Gadgetory

this news it just got to us we have enough to get you some of the basics the Titan V is a three thousand dollar Nvidia graphics card it is not meant for gamers it is a scientific workload and computational card including abilities for deep learning and machine learning data sets where you need a lot of memory this has 12 gigabytes of HP m2 and you also need some specialized hardware like tensor cores which were detailed when the Volta architecture was shown at GDC this year before that this video is brought to you by thermal Grizzly makers of the conductor hot liquid metal that we recently used to drop 20 degrees off of our temperatures thermal grizzly also makes traditional thermal compounds we use on top of the IHS like cryo not and Hydra not pastes learn more at the link below as for core specs the card is a 250 watt TDP it uses a Volta GV 100 GPU and is powered by a single 8 pin and single 6 pin power connector this is cooled by the standard nvidia heatsink and fan using a vapor chamber and the cooler is pretty much added to limits in terms of cooling capabilities at this TDP as for the other specs it's got 50 120 cuda cores 640 tensor cores alongside those and the clock on the cuda cores is 12 hundred megahertz at base or 1455 megahertz boost so we're bit lower than something like a 1080i but targeting a much different market where things are perhaps less frequency intensive then for example games they want more cores instead the card also hosts 320 TM use or texture map units and we've got a block diagram from the original Volta announcement that we can show this is from earlier this year the V 100 and you can just see how packed in the SMS are on each of the G pcs but this is somewhat standard if we drill down deeper there and look at the SM layout this is also from the Volta 100 announcement you'll see how the SMS are architected as opposed to the usual layout typically in something like a gaming card you don't see the heavy saturation of F 64 into FP 32 and tensor Coors all-in-one you basically just see FP 32 for something like gaming with maybe a little bit of FP 64 on the side if it's one of the older architectures especially when there was more focus on double precision one of the major examples of the differences between this and one of the modern gaming architectures like Pascal is that the FP 32 saturation here is only 64 units for the entire SM so one SM gives you 64 units of FP 32 similar to what we've seen on some other scientific architectures whereas the 1080i for example runs 128 units that would be on the same SM so the SM architecture is much different on a scientific card than it is on a gaming card you really wouldn't have obviously a great gaming experience on a $3,000 scientific card but this is the first step to Volta coming out to the mainstream and that's probably going to be next year as for other information that might interest you this reference board has a 16 phase vrm it's unlikely that Nvidia will produce or allow partner models of this card but the reference board has plenty of phases to work with the Titan v is also equipped with the same combined l1 and shared memory as seen on GB 100 architectures earlier announced and that means that the developers should theoretically have an easier time working on CUDA or tensor workload software development by leveraging the combined l1 cache and shared memory and for one final bit of perspective here for you the Titan V host 21 point 1 billion transistors the GTX 1080 TI runs 12 billion and the previous Pascal predecessor to the V 100 the GP 100 runs at fifteen point three billion transistors so the increase in transistor count is noteworthy as is presumably the increase in performance in targeted workloads that like machine learning and deep learning hence the tensor integration with Volta we'll have more information on all of this as we get more this is a brand new announcement we weren't even embargoed on it so I'm learning this as as you are right now but there's more to learn about Volta of course rolling into next year otherwise stay tuned to the channel subscribe for more you can catch the follow up coverage check the article linked in the description below we will be updating it if any more information comes out and patreon.com/scishow stuff is out directly or go to store it out gamers nexus net to become a shirt like this one if you'd rather support us there thank you for watching I'll see you all next time