Ask GN 112: How Can AMD Stay Competitive? GPU Price Inflation vs. Graphics?
video description
Date: 2020-05-06
Related videos
Comments and reviews: 10
Dustin
One thing AMD could do to easily lose its advantage would be if they lost Dr. Su as CEO and replaced her with a standard business drone MBA. That's what killed them in the first place, letting business types run the business, which they only did straight into the ground. Putting a PhD-holding engineer in the top seat in a last ditch hail-mary desperate move was unintentionally genius. And it makes the MBAs look bad. I guarantee you there are contingents of their shareholders who are screaming and begging to get an MBA back in the power position. If they ever get their way, dump AMD stock that day because it will mark the beginning of their downfall. To the person who asked about the size of caches and why L3 keeps growing but L2 does not. cache is very, very tricky. Having 'too much' is completely possible and is bad. The more cache you have, the harder cache misses or branch midpredictions hurt. If it loads up 512MB of data and then determines it will not need any of it, flushing that and loading the data you've just realized you are more likely to need is much faster than if you were having to blow out 4GB of cache. Even if the cost of adding more was cheaper economically, the design of the processor and the workload the processor will typically see has to be considered. You can't really decide what size of cache makes sense unless you know how heavily pipelined you are going to be able to build things. If the processor is always going to be doing one type of work, you can fine-tune it very well, but it will make it worse at running workloads that have a lot of differences, and the reverse is also true. On generating enough work to fill everybodys hours. you can easily make that far simpler. Make your people salaried rather than hourly. Then you judge them based upon getting the job done. If they can get it done in a shorter amount of time, that's just a benefit to them. If they already ARE salary, but you want to 'fully load' them with 40 hours a week anyway. stop it. That's a destructive, wrongheaded mindset. That's not what salary positions are for. Salary is when someone is paid to do the work, hourly is when you pay for their time. You're not paying for their time if you're paying them salary. So stop counting it.
reply
One thing AMD could do to easily lose its advantage would be if they lost Dr. Su as CEO and replaced her with a standard business drone MBA. That's what killed them in the first place, letting business types run the business, which they only did straight into the ground. Putting a PhD-holding engineer in the top seat in a last ditch hail-mary desperate move was unintentionally genius. And it makes the MBAs look bad. I guarantee you there are contingents of their shareholders who are screaming and begging to get an MBA back in the power position. If they ever get their way, dump AMD stock that day because it will mark the beginning of their downfall. To the person who asked about the size of caches and why L3 keeps growing but L2 does not. cache is very, very tricky. Having 'too much' is completely possible and is bad. The more cache you have, the harder cache misses or branch midpredictions hurt. If it loads up 512MB of data and then determines it will not need any of it, flushing that and loading the data you've just realized you are more likely to need is much faster than if you were having to blow out 4GB of cache. Even if the cost of adding more was cheaper economically, the design of the processor and the workload the processor will typically see has to be considered. You can't really decide what size of cache makes sense unless you know how heavily pipelined you are going to be able to build things. If the processor is always going to be doing one type of work, you can fine-tune it very well, but it will make it worse at running workloads that have a lot of differences, and the reverse is also true. On generating enough work to fill everybodys hours. you can easily make that far simpler. Make your people salaried rather than hourly. Then you judge them based upon getting the job done. If they can get it done in a shorter amount of time, that's just a benefit to them. If they already ARE salary, but you want to 'fully load' them with 40 hours a week anyway. stop it. That's a destructive, wrongheaded mindset. That's not what salary positions are for. Salary is when someone is paid to do the work, hourly is when you pay for their time. You're not paying for their time if you're paying them salary. So stop counting it.
reply
Chase
To the L2 vs L3 question. L2 functions really to cache things a bit quicker that you don't want to be in L3 closer to a single core or pair of threads etc. L3 caches things for the whole CPU complex. L3 is always the most beneficial place to increase cache as all cores can benefit larger L3 means better multi-threading performance generally. while L2 increases aren't needed as much because they mainly are there to take up slack for the L3 and the performance characteristics between L2/L3 haven't changed too much, say for instance your L3 on the new process had to be 2x slower, well then you would need to reevaluate this relationship and build a larger L2 to compensate for the slower L3. if L3 gets faster you could even make the L2 a bit smaller! Think of the Main memory like a Sea, flowing into a lake and then into a water tower and then into your glass (the CPU) Each smaller reservoir only needs to be large enough to not get drained, if you have a memory access miss prediction there isn't a lot you can do, but if the prediction is working you can keep what you will need in the cache and keep it flowing to the CPU and all of the caches working optimally. The water towers aren't a perfect analogy since data flows in both directions through the memory hierarchy but it gets the idea across I think. For instance if something doesn't fit in cache it must be spilled back into main memory and flow back upstream.
reply
To the L2 vs L3 question. L2 functions really to cache things a bit quicker that you don't want to be in L3 closer to a single core or pair of threads etc. L3 caches things for the whole CPU complex. L3 is always the most beneficial place to increase cache as all cores can benefit larger L3 means better multi-threading performance generally. while L2 increases aren't needed as much because they mainly are there to take up slack for the L3 and the performance characteristics between L2/L3 haven't changed too much, say for instance your L3 on the new process had to be 2x slower, well then you would need to reevaluate this relationship and build a larger L2 to compensate for the slower L3. if L3 gets faster you could even make the L2 a bit smaller! Think of the Main memory like a Sea, flowing into a lake and then into a water tower and then into your glass (the CPU) Each smaller reservoir only needs to be large enough to not get drained, if you have a memory access miss prediction there isn't a lot you can do, but if the prediction is working you can keep what you will need in the cache and keep it flowing to the CPU and all of the caches working optimally. The water towers aren't a perfect analogy since data flows in both directions through the memory hierarchy but it gets the idea across I think. For instance if something doesn't fit in cache it must be spilled back into main memory and flow back upstream.
reply
Milo
The QUESTION I WOULD LIKE ANSWERED WOULD BE THIS, I have two RX 480 8GB XFX Black Edition GPUs in my system and i have had them for about 3 years now so i am thinking about a GPU upgrade. I DO NOT WANT TO SPEND 500+ out of principle, i usually like to spend 300 to 500 on my GPUs, ith that in mind, What GPU should i purchase? My rig is running a GA Aorus Master Z390, 128GB of Corsaie Vengeance LPX RAM, Intel i9-9900KS, Cooler Master Hyper 212 Cooler, the large Cooler Master HAF case with 6 extra fans, running Windows 10 Pro off a Samsung 2TB SSD and several 12TB HDD all powered by a Corsair AX1500i PSU. I rarely game, but it is nice to have the ability, so i dont want to jump into a workstation GPU but i do a lot of phot editing, some video editing, website building, lots of research, it is not unusual to have 200-400 browser tabs open at any time and i have 5 LCD panels and 1 55 HDTV connected to this system. My big thin besides compiling and basic performance is not having to wait more then a 1/2 second for anything to do something. With all that in mind, if i could get some advice from Kingpin or yourself, that would be very kind and appreciated. Safe Travels
reply
The QUESTION I WOULD LIKE ANSWERED WOULD BE THIS, I have two RX 480 8GB XFX Black Edition GPUs in my system and i have had them for about 3 years now so i am thinking about a GPU upgrade. I DO NOT WANT TO SPEND 500+ out of principle, i usually like to spend 300 to 500 on my GPUs, ith that in mind, What GPU should i purchase? My rig is running a GA Aorus Master Z390, 128GB of Corsaie Vengeance LPX RAM, Intel i9-9900KS, Cooler Master Hyper 212 Cooler, the large Cooler Master HAF case with 6 extra fans, running Windows 10 Pro off a Samsung 2TB SSD and several 12TB HDD all powered by a Corsair AX1500i PSU. I rarely game, but it is nice to have the ability, so i dont want to jump into a workstation GPU but i do a lot of phot editing, some video editing, website building, lots of research, it is not unusual to have 200-400 browser tabs open at any time and i have 5 LCD panels and 1 55 HDTV connected to this system. My big thin besides compiling and basic performance is not having to wait more then a 1/2 second for anything to do something. With all that in mind, if i could get some advice from Kingpin or yourself, that would be very kind and appreciated. Safe Travels
reply
pino
Amd really needs to work hard on a net architecture for the future for both cpu and cpu. zen has 2 more years at least but after that they reach intel 7xxx territory. more cores is irrelevant for end users at home when software does not support it so we need faster cores. so new architecture 8 cores and up should be the goal for max 3-4 years from now. The things thatn put pu off on AMD was the way boost works compared to intel which is just a solid = temp = ok max boost! and then theres bios and driver issues which were a mess. for now i dont really care what the do or dont do anymore. i have a 9900k which will serve me for atleast 4 years easy. so next gen for both intel and amd will be interesting again. I think games will keep getting better and better and to a part resolutiuon will cripple fps. the adoption to 1440p is stil bad as it required 2070 super territory cards to play current games on average fps on 1440p Look at RDR2 and outer worlds and borderlands 3 that where barely running ultra on a 2070 1080p.
reply
Amd really needs to work hard on a net architecture for the future for both cpu and cpu. zen has 2 more years at least but after that they reach intel 7xxx territory. more cores is irrelevant for end users at home when software does not support it so we need faster cores. so new architecture 8 cores and up should be the goal for max 3-4 years from now. The things thatn put pu off on AMD was the way boost works compared to intel which is just a solid = temp = ok max boost! and then theres bios and driver issues which were a mess. for now i dont really care what the do or dont do anymore. i have a 9900k which will serve me for atleast 4 years easy. so next gen for both intel and amd will be interesting again. I think games will keep getting better and better and to a part resolutiuon will cripple fps. the adoption to 1440p is stil bad as it required 2070 super territory cards to play current games on average fps on 1440p Look at RDR2 and outer worlds and borderlands 3 that where barely running ultra on a 2070 1080p.
reply
Dillon
Whats going to cause a graphics push in relation to the 6th topic is the release of game consoles. Just like we saw before the the ps4 and xbone release. Mid end or old high end hardware was able to nearly max most games at that time. I was gaming on a HD7870 2GB for the lonnnnnngest time and was able to max or nearly max lot of games in 1080p. Minus stuff like AA and a couple settings that made almost no graphics improvement turned off but stuff started picking up the last couple of years so I picked up an RX580 8gb used. Ex mining card on ebay but its been good so far. At that time I had purchased the hd7870 2gb for 80 used. Aside from an overheating issue due to a poor fan profile it was a good gpu with some fresh paste. Acted odd on certain mobos tho. In intel machines it didnt mind higher temps 80C+ before artifacting. In amd machines 70C was the shutdown point lol was odd but as long as I kept the fan profile tweaked it worked great.
reply
Whats going to cause a graphics push in relation to the 6th topic is the release of game consoles. Just like we saw before the the ps4 and xbone release. Mid end or old high end hardware was able to nearly max most games at that time. I was gaming on a HD7870 2GB for the lonnnnnngest time and was able to max or nearly max lot of games in 1080p. Minus stuff like AA and a couple settings that made almost no graphics improvement turned off but stuff started picking up the last couple of years so I picked up an RX580 8gb used. Ex mining card on ebay but its been good so far. At that time I had purchased the hd7870 2gb for 80 used. Aside from an overheating issue due to a poor fan profile it was a good gpu with some fresh paste. Acted odd on certain mobos tho. In intel machines it didnt mind higher temps 80C+ before artifacting. In amd machines 70C was the shutdown point lol was odd but as long as I kept the fan profile tweaked it worked great.
reply
Milo
Your funny Steve. Go into Best Buy and ask about a computer. , yeah not in Michigan, if you can get them to tell you which systems have a dedicated vs an integrated only GPU then that is about as good as you can get, as for telling you about the systems they sell, most of it is them grabbing the tag and looking up the specs, or casually pretending that they know what their talking about and softly sweeping their hand across the stickers on the case noting that this system features a power Inel or AMD (which ever one maybe in that system) processor and a massive 16GB of RAM (as if that is massive, and its running the latest version of Windows 10, so i would not say that AMD would need to push to much money into training as it doesn't seem to be doing Intel vert much good and the few people that do know what they are talking about usually have a preference and try to push it on you instead of gauging what would suit you best.
reply
Your funny Steve. Go into Best Buy and ask about a computer. , yeah not in Michigan, if you can get them to tell you which systems have a dedicated vs an integrated only GPU then that is about as good as you can get, as for telling you about the systems they sell, most of it is them grabbing the tag and looking up the specs, or casually pretending that they know what their talking about and softly sweeping their hand across the stickers on the case noting that this system features a power Inel or AMD (which ever one maybe in that system) processor and a massive 16GB of RAM (as if that is massive, and its running the latest version of Windows 10, so i would not say that AMD would need to push to much money into training as it doesn't seem to be doing Intel vert much good and the few people that do know what they are talking about usually have a preference and try to push it on you instead of gauging what would suit you best.
reply
sparkyenergia
Unfunk: If you read this. The reason for the L2 vs L2 discrepancy is that most caches are what they call cache inclusive. What this means is that if you have something in L1 cache it is also in L2 and also in L3. If you do some mental gymnastics you can quickly figure out that if you had a 1mb L1 with a 1mb L2 and a 1 mb L3 that you would have 1mb of usable cache. The best arrangement is to have the last level cache be the biggest and the L1/L2 be as small as possible. The next thing is that as that size of a cache increases so does the latency. So you don't wont your L2 to be so small as to put everything into the higher latency L3 cache. That balance has lead to 512k being the go to for quite a while. AMD up until Ryzen used a cache exclusive design meaning data was not replicated this allowed AMD to have 512kb cache with the first Athlon 64's.
reply
Unfunk: If you read this. The reason for the L2 vs L2 discrepancy is that most caches are what they call cache inclusive. What this means is that if you have something in L1 cache it is also in L2 and also in L3. If you do some mental gymnastics you can quickly figure out that if you had a 1mb L1 with a 1mb L2 and a 1 mb L3 that you would have 1mb of usable cache. The best arrangement is to have the last level cache be the biggest and the L1/L2 be as small as possible. The next thing is that as that size of a cache increases so does the latency. So you don't wont your L2 to be so small as to put everything into the higher latency L3 cache. That balance has lead to 512k being the go to for quite a while. AMD up until Ryzen used a cache exclusive design meaning data was not replicated this allowed AMD to have 512kb cache with the first Athlon 64's.
reply
Vernacular
29: 58 If I can't do it really well [. ] then I don't do it. Thanks for addressing my question. I'm still disappointed I couldn't prod GN into doing something with SFF again, but your small team is working on many things so there's a lot on their plate already. I saw your list of pitfalls for CPU cooler testing and for your own sanity I don't recommend being that rigorous with absolutely everything. I mean, you didn't test thermals for the Cute Pet case since that was more focused on the novelty and build quality. Perhaps I just need to find an SFF case under 20 liters shaped like a cat or with a waifu printed on the side and boom, we got some SFF content again! Joking aside, Steve laid it all out clearly and even said Sorry at one point so although I really want SFF content I can't really be mad about it.
reply
29: 58 If I can't do it really well [. ] then I don't do it. Thanks for addressing my question. I'm still disappointed I couldn't prod GN into doing something with SFF again, but your small team is working on many things so there's a lot on their plate already. I saw your list of pitfalls for CPU cooler testing and for your own sanity I don't recommend being that rigorous with absolutely everything. I mean, you didn't test thermals for the Cute Pet case since that was more focused on the novelty and build quality. Perhaps I just need to find an SFF case under 20 liters shaped like a cat or with a waifu printed on the side and boom, we got some SFF content again! Joking aside, Steve laid it all out clearly and even said Sorry at one point so although I really want SFF content I can't really be mad about it.
reply
ageofbogyo
They have to focus on software too, windows patches years later their cpus need it, drivers 6-12 months later after a product launch still having lots of issues. Most games being optimized for NVidia and not for their cards (so the true tflops power of their cards just shows in professional special work apps) etc. They must focus on laptops/mobile uses and making profit. DYI desktop market is a dying small fringe segment. They could take the whole gaming laptop market from intel+nv with their console APU tech. Though i have a feeling we will have cloud gaming be a thing by the time it would happen. Keep giving up the top marketing flagship 1% gpu market. They cannot compete against NVidia and intel 2 giants at the same time on multiple fronts (consumer, professional, data center/server/AI)
reply
They have to focus on software too, windows patches years later their cpus need it, drivers 6-12 months later after a product launch still having lots of issues. Most games being optimized for NVidia and not for their cards (so the true tflops power of their cards just shows in professional special work apps) etc. They must focus on laptops/mobile uses and making profit. DYI desktop market is a dying small fringe segment. They could take the whole gaming laptop market from intel+nv with their console APU tech. Though i have a feeling we will have cloud gaming be a thing by the time it would happen. Keep giving up the top marketing flagship 1% gpu market. They cannot compete against NVidia and intel 2 giants at the same time on multiple fronts (consumer, professional, data center/server/AI)
reply
dexter2433
i have 2 systems one Ryzen 7 2700X 32 gig DDR 4 3200 and a 1TB M. 2 nvme and XFX 5700 and my internet rig for all use other than gaming is a Ryzen 3 2200g with refurbished Nvidia 1650 none ti 140 with 16 gig ddr 4 2400 and a 3 year old 320gig ssd sata drive i had laying around and i share a 32 MSI 1440 144hz screen and i have tossed a few games on the internet rig just to see and I was surprised the 1650 and 2200g can play quiet a few games just fine. not maxed out or everything on ultra like the gaming rig but plenty good that you can enjoy gaming on it. and you can put together for less than 500 easy it even has a 20 VS450 psu in a 27 case and it runs great rock solid never crashes but it keeps all the junk off of my gaming rig and that is why o tossed it together.
reply
i have 2 systems one Ryzen 7 2700X 32 gig DDR 4 3200 and a 1TB M. 2 nvme and XFX 5700 and my internet rig for all use other than gaming is a Ryzen 3 2200g with refurbished Nvidia 1650 none ti 140 with 16 gig ddr 4 2400 and a 3 year old 320gig ssd sata drive i had laying around and i share a 32 MSI 1440 144hz screen and i have tossed a few games on the internet rig just to see and I was surprised the 1650 and 2200g can play quiet a few games just fine. not maxed out or everything on ultra like the gaming rig but plenty good that you can enjoy gaming on it. and you can put together for less than 500 easy it even has a 20 VS450 psu in a 27 case and it runs great rock solid never crashes but it keeps all the junk off of my gaming rig and that is why o tossed it together.
reply
Add a review, comment
Other channel videos
