AMD R9 3950X Frequency vs. Temperature Scaling & 5. 3GHz OC Stream Recap
video description
Date: 2020-05-06
Related videos
Comments and reviews: 10
Tiki832
Wanted to check something, looking at the Blender results as a hard data table I've seen quite a few reviewers commenting that the 4. 4Ghz all core 1. 3-1. 34v depending on specific chip should be achievable with a higher end aircooler or closed loop with no need for extreme cooling solutions making it actually day to day achievable for a lot of users who would be buying the 3950x. Considering in some use cases this brings it to 9980XE equivalent performance that's quite a return, I know you guys at GN speculated in your initial review on what might be achievable on a better air cooler or closed loop, but after doing your LN2 run and seeing how your specific 3950x scales, would you say that's realistic and on a side point has your earlier thoughts on what it could be pushed to on larger air coolers/closed loops changed now since that review video?
reply
Wanted to check something, looking at the Blender results as a hard data table I've seen quite a few reviewers commenting that the 4. 4Ghz all core 1. 3-1. 34v depending on specific chip should be achievable with a higher end aircooler or closed loop with no need for extreme cooling solutions making it actually day to day achievable for a lot of users who would be buying the 3950x. Considering in some use cases this brings it to 9980XE equivalent performance that's quite a return, I know you guys at GN speculated in your initial review on what might be achievable on a better air cooler or closed loop, but after doing your LN2 run and seeing how your specific 3950x scales, would you say that's realistic and on a side point has your earlier thoughts on what it could be pushed to on larger air coolers/closed loops changed now since that review video?
reply
Toms
By definition going faster as you get cooler=going slower as you get hotter=throttling. Nvidias gpu boost above 64 c therefore throttles by design, it just throttles allot more above 89 c, This is the same. Actual boosting should in theory happen at a fixed temperature like intels turbo boost and possibly RDNA. The hd7950boost for exmaple had a real boost where it didn't slow down as temps increased in normal operating temps. It's like how TDPs are a lie, Say amd make a 3950xs official tdp of 45w at 2. 2Ghz, Turbo boost clocks it up to 4Ghz (125w) but as it gets hotter is slows down to say 3. 5Ghz boost (at 95w, What's its TDP then? Also under what circumstances is base clock relevant? If colder making it go faster is the new definition of boost then what is the new definition of throttling?
reply
By definition going faster as you get cooler=going slower as you get hotter=throttling. Nvidias gpu boost above 64 c therefore throttles by design, it just throttles allot more above 89 c, This is the same. Actual boosting should in theory happen at a fixed temperature like intels turbo boost and possibly RDNA. The hd7950boost for exmaple had a real boost where it didn't slow down as temps increased in normal operating temps. It's like how TDPs are a lie, Say amd make a 3950xs official tdp of 45w at 2. 2Ghz, Turbo boost clocks it up to 4Ghz (125w) but as it gets hotter is slows down to say 3. 5Ghz boost (at 95w, What's its TDP then? Also under what circumstances is base clock relevant? If colder making it go faster is the new definition of boost then what is the new definition of throttling?
reply
Jimmy
Eye-opening results there Steve - Well done! I think we all need to stand back and take some time to realise and fully appreciate just what AMD have done with this game-changing process. EVERYONE benefits - AMD hopefully and justifiably gets the profit they clearly deserve, consumers get much more for less cost (compared to recent past Intel, Intel gets a much needed kick up the ass and are now instructed to PROPERLY innovate. WE ALL benefit - all thanks to AMD in this cycle. If this same binning is seen across the TR3 lineup then folks are going to be very, very happy. A home TR3 system that can game the same if not better than a 3950x would be truly something - all bases (non gaming/gaming) covered. Of course, don't ever consider a TR3 system for gaming-only workloads.
reply
Eye-opening results there Steve - Well done! I think we all need to stand back and take some time to realise and fully appreciate just what AMD have done with this game-changing process. EVERYONE benefits - AMD hopefully and justifiably gets the profit they clearly deserve, consumers get much more for less cost (compared to recent past Intel, Intel gets a much needed kick up the ass and are now instructed to PROPERLY innovate. WE ALL benefit - all thanks to AMD in this cycle. If this same binning is seen across the TR3 lineup then folks are going to be very, very happy. A home TR3 system that can game the same if not better than a 3950x would be truly something - all bases (non gaming/gaming) covered. Of course, don't ever consider a TR3 system for gaming-only workloads.
reply
Chuu
1 GHz on one is not the same as 1 GHz on the other -- I get what you're saying, and I'm not saying you're outright wrong, but you are objectively partially wrong. On any modern out-of-order CPU (not just Intel/AMD, simple ALU ops always have a one-cycle latency, so on tight ALU dependency chains, 1 GHz on one really is the same as 1 GHz on the other (and eg. the vast majority of modern L1 caches similarly do 4-cycle load latency. Yes, most code is not composed solely of tight ALU dependency chains, but those kinds of tight loops (and/or loops that scale similarly to it) do exist in reality and do matter, so it's not really something that can just be dismissed outright.
reply
1 GHz on one is not the same as 1 GHz on the other -- I get what you're saying, and I'm not saying you're outright wrong, but you are objectively partially wrong. On any modern out-of-order CPU (not just Intel/AMD, simple ALU ops always have a one-cycle latency, so on tight ALU dependency chains, 1 GHz on one really is the same as 1 GHz on the other (and eg. the vast majority of modern L1 caches similarly do 4-cycle load latency. Yes, most code is not composed solely of tight ALU dependency chains, but those kinds of tight loops (and/or loops that scale similarly to it) do exist in reality and do matter, so it's not really something that can just be dismissed outright.
reply
Moshman
Hey, just wanted to drop a line and flex about how I got my 24-year-old Pentium 75 OC'd to 75 THz. that's a T there, peasants, (as in terahertz) not a typo. I was able to bypass the law of thermodynamics by fanning the CPU with a dryer sheet as active cooling supplemented by a stack of pennies as the heatsink. So I still got a Pentium 75, just x10, 000 of them. This thing is so fast, I had it render all the CGI in Avatar in the time it took me to start it, go take a poop and come back after. It's so fast, it's already cured 6 diseases in Folding and is now curing diseases that don't exist yet. So I say to you, Gamers Nexus: Git gud.
reply
Hey, just wanted to drop a line and flex about how I got my 24-year-old Pentium 75 OC'd to 75 THz. that's a T there, peasants, (as in terahertz) not a typo. I was able to bypass the law of thermodynamics by fanning the CPU with a dryer sheet as active cooling supplemented by a stack of pennies as the heatsink. So I still got a Pentium 75, just x10, 000 of them. This thing is so fast, I had it render all the CGI in Avatar in the time it took me to start it, go take a poop and come back after. It's so fast, it's already cured 6 diseases in Folding and is now curing diseases that don't exist yet. So I say to you, Gamers Nexus: Git gud.
reply
javierortiz82
16: 00 omg that's so true Steve, this week a guy was asking in a forum for an upgrade over his 6 gb 1060 and I had people with torches answering me when I told him to wait for at least a year and that the 2060 is the worst value because of how anhemic for its price it is and how immature the rtx technology currently is, all because people some people who bought that card seem to need self validation on their purchases. People need to grow up and accept that a, we all have experienced buyer's remorse; b an apples to apples comparison between different architectures is impossible all the time; and c, an opinion is an opinion.
reply
16: 00 omg that's so true Steve, this week a guy was asking in a forum for an upgrade over his 6 gb 1060 and I had people with torches answering me when I told him to wait for at least a year and that the 2060 is the worst value because of how anhemic for its price it is and how immature the rtx technology currently is, all because people some people who bought that card seem to need self validation on their purchases. People need to grow up and accept that a, we all have experienced buyer's remorse; b an apples to apples comparison between different architectures is impossible all the time; and c, an opinion is an opinion.
reply
Klaus
Well, the results everyone seems to be getting from the 3950x, especially when it comes to thermals and power compared to the 3900x and below, confirm what AdoredTV has been saying for many months: that most Ryzen 3000 CPUs use bottom of the barrel silicon, as the good ones go for EPYC for the most part. I'm not saying they are bad by any means. But it makes you wonder what could be done for an ultra-low powered machine like an ultrabook if they just used silicon as good as in the 3950x.
reply
Well, the results everyone seems to be getting from the 3950x, especially when it comes to thermals and power compared to the 3900x and below, confirm what AdoredTV has been saying for many months: that most Ryzen 3000 CPUs use bottom of the barrel silicon, as the good ones go for EPYC for the most part. I'm not saying they are bad by any means. But it makes you wonder what could be done for an ultra-low powered machine like an ultrabook if they just used silicon as good as in the 3950x.
reply
uZiiTV
Lets say I have hundreds of millions of dollars and I want to make the ultimate rig that uses non stop LN2 to keep the 3950X OCed at 5. 3 for everyday use. Also lets say I build a system that automatically provides LN2 at a set pace so CPU gets it when it needs it so I don't have to manually provide it. Is this even a possibility? Can this work or will the system die quickly? Please see this I'm very interested as it would be a dream to do something like that. Love your content.
reply
Lets say I have hundreds of millions of dollars and I want to make the ultimate rig that uses non stop LN2 to keep the 3950X OCed at 5. 3 for everyday use. Also lets say I build a system that automatically provides LN2 at a set pace so CPU gets it when it needs it so I don't have to manually provide it. Is this even a possibility? Can this work or will the system die quickly? Please see this I'm very interested as it would be a dream to do something like that. Love your content.
reply
Davey
Watching this on my non OC'ed 1950x wishing I could still keep up. OC'ed it 4GHz all cores 68c-72c 1. 383v average. BUT it was 1k so Ill back off it was an unnecessary OC. Idling at 38c on my 360mm radiator. I way overshot what the 1950x needed for cooling stock. Funny seeing the CPU stick out of the sides. I hope AMD dosent get rid of the TR CPU. It's fine for gaming, but when I need to do real CAD work it just screams over my old i7 4930k OC 4. 5Ghz.
reply
Watching this on my non OC'ed 1950x wishing I could still keep up. OC'ed it 4GHz all cores 68c-72c 1. 383v average. BUT it was 1k so Ill back off it was an unnecessary OC. Idling at 38c on my 360mm radiator. I way overshot what the 1950x needed for cooling stock. Funny seeing the CPU stick out of the sides. I hope AMD dosent get rid of the TR CPU. It's fine for gaming, but when I need to do real CAD work it just screams over my old i7 4930k OC 4. 5Ghz.
reply
Common
Delid the sucker and use a liquid cooler with liquid metal straight between the die and the block and see what happens? use the derbauer bare die frame for it, so all the parts can be purschased if you want, of course you need a heater to delid it because its soldered. Buyt you can get those easily enough. And about the liquid metal application, use it so it only kind of dyes the surfaces, im talking about 0. 05mm layer or less of liquid metal.
reply
Delid the sucker and use a liquid cooler with liquid metal straight between the die and the block and see what happens? use the derbauer bare die frame for it, so all the parts can be purschased if you want, of course you need a heater to delid it because its soldered. Buyt you can get those easily enough. And about the liquid metal application, use it so it only kind of dyes the surfaces, im talking about 0. 05mm layer or less of liquid metal.
reply
Add a review, comment
Other channel videos
