If you discover an error in the final product you can change the relevant clip in the timeline while the rest of your files remain rendered. It seems like this could save you a lot of time. It might be difficult to get a powerful Mac to do testing, but I would strongly recommend looking into FCPX, since the overall generated media implementation is still much better than even Resolve.
reply
Wh0_Am_
Well, there is some botched testing methodology. Nicehash is a tool/pool that provides easy access to multiple mining algorithms and switches between them as profitability rates change improving efficiency (for a more in depth explanation UFD Tech has some good videos from 2 or so years ago on the subject, each of these mining algorithms have different optimizations for different hardware, much like separate games such as Sniper elite and GTA V to put it in a gaming perspective throwing Nicehash onto a graph for temps or benchmarks is much like throwing Steam onto that very same chart and expecting it to be useful. For example, I have mined with the CrytoNight algorithm and never saw more than 60C on my (air cooled) GPUs meanwhile Lyra2Rev2 and Neoscrypt would easily peak over 90C if I let them. (for reference at the time I used a GTX 1050 LP from MSI, 1 GTX 1060 3GB SC from EVGA and another 1060 3GB GT OC from MSI, and in the end nearly got a return on my investment, and now have somewhat decent cards)
reply
Well, there is some botched testing methodology. Nicehash is a tool/pool that provides easy access to multiple mining algorithms and switches between them as profitability rates change improving efficiency (for a more in depth explanation UFD Tech has some good videos from 2 or so years ago on the subject, each of these mining algorithms have different optimizations for different hardware, much like separate games such as Sniper elite and GTA V to put it in a gaming perspective throwing Nicehash onto a graph for temps or benchmarks is much like throwing Steam onto that very same chart and expecting it to be useful. For example, I have mined with the CrytoNight algorithm and never saw more than 60C on my (air cooled) GPUs meanwhile Lyra2Rev2 and Neoscrypt would easily peak over 90C if I let them. (for reference at the time I used a GTX 1050 LP from MSI, 1 GTX 1060 3GB SC from EVGA and another 1060 3GB GT OC from MSI, and in the end nearly got a return on my investment, and now have somewhat decent cards)
reply
rings
Great Vid Steve! A roll and B roll rendering testing was very enlightening. Sorry to ask you of yet another thing, but can you share results with both your current, and future extra ram performance? Has been a discussion point (my fault) of extra ram, vs the taxation of memory controllers, and thus ACTUAL performance in benchmarks and ACTUAL real world video editing performance, and maybe perhaps Vray digital rendering. I apologize for what will clearly be a very yet another very detailed report. The industry owes you YET again. Will be extra enlightening on 2/7 with your 3990X testing. Can't wait. :) You are an industry hero! Thank you for your very detailed work! I guess you just made another video here with thermal paste testing on the Threadripper line. With said equipment. Love it!
reply
Great Vid Steve! A roll and B roll rendering testing was very enlightening. Sorry to ask you of yet another thing, but can you share results with both your current, and future extra ram performance? Has been a discussion point (my fault) of extra ram, vs the taxation of memory controllers, and thus ACTUAL performance in benchmarks and ACTUAL real world video editing performance, and maybe perhaps Vray digital rendering. I apologize for what will clearly be a very yet another very detailed report. The industry owes you YET again. Will be extra enlightening on 2/7 with your 3990X testing. Can't wait. :) You are an industry hero! Thank you for your very detailed work! I guess you just made another video here with thermal paste testing on the Threadripper line. With said equipment. Love it!
reply
Mythricia
In Blender 2. 8, GPU has almost no penalty with small tile sizes (massively different from 2. 79 and before. The only optimization is you want rectangular tiles, in a 1: 2 ratio. I found that 16x64y (so, tall & skinny) was optimal for my 1080Ti, and the CPU doesn't care at all about tile shape, just the size (so, 16x64 is the same size as 32x32, which was optimal for my 3900X. Larger GPU tiles (bigger than 128x256) in 2. 80++ are actually significantly slower than smaller tiles. That said, separate tile sizes per device would still be nice in future updates, since optimal GPU tile size is still bigger than CPU (128x257 vs 32x32.
reply
In Blender 2. 8, GPU has almost no penalty with small tile sizes (massively different from 2. 79 and before. The only optimization is you want rectangular tiles, in a 1: 2 ratio. I found that 16x64y (so, tall & skinny) was optimal for my 1080Ti, and the CPU doesn't care at all about tile shape, just the size (so, 16x64 is the same size as 32x32, which was optimal for my 3900X. Larger GPU tiles (bigger than 128x256) in 2. 80++ are actually significantly slower than smaller tiles. That said, separate tile sizes per device would still be nice in future updates, since optimal GPU tile size is still bigger than CPU (128x257 vs 32x32.
reply
YOUDIEMOFO
Why oh just why do people insist on leaving air inside their loops. Especially right where it matters most inside the CPU block itself. Just love seeing these high end builds with custom loops by semi professional builders having air left inside them. Come on and take five more minutes and rotate the build or cycle the pumps. Something. Might even improve your temps some. Not sure if people just take B roll right away after building just to get it, or just don't care to bleed loops. There should be no air AT ALL cycling through your loop/pump/in the blocks themselves. Facts.
reply
Why oh just why do people insist on leaving air inside their loops. Especially right where it matters most inside the CPU block itself. Just love seeing these high end builds with custom loops by semi professional builders having air left inside them. Come on and take five more minutes and rotate the build or cycle the pumps. Something. Might even improve your temps some. Not sure if people just take B roll right away after building just to get it, or just don't care to bleed loops. There should be no air AT ALL cycling through your loop/pump/in the blocks themselves. Facts.
reply
NicoD
What you guys do on a system of 700W and a price that I can't pay. I do on a 100 ARM single board computer that consumes 10W. The Khadas VIM3. I edit and render my Youtube videos with it. I even use it when traveling, powered by power banks charged with 5V USB solar panels. And for Blender I've got a farm with as many high performance ARM SBCs as I need. I do have +30 SBCs since I review them. I like seeing what's possible with X86, but I don't see any advantage in using it. It's just a space heater with very little performance/Watt. I think you guys are a bit ignorant.
reply
What you guys do on a system of 700W and a price that I can't pay. I do on a 100 ARM single board computer that consumes 10W. The Khadas VIM3. I edit and render my Youtube videos with it. I even use it when traveling, powered by power banks charged with 5V USB solar panels. And for Blender I've got a farm with as many high performance ARM SBCs as I need. I do have +30 SBCs since I review them. I like seeing what's possible with X86, but I don't see any advantage in using it. It's just a space heater with very little performance/Watt. I think you guys are a bit ignorant.
reply
0zzm0dious
So is it out of place to say that the soft tubing job for that loop looks really. half assed? It surely doesnt look professional by any means. doubling and trippling up of fitting to fitting to fitting just looks sloppy. And its not like he was trying to meet up with hard lines. I dunno. I know it doesnt affect anything, it just seems like flying a guy out there to help out, youd think they would have brought their A game and made that cooling look super clean. I dunno. maybe it's just me. anyone else think it could have been done with a bit more care?
reply
So is it out of place to say that the soft tubing job for that loop looks really. half assed? It surely doesnt look professional by any means. doubling and trippling up of fitting to fitting to fitting just looks sloppy. And its not like he was trying to meet up with hard lines. I dunno. I know it doesnt affect anything, it just seems like flying a guy out there to help out, youd think they would have brought their A game and made that cooling look super clean. I dunno. maybe it's just me. anyone else think it could have been done with a bit more care?
reply
SomeOldTHICCDude
With all the emphasis on video quality out of the way now, are we going to start noticing better audio levels in videos? It s regular enough that I hear clipping in the audio tracks, voiceover specifically. Regardless of the volume and output methods used on my end. To a point, where it s a safe bet that I ll currently NEVER use headphones with GN videos. Because the audio is just so damned harsh to listen to, especially in such immediate proximity to my ears with even a basic pair of headphones, let alone decent ones.
reply
With all the emphasis on video quality out of the way now, are we going to start noticing better audio levels in videos? It s regular enough that I hear clipping in the audio tracks, voiceover specifically. Regardless of the volume and output methods used on my end. To a point, where it s a safe bet that I ll currently NEVER use headphones with GN videos. Because the audio is just so damned harsh to listen to, especially in such immediate proximity to my ears with even a basic pair of headphones, let alone decent ones.
reply
Dr
LoL. The shelves are an Nvidia marketing wet dream. You've got Teddy Bear RX 570 as the only complete Radeon, two dead Vega 10 and 20 dies. The rest of the GPUs and the boxes displayed are all Nvidia. Either baiting or not even trying to hide. I was honestly ok with you hating on the VII because it was never a gaming card, but it was a beast for compute and Nvidia has nothing to answer in price. Radeon VII 3. 360 TFLOPS FP64 700 vs RTX 5000 0. 348TFLOPS FP64 2, 300 Helps when you don't pay for it though right? LOL
reply
LoL. The shelves are an Nvidia marketing wet dream. You've got Teddy Bear RX 570 as the only complete Radeon, two dead Vega 10 and 20 dies. The rest of the GPUs and the boxes displayed are all Nvidia. Either baiting or not even trying to hide. I was honestly ok with you hating on the VII because it was never a gaming card, but it was a beast for compute and Nvidia has nothing to answer in price. Radeon VII 3. 360 TFLOPS FP64 700 vs RTX 5000 0. 348TFLOPS FP64 2, 300 Helps when you don't pay for it though right? LOL
reply
Add a review, comment
Other channel videos
