Ask GN 111: 3950X & Threadripper Paper Launch? Reusing Thermal Pads?
video description
Date: 2020-05-06
Related videos
Comments and reviews: 10
Islam
Ask GN questions: 1- this gonna be long: I am on solar system with batteries and a DC supplied gaming rig using a 250W DC power supply so power is the most important matter to me. I need the most power efficient CPU that I can find today so I looked at the I3-8100 that I already own. it takes something like 6-31 watts and that looks good, but I need better one with higher clocks, more cores and more gaming performance per watt. I only game on my rig so I don't care about other performance benches. anyway I am looking at the I9-9900T a 35W TDP as advertised and can be set to 25W cTDP. The question is: I think that the low base clock and high boost clock on Intel CPUs can tell how much the CPU is power efficient, correct me if this is wrong. 2- I thought about the I9-9900k and then underclock/undervolt, but I don't know if that is possible The question is: can you make a 9900T efficiency out of 9900K? and how to do so? and do I need a Z motherboard for these tweakes? 3- I thought about the 7 nm AMD Ryzen 3900X then undervolt/underclock it, disable PBO and edit the clocks, but I have seen your video about AMD Ryzen CPU realtime clocks not behaving well with undervolts. the question is: can I get a more power efficient AMD CPU (than intel) with better gaming performance per watt and what should I get? should I wait for the new generations so that I may get better 10900k or 10900T. Another important matter here is that intel maybe maybe consume less power on some old applications that uses few cores and set it's own low multipliers so maybe intel has better C-states and consume less at low worload. Do AMD Ryzen have the same efficiency on idle or low loads? 4- more cores is more future proof not sure so I would like to clear that for me if it is true or not. 5- lastly I tried to change boost clock on AMD Ryzen 3400 before, but I couldn't because I found no option for that in ASUS B450 mobo and did not find that either in Ryzen master application, so I want to know why changing boost clocks is not possible?
reply
Ask GN questions: 1- this gonna be long: I am on solar system with batteries and a DC supplied gaming rig using a 250W DC power supply so power is the most important matter to me. I need the most power efficient CPU that I can find today so I looked at the I3-8100 that I already own. it takes something like 6-31 watts and that looks good, but I need better one with higher clocks, more cores and more gaming performance per watt. I only game on my rig so I don't care about other performance benches. anyway I am looking at the I9-9900T a 35W TDP as advertised and can be set to 25W cTDP. The question is: I think that the low base clock and high boost clock on Intel CPUs can tell how much the CPU is power efficient, correct me if this is wrong. 2- I thought about the I9-9900k and then underclock/undervolt, but I don't know if that is possible The question is: can you make a 9900T efficiency out of 9900K? and how to do so? and do I need a Z motherboard for these tweakes? 3- I thought about the 7 nm AMD Ryzen 3900X then undervolt/underclock it, disable PBO and edit the clocks, but I have seen your video about AMD Ryzen CPU realtime clocks not behaving well with undervolts. the question is: can I get a more power efficient AMD CPU (than intel) with better gaming performance per watt and what should I get? should I wait for the new generations so that I may get better 10900k or 10900T. Another important matter here is that intel maybe maybe consume less power on some old applications that uses few cores and set it's own low multipliers so maybe intel has better C-states and consume less at low worload. Do AMD Ryzen have the same efficiency on idle or low loads? 4- more cores is more future proof not sure so I would like to clear that for me if it is true or not. 5- lastly I tried to change boost clock on AMD Ryzen 3400 before, but I couldn't because I found no option for that in ASUS B450 mobo and did not find that either in Ryzen master application, so I want to know why changing boost clocks is not possible?
reply
Peter
Many people confuse AI and deep learning, those are two very different things. At this moment GPU's can handle deep learning, they can't handle AI. Example of deep learning: show tens of thousands of pictures of bees, tell the system that it sees bees and then show a picture of a bee without telling that it is a bee and the system can tell that it is a bee. This on a much larger scale: object-recognition. Also for example for autonomous driving, the system recognizes the lines on the road, signs on the way. and based on that it can make the right decisions as long as nothing deviates from the ordinary (sunlight, rain, mud, snow. The more pictures the better it gets trained. The more you let those cars drive autonomously the better the system gets at it. Just one example of deep learning but the most known one. Another example is to learn computers recognize tumors, computers already are better at it than oncologists. How does a system do this? It looks at patterns, similarities between pictures which get the same tag. In short, deep learning is NOT a hype. I won't make any claim about AI because I don't know enough about it, I can only say that I never have been convinced of any concrete example of AI so far. The YouTube algorithm definitely is not AI, it just is an algorithm, more or less refined (the code is not open. I would be careful with calling that SC2-AI an AI. For AI you need more than just a lot of if-else conditions. Maybe we shouldn't compare chips with the human brain this way at all. There are many similarities (basically the brain also uses 1's and 0's) but also many differences. A computer is much better at calculations than a brain but the brain is much better at visual processing and audio-processing, for now anyway.
reply
Many people confuse AI and deep learning, those are two very different things. At this moment GPU's can handle deep learning, they can't handle AI. Example of deep learning: show tens of thousands of pictures of bees, tell the system that it sees bees and then show a picture of a bee without telling that it is a bee and the system can tell that it is a bee. This on a much larger scale: object-recognition. Also for example for autonomous driving, the system recognizes the lines on the road, signs on the way. and based on that it can make the right decisions as long as nothing deviates from the ordinary (sunlight, rain, mud, snow. The more pictures the better it gets trained. The more you let those cars drive autonomously the better the system gets at it. Just one example of deep learning but the most known one. Another example is to learn computers recognize tumors, computers already are better at it than oncologists. How does a system do this? It looks at patterns, similarities between pictures which get the same tag. In short, deep learning is NOT a hype. I won't make any claim about AI because I don't know enough about it, I can only say that I never have been convinced of any concrete example of AI so far. The YouTube algorithm definitely is not AI, it just is an algorithm, more or less refined (the code is not open. I would be careful with calling that SC2-AI an AI. For AI you need more than just a lot of if-else conditions. Maybe we shouldn't compare chips with the human brain this way at all. There are many similarities (basically the brain also uses 1's and 0's) but also many differences. A computer is much better at calculations than a brain but the brain is much better at visual processing and audio-processing, for now anyway.
reply
Yeest540
#askgn ASK GN: #1 - The main trends these days seem to be mini-ITX, and Mid Tower cases. Mid Tower cases range from budget to expensive and can also vary in size. What happened to case manufacturer full tower line ups excluding specialty cases? Are they more expensive to make with smaller profit margins or people just don't want them? I personally wouldn't mind a bigger box with some more space to build in, but there's nothing new in the category to compete with the nicer Mid Towers. #2 - Overall wouldn't it be better if case manufacturers for the DIY space collectively stopped selling any fans with all their cases? Overall companies would be able to either reduce the prices of their cases, make more money, or put the costs saving into more features. In theory it could also reduce manufacturing/assembly costs. It is not uncommon these days for people replace stock fans, even if they are decent RGB ones with ones of their choice. In the end this would save people money in most cases as people who don't care about looks or rgb can purchase cheap but effective fans, those who do can choose the fans of their choice, and those who have old fans that are still good can continue to use them. I also feel like this can potentially reduce waste. If all the case manufacturers did this, choosing/buying your case fans when you're buying a case would become normal. Keep up the good work!
reply
#askgn ASK GN: #1 - The main trends these days seem to be mini-ITX, and Mid Tower cases. Mid Tower cases range from budget to expensive and can also vary in size. What happened to case manufacturer full tower line ups excluding specialty cases? Are they more expensive to make with smaller profit margins or people just don't want them? I personally wouldn't mind a bigger box with some more space to build in, but there's nothing new in the category to compete with the nicer Mid Towers. #2 - Overall wouldn't it be better if case manufacturers for the DIY space collectively stopped selling any fans with all their cases? Overall companies would be able to either reduce the prices of their cases, make more money, or put the costs saving into more features. In theory it could also reduce manufacturing/assembly costs. It is not uncommon these days for people replace stock fans, even if they are decent RGB ones with ones of their choice. In the end this would save people money in most cases as people who don't care about looks or rgb can purchase cheap but effective fans, those who do can choose the fans of their choice, and those who have old fans that are still good can continue to use them. I also feel like this can potentially reduce waste. If all the case manufacturers did this, choosing/buying your case fans when you're buying a case would become normal. Keep up the good work!
reply
depth386
I have a first world problem. If I have a bunch of crap open, including Google Chrome, if I play The Division 2 with all those other tasks in the background it s fine for like a 15 minutes to an hour and then I get a random freeze, followed by dark screen, and unrequested reboot. It can play the game all day after a reboot. It isn t memory because I built a monster system with 64GB of RAM, 9900K, Gigabyte Z390 Aorus Elite, Radeon VII, Samsung 860 EVO 4TB. Cooling is a ridiculous custom loop with two 420x140x60 rads. It is designed for a lazy life where I leave office, chrome, more chrome, and every game store/launcher in existence open while I game. So this definitely irks me. I have been frequebtly updating the AMD Radeon Software thinking it was just drivers. I also lowered my overclocks majorly to guarantee stability and it still happened. I can do aforementioned lazy life with a serious OC on both the CPU and GPU with other games. I am beginning to blame the game it s self instead of AMD s drivers. Would love to know if anyone here can offer insight. Maybe someone who plays The Division 2 will read this comment. Thanks guys.
reply
I have a first world problem. If I have a bunch of crap open, including Google Chrome, if I play The Division 2 with all those other tasks in the background it s fine for like a 15 minutes to an hour and then I get a random freeze, followed by dark screen, and unrequested reboot. It can play the game all day after a reboot. It isn t memory because I built a monster system with 64GB of RAM, 9900K, Gigabyte Z390 Aorus Elite, Radeon VII, Samsung 860 EVO 4TB. Cooling is a ridiculous custom loop with two 420x140x60 rads. It is designed for a lazy life where I leave office, chrome, more chrome, and every game store/launcher in existence open while I game. So this definitely irks me. I have been frequebtly updating the AMD Radeon Software thinking it was just drivers. I also lowered my overclocks majorly to guarantee stability and it still happened. I can do aforementioned lazy life with a serious OC on both the CPU and GPU with other games. I am beginning to blame the game it s self instead of AMD s drivers. Would love to know if anyone here can offer insight. Maybe someone who plays The Division 2 will read this comment. Thanks guys.
reply
Jeffrey
Machine learning example First, there's different types of AI/ML but a very important approach is based on EVOLUTION or more specifically SURVIVAL OF THE FITTEST. An example of this is creating medicines to cure disease. Once they have the processing power and software to enable this, they will have the PERIODIC TABLE etc available then you can say here's this person's cancer cell at which point you give it some saline solution or whatever as a baseline then see how long the cell lasts. then apply medicine and see how long that lasts. if the cell lasts LONGER keep that new molecule. if it dies earlier drop that approach. modify the molecule that works (by adding random atoms to random sites for example) as well as trying different molecules entirely. give more processing time to modifying molecular variants that produce better results. by this method in theory the virtual processing could tailor the PERFECT SOLUTION for your illness every time. and it ACCELERATES with cross-over as you can start with solutions that work well for similar illnesses. Truly amazing stuff ahead!
reply
Machine learning example First, there's different types of AI/ML but a very important approach is based on EVOLUTION or more specifically SURVIVAL OF THE FITTEST. An example of this is creating medicines to cure disease. Once they have the processing power and software to enable this, they will have the PERIODIC TABLE etc available then you can say here's this person's cancer cell at which point you give it some saline solution or whatever as a baseline then see how long the cell lasts. then apply medicine and see how long that lasts. if the cell lasts LONGER keep that new molecule. if it dies earlier drop that approach. modify the molecule that works (by adding random atoms to random sites for example) as well as trying different molecules entirely. give more processing time to modifying molecular variants that produce better results. by this method in theory the virtual processing could tailor the PERFECT SOLUTION for your illness every time. and it ACCELERATES with cross-over as you can start with solutions that work well for similar illnesses. Truly amazing stuff ahead!
reply
Nexodus
Hi Steve, based on personal opinion or actual knowledge, where do you think the future of Graphics will go from here? (Ignoring existing GPU solutions and soon to be released cards from Nvidia and AMD? There will surely be a time where companies like Nvidia and AMD won't be able to go anything beyond a certain level of GPU performance with there dedicated GPU PCI-e cards, whether it's Moore's Law of a lack of benefit between consistently making newer generations of GPU architecture, i'm not sure. Anyhow, do you think integrated GPU's will come out as being the main target of productions for all forms of graphics for PC usage? better yet, will GPU's based on a direct PCIe connection continue to be the future of Graphics? Also, what may the plans be for memory usage for such iGPU's be? For example would they use one set of GDDR memory that c\serves as direct memory for the CPU and iGPU like consoles have traditionally, or perhaps even a graphics oriented version of High Bandwidth Memory? Thanks Steve, your rad if you read this!
reply
Hi Steve, based on personal opinion or actual knowledge, where do you think the future of Graphics will go from here? (Ignoring existing GPU solutions and soon to be released cards from Nvidia and AMD? There will surely be a time where companies like Nvidia and AMD won't be able to go anything beyond a certain level of GPU performance with there dedicated GPU PCI-e cards, whether it's Moore's Law of a lack of benefit between consistently making newer generations of GPU architecture, i'm not sure. Anyhow, do you think integrated GPU's will come out as being the main target of productions for all forms of graphics for PC usage? better yet, will GPU's based on a direct PCIe connection continue to be the future of Graphics? Also, what may the plans be for memory usage for such iGPU's be? For example would they use one set of GDDR memory that c\serves as direct memory for the CPU and iGPU like consoles have traditionally, or perhaps even a graphics oriented version of High Bandwidth Memory? Thanks Steve, your rad if you read this!
reply
Bruit
#askgn Hey Steve and team. So, having just bought an RX5600XT; I've been reflect on whether the purchase of a 6GB VRAM card was a particular wise choice ( although i got it at a pre-order price, which was equivalent to a lower tier GTX1660 super. My question is regarding GPU VRAM and NVMe drives: Will an NVMe drive, used as gaming only storage, have any effect on VRAM loading, in-game. Say if i exceed VRAM limits during gameplay, and the GPU requires new data to be loaded from Storage. Presumably an NVMe drive would load this data quicker than a SATA SSD, or HDD. But would it have an appreciable impact in-game e. g. a noticeable reduction in stuttering, 1% lows, etc. I assume CPU and system RAM also influence this process, but are they the limiting factor on transfer speed. If yes, does it make an NVMe gaming drive a waste of money, when compared to a slower/cheaper SATA SSD. Please forgive my ignorance if this is not how GPU memory works during game play. Cheers Chris
reply
#askgn Hey Steve and team. So, having just bought an RX5600XT; I've been reflect on whether the purchase of a 6GB VRAM card was a particular wise choice ( although i got it at a pre-order price, which was equivalent to a lower tier GTX1660 super. My question is regarding GPU VRAM and NVMe drives: Will an NVMe drive, used as gaming only storage, have any effect on VRAM loading, in-game. Say if i exceed VRAM limits during gameplay, and the GPU requires new data to be loaded from Storage. Presumably an NVMe drive would load this data quicker than a SATA SSD, or HDD. But would it have an appreciable impact in-game e. g. a noticeable reduction in stuttering, 1% lows, etc. I assume CPU and system RAM also influence this process, but are they the limiting factor on transfer speed. If yes, does it make an NVMe gaming drive a waste of money, when compared to a slower/cheaper SATA SSD. Please forgive my ignorance if this is not how GPU memory works during game play. Cheers Chris
reply
Matthew
i've always had problems with how delidding is portrayed, it's made and advertised to look easy and give you SOOOO much extra performance. in reality the extra performance you get is more dependent on how lucky you are in the silicone lottery more than how good your thermals are. in addition to that you are not going to get a HUGE improvement in any everyday task you use it for, be it gaming or other wise. As far as i am concerned the ONLY reason to delid is if you're going for some kind of record. personally i find for the avg user it is totally not worth it. just get some good thermal paste and call that a day. I see TOO many people in gaming groups destroy perfectly good hardware because they watch videos promoting it and it looked easy. As far as i am concerned the ONLY reason to delid is if you're going for some kind of record. the extra 50-200mhz is not worth the risk
reply
i've always had problems with how delidding is portrayed, it's made and advertised to look easy and give you SOOOO much extra performance. in reality the extra performance you get is more dependent on how lucky you are in the silicone lottery more than how good your thermals are. in addition to that you are not going to get a HUGE improvement in any everyday task you use it for, be it gaming or other wise. As far as i am concerned the ONLY reason to delid is if you're going for some kind of record. personally i find for the avg user it is totally not worth it. just get some good thermal paste and call that a day. I see TOO many people in gaming groups destroy perfectly good hardware because they watch videos promoting it and it looked easy. As far as i am concerned the ONLY reason to delid is if you're going for some kind of record. the extra 50-200mhz is not worth the risk
reply
DR.NEGA
a program is a program fed by us with instructions and code. like starcitzen game. developers just feeding planet template with 6 types of variation then AI we made will learn to code itself by building its owned planets all its owned within the parameters without user input of making 100 planets or 10 billions of them. cause it be very tedious for developers. same-thing for satellite programing tv. one access card automatic update itself without user input. thats call EMM managing ECM keys on its owned. all to make it simple for users not to put too much input. the computer do the rest. thats ML. learns what we teach then AI a program waiting what we tell we want. build entire universe. it will procedural generate entire universe by parameters of options is following the 3 laws in programming. distance, star type, and temp. main laws every AI follows by code we put
reply
a program is a program fed by us with instructions and code. like starcitzen game. developers just feeding planet template with 6 types of variation then AI we made will learn to code itself by building its owned planets all its owned within the parameters without user input of making 100 planets or 10 billions of them. cause it be very tedious for developers. same-thing for satellite programing tv. one access card automatic update itself without user input. thats call EMM managing ECM keys on its owned. all to make it simple for users not to put too much input. the computer do the rest. thats ML. learns what we teach then AI a program waiting what we tell we want. build entire universe. it will procedural generate entire universe by parameters of options is following the 3 laws in programming. distance, star type, and temp. main laws every AI follows by code we put
reply
mod
I was just watching footage of the last starlink launch by Space X and had a thought about game streaming services going forward. What do you think the chances are that Space X puts the servers for game streaming directly on the satellites so that they can offer low latency game streaming services to their customers? This might sound crazy at first, but considering the expected life of the satellites is 5 years it would line up pretty well with the average life of a console at 7 years, if the roll out of the satellites happens over 2 years. It would put them in a very unique situation to offer good video game streaming no matter where you are in the world. Obviously this will likely not be present on the first gen satellites, but going forward it could be an interesting way to avoid a lot of the inherent problems with real time streaming.
reply
I was just watching footage of the last starlink launch by Space X and had a thought about game streaming services going forward. What do you think the chances are that Space X puts the servers for game streaming directly on the satellites so that they can offer low latency game streaming services to their customers? This might sound crazy at first, but considering the expected life of the satellites is 5 years it would line up pretty well with the average life of a console at 7 years, if the roll out of the satellites happens over 2 years. It would put them in a very unique situation to offer good video game streaming no matter where you are in the world. Obviously this will likely not be present on the first gen satellites, but going forward it could be an interesting way to avoid a lot of the inherent problems with real time streaming.
reply
Add a review, comment
Other channel videos
