@@adoredtv Was it you that. a while ago,, claimed that Vegas power draw was the reason AMD went HBM2, since the cards otherwise would just draw far too much power? It was nice to see the difference spelled out hiere in detail.
Well thank you for that video. Short, informative and nicely structured. It really helped me to differentiate between those two types of memory. For a future video I would love to learn more about dark silicon, inactive parts of microchip-area due to the extreme density of modern fabrication processes and the energy cost and density that comes with it. Sophie Wilson mentioned it in her talk "The Future of Microprocessors" that also raised some other interesting points. For example I was surprised that older process nodes are still very viable and perfectly fine for a ton of system designs. Anyways... thanks for the video and the fine people of reddit linking to it. :-)
That's what I wonder as well. Titan RTX get 672 GB/s from 384 bits of GDDR6 while Tesla V100 get 900GB/s from 4096 bits of HBM2. Since they are so close in bandwidth it would be interesting to know how other factors like latency impacts performance when the reading and writing gets really random. It's in latency we see the big difference between 3DXpoint and NAND for example.
@@mokujin29 it sucked at the release but it aged well. drivers are rock stable now and with the right tweaks made this card performs at the level of a GTX 1080/RTX 2060.
Not so poor for NGREEDIA. The only GPUs where they used HBM2 are Titan V (3k $), Tesla V100 (10k $) and DGX2 (68k $), not even the latest RTX Quadro 8000 (10k $) has HBM2. On the other hand, AMD has been using HBM2 for several years and even put 16GB of them on their latest Radeon VII. People just haven't the technical knowledge to fully appreciate the $$$ factor aka the extra time and money behind such kind of memory in a consumer product. Unfortunately, the rumors say AMD will switch to GDDR6 with Navi 7nm... Maybe the $$$ they save will be used to price their GPU lower and thus be more appealing for the consumers. I can understand why they would switch, as the general consumer can't really appreciate the presence of the HBM2 memory, but he can certainly appreciate a lower price tag. Still, as a tech savvy, it's annoying to see such great technology confined only to the enterprise and datacenters/AI industry...
@@lostremnant1140 -- And this is why us consumers needs to buy GPUs with HBM2 in effort to bring down the price for better graphic cards. GDDR memory is such an old concept that it will eventually need to be replaced. Why wait?
Great info, thanks.
New fodder for you new videos?
@@DzheiSilis Actually yes, or at least I was hunting specific information on power draw. ;)
Wild AdoredTV appeared.
@@trastewere AdoredTV used Comment Teaser! It's super effective!
@@adoredtv Was it you that. a while ago,, claimed that Vegas power draw was the reason AMD went HBM2, since the cards otherwise would just draw far too much power? It was nice to see the difference spelled out hiere in detail.
Well thank you for that video. Short, informative and nicely structured. It really helped me to differentiate between those two types of memory.
For a future video I would love to learn more about dark silicon, inactive parts of microchip-area due to the extreme density of modern fabrication processes and the energy cost and density that comes with it. Sophie Wilson mentioned it in her talk "The Future of Microprocessors" that also raised some other interesting points. For example I was surprised that older process nodes are still very viable and perfectly fine for a ton of system designs. Anyways... thanks for the video and the fine people of reddit linking to it. :-)
Thank you Steven. This was very informative especially for us laymen who have surface knowledge of the subject. It's appreciated!
Thanks for posting this, I was just trying to look up earlier this week the differences between the two.
Awesome video.
What are the latency comparisons for GDDR vs. HBM2?
That's what I wonder as well. Titan RTX get 672 GB/s from 384 bits of GDDR6 while Tesla V100 get 900GB/s from 4096 bits of HBM2. Since they are so close in bandwidth it would be interesting to know how other factors like latency impacts performance when the reading and writing gets really random. It's in latency we see the big difference between 3DXpoint and NAND for example.
they have explained it in another video
Thanks for making this video! Now that Apple is using this in their 2020 MBP this video will spike 😁
Oh yes my Vega 56
Sucks
@@mokujin29 it sucked at the release but it aged well. drivers are rock stable now and with the right tweaks made this card performs at the level of a GTX 1080/RTX 2060.
Absolutly true still have one with some slight tweaks you can still run most games at high settings on 60fps at 1440p resolution in my case
Will the future cpu make this kind of design?
WHy ? It is worse than CPU tech.
pastebin ps5 rumor brought me here
That was a fake one then :D
PS5 only 2x more RAM of only 2.55x higher GB/s. Very disappointing after 7 years, much below industry standards.
hbm2 yyds
Hbm just works. Nothing but memory errors and problems with gddr.
GDDR6 is for poor people.
Not so poor for NGREEDIA. The only GPUs where they used HBM2 are Titan V (3k $), Tesla V100 (10k $) and DGX2 (68k $), not even the latest RTX Quadro 8000 (10k $) has HBM2. On the other hand, AMD has been using HBM2 for several years and even put 16GB of them on their latest Radeon VII. People just haven't the technical knowledge to fully appreciate the $$$ factor aka the extra time and money behind such kind of memory in a consumer product. Unfortunately, the rumors say AMD will switch to GDDR6 with Navi 7nm... Maybe the $$$ they save will be used to price their GPU lower and thus be more appealing for the consumers. I can understand why they would switch, as the general consumer can't really appreciate the presence of the HBM2 memory, but he can certainly appreciate a lower price tag. Still, as a tech savvy, it's annoying to see such great technology confined only to the enterprise and datacenters/AI industry...
@@lostremnant1140 -- And this is why us consumers needs to buy GPUs with HBM2 in effort to bring down the price for better graphic cards. GDDR memory is such an old concept that it will eventually need to be replaced. Why wait?
confirmed - navi will use gddr6
@@ahah1785 Almost guaranteed the next consoles will stick to GDDR6.
@@usul573 on a samsung 7nm manufacturing