WEBVTT 00:00:00.160 --> 00:00:06.399 it's here it is ripping fast and it's 00:00:03.600 --> 00:00:11.880 $2,000 but in exchange for your least favoritest of kidneys Invidia promises 00:00:08.840 --> 00:00:14.200 that their brand new GeForce RTX 590 00:00:11.880 --> 00:00:21.600 will deliver a level of performance that obliterates their only real competition 00:00:17.439 --> 00:00:25.039 NVIDIA more GPU cores boom more vram and 00:00:21.600 --> 00:00:28.320 faster vram boom boom enhanced RT and AI 00:00:25.039 --> 00:00:30.720 course wider memory bus and PCI Gen 5 00:00:28.320 --> 00:00:36.000 boom boom boom on top of that NVIDIA has packed in 00:00:33.680 --> 00:00:40.280 a deep learning super shedload of new features that I would love to tell you 00:00:37.600 --> 00:00:44.719 about but unless NVIDIA also invented AI teeth extraction I won't be able to so 00:00:42.760 --> 00:00:48.879 instead I leave this review in your capable hands see you 00:00:48.879 --> 00:00:55.320 later I got it I got it I got it first up graphics performance let's cut to the 00:00:53.440 --> 00:00:59.239 chase instead of being scammed online I found a way to prevent you from falling 00:00:56.840 --> 00:01:03.640 into traps when you buy and sell gpus our sponsored job is a legit Marketplace 00:01:01.239 --> 00:01:06.400 to score some deals Ang Jawa is always real never 00:01:13.439 --> 00:01:20.240 cake let's get right to Raw gaming results no rate tracing no upscaling and 00:01:18.000 --> 00:01:25.119 we're starting with 1440p across our suite of games at 1440p the 90 never 00:01:22.560 --> 00:01:29.600 Falls flat on its face obviously but still manages to be underwhelming in the 00:01:27.880 --> 00:01:34.520 Vulcan based Red Dead Redemption 2 we see less than a 10% improvement over the 00:01:32.360 --> 00:01:39.880 490 and that lackluster uplift is repeated in f124 more problematic is the 00:01:37.360 --> 00:01:46.880 embarrassingly small 30% lead over the 7900 XTX which frequently goes for a 00:01:43.240 --> 00:01:48.840 little over 2 fifths of the price o and 00:01:46.880 --> 00:01:53.119 veral doesn't bring better news but as we move on to newer more graphically 00:01:50.719 --> 00:01:57.240 intensive games the 1590 does start to pull away from the pack in the gorgeous 00:01:55.000 --> 00:02:02.960 Thriller Allen Wick 2 it beats the 3090 TI by more than double and looks great 00:01:59.920 --> 00:02:05.680 in Blackmouth Wukong beating the 490 by 00:02:02.960 --> 00:02:09.119 27% cyberpunk is another strong point compared to the previous generations but 00:02:07.520 --> 00:02:13.360 as low as they are on the chart it's worth noting AMD's strong performance 00:02:11.200 --> 00:02:17.560 per dollar in this game at least when R tracing isn't enabled we'll get to that 00:02:15.239 --> 00:02:23.000 later for now this might be obvious but if you are a 1440p player the 90 is 00:02:20.760 --> 00:02:27.040 overkill with the current crop of CPUs if you're on the latest 9800 x3d you 00:02:24.959 --> 00:02:30.680 might find that the 590 exerts a little bit more of a commanding lead but I 00:02:28.800 --> 00:02:34.519 think that anyone with this setup should be putting their money into a new 00:02:31.879 --> 00:02:37.879 monitor rather than a new CPU let's move on to 4K testing where the CPU 00:02:36.120 --> 00:02:43.319 bottlenecks are less likely to rear their ugly heads in cyberpunk the 590 is 00:02:40.879 --> 00:02:50.080 the first card to ever crack triple digits at our Ultra preset scoring a 00:02:45.879 --> 00:02:51.879 30fps lead over the 490 in Allen wake 2 00:02:50.080 --> 00:02:55.239 the story remains largely the same offering a noticeable difference in 00:02:53.239 --> 00:02:59.879 playability compared to any previous Flagship black myth Wukong at cinematic 00:02:57.680 --> 00:03:06.200 settings however is the Everest lik Summit where even the 59 falls short of 00:03:03.120 --> 00:03:08.480 60 FPS average perhaps an overclocking 00:03:06.200 --> 00:03:12.760 Sherpa could get us to the peak but that's a subject for another day in Red 00:03:10.200 --> 00:03:17.319 Dead Redemption 2 the 590 does not impress especially when you consider its 00:03:14.519 --> 00:03:20.519 price and in f124 the 90 continues to operate at that kind of level of 00:03:18.480 --> 00:03:24.560 performance that no one else can touch the problem is that for all the hype the 00:03:22.920 --> 00:03:29.920 performance bump is roughly on par with the price bump making the 590 look less 00:03:27.239 --> 00:03:35.360 like a truly Next Generation product and more like a 490 Super GT zai buai but 00:03:33.720 --> 00:03:38.959 what's the deal I thought that Blackwell was supposed to be the giant leap 00:03:36.879 --> 00:03:43.319 forward with all that flip metering and neural rendering and increased rate 00:03:41.000 --> 00:03:47.439 triangle intersections what the [ __ ] whoa whoa all right hold on that's a lot 00:03:45.560 --> 00:03:52.400 of words but to understand how they are going to impact performance and they 00:03:49.879 --> 00:03:58.000 will we need to understand what they mean see Blackwell brings so many new 00:03:55.799 --> 00:04:03.519 enhancements that inidia marketing doesn't even call it a GPU U 00:04:00.480 --> 00:04:08.319 architecture no it's called a neural 00:04:03.519 --> 00:04:10.519 rendering architecture so what is that 00:04:08.319 --> 00:04:15.360 as far as we can tell it's about equal parts genuine Innovation and marketing 00:04:12.879 --> 00:04:20.000 fluff we'll start with the Innovation up until this point NVIDIA's AI 00:04:17.359 --> 00:04:25.040 accelerating tensor course could not be accessed by a graphics API like Vulcan 00:04:22.680 --> 00:04:29.720 or DirectX but through collaboration with Microsoft directex now has the 00:04:27.479 --> 00:04:35.000 Cooperative vector's API which means means that Gamers can use neural shaders 00:04:32.919 --> 00:04:39.600 unlike typical shaders this allows geometry to be imbued with extra 00:04:37.199 --> 00:04:44.160 properties but now that extra property could be a small neural network that 00:04:41.759 --> 00:04:49.600 could generate more geometry or help ease Ray tracing calculations for 00:04:46.600 --> 00:04:52.160 instance Mega geometry this one allows 00:04:49.600 --> 00:04:58.080 for realtime generation of level of detail steps without requiring any 00:04:55.000 --> 00:05:00.320 normal Maps think of like UE 5's nanite 00:04:58.080 --> 00:05:05.360 which helps ease jarring l change effects and saves developer time but 00:05:02.600 --> 00:05:09.759 with you know AI to take advantage of these features NVIDIA loaded the 590 00:05:07.520 --> 00:05:14.639 with the hardware it needs to accelerate them it's got fifth gen tensor cores 00:05:12.400 --> 00:05:20.280 which drastically reduces memory usage for simpler AI models that don't require 00:05:17.600 --> 00:05:24.360 High precision as for the non- AI stuff we get upgraded fourth gen RT cores 00:05:22.479 --> 00:05:29.039 which now double the ray triangle intersection rate with just 75% of the 00:05:27.039 --> 00:05:33.600 memory footprint and as for the regular old Cuda course well those just don't 00:05:31.199 --> 00:05:38.720 seem to have changed very much so far the 90 has managed a best case scenario 00:05:36.080 --> 00:05:43.960 of about 30% faster than his predecessor seemingly entirely thanks to the 33% 00:05:41.680 --> 00:05:48.600 higher GPU core camp this combined with their reuse of tsmc's 4N process node 00:05:46.639 --> 00:05:52.039 from last gen explains why the new chip is so big and why NVIDIA had to 00:05:50.319 --> 00:05:57.319 sacrifice some clock speed to keep their yields and therefore pricing still 00:05:54.360 --> 00:06:01.479 attainable to the 1% GDDR 7 on the other hand is kind of a big deal it boasts 00:05:59.199 --> 00:06:06.160 doubled the data rate of gddr6 while using half as much power per bit this is 00:06:04.039 --> 00:06:11.440 an enlarged part thanks to the shift to Pam 3 signaling Pam short for pulse 00:06:09.160 --> 00:06:16.560 amplitude modulation is akin to how we store data in multi-level cell flash 00:06:13.479 --> 00:06:18.720 storage gddr6 uses Pam four meaning that 00:06:16.560 --> 00:06:23.120 each clock can be encoded for four different states rather than just two 00:06:20.960 --> 00:06:27.240 but it came with a big trade-off the error rate since the signals are so 00:06:25.240 --> 00:06:30.199 similar in amplitude that sometimes they can be hard to tell apart especially 00:06:28.919 --> 00:06:34.400 when there's interf Pam 3 improves the situation by just 00:06:32.599 --> 00:06:37.960 trying to handle three states instead of four giving a little bit more room 00:06:36.400 --> 00:06:43.039 between each of them this improves signal Integrity allowing GDDR 7 to run 00:06:40.400 --> 00:06:46.919 at higher frequency while consuming less power to make up for the trade-offs oh 00:06:44.840 --> 00:06:53.479 and let's not forget that we finally got 32 gigs of vram this will be a huge jump 00:06:49.639 --> 00:06:54.880 for AI dorks and maybe Gamers someday 00:06:53.479 --> 00:06:59.199 but there are some other cool things like NVIDIA's new 9th gen nank Hardware 00:06:57.560 --> 00:07:05.080 video encoders which support higher quality 422 10bit video this for the 00:07:02.560 --> 00:07:09.400 right people is a huge deal and might make Blackwell a must-have upgrade and 00:07:07.599 --> 00:07:15.120 for the folks out there who own monitors hi plof we finally get a card that can 00:07:11.639 --> 00:07:17.000 actually take advantage of DP 2.1 uh2 a 00:07:15.120 --> 00:07:21.639 new DisplayPort standard that can drive 4K 240 HZ without display stream 00:07:19.720 --> 00:07:27.479 compression and all of this while talking to your computer at PCIe Gen 5 00:07:25.080 --> 00:07:31.960 it's 2025 and rate tracing is no longer an afterthought or even a choice in some 00:07:29.960 --> 00:07:37.120 cases with the new Jones being the first game to outright require support so 00:07:35.000 --> 00:07:40.840 let's talk about it for RT testing we use the highest settings starting at 00:07:38.919 --> 00:07:45.680 1440 uh and I want to get this out of the way AMD does not R Trace well n wake 00:07:43.680 --> 00:07:50.599 2 makes for a very playable experience on the 590 with 1% lows well above 60 00:07:48.280 --> 00:07:55.159 FPS numbers it can't quite hit yet on the absolutely brutal black myth Wukong 00:07:53.039 --> 00:08:02.440 though it is playable unlike the poor 7900 XTX ouch in cyberpunk the 59 has 00:07:59.759 --> 00:08:06.560 just a 20% lead over the 490 but compared to the 480 super it maintains 00:08:04.639 --> 00:08:10.240 its price to Performance ratio which I generally consider to be pretty darn 00:08:08.120 --> 00:08:15.800 acceptable for a Halo class card in the lightly rate traced f124 AMD comes back 00:08:13.280 --> 00:08:20.599 to life a little performing well against the 480 super and the same can be said 00:08:17.960 --> 00:08:26.479 for returnal but there's no question that the 590 is King for RT at 00:08:24.080 --> 00:08:30.840 1440p with a crown that only gets more dazzling at 4K black myth Wukong Falls 00:08:28.919 --> 00:08:35.320 below what we consider playable for an intense action game don't worry we'll 00:08:32.680 --> 00:08:39.760 talk about AI upscaling later cuz first dang look at this thing maintaining 00:08:37.680 --> 00:08:44.519 performance in the 50s at these settings in this economy dang NVIDIA that's 00:08:42.560 --> 00:08:49.200 pretty impressive and if you care more about absolute Cinema then frame rate 00:08:46.920 --> 00:08:53.480 well it holds above 30 FPS in Allen wake 2 which should go great with your 00:08:51.240 --> 00:08:58.720 popcorn f124 and returnal are similar stories as the 1440p results just with 00:08:56.240 --> 00:09:03.279 more pixels and fewer FPS all of this taken together means we're looking at a 00:09:00.519 --> 00:09:08.720 greater than 30% lead over last gen at a 25% higher price meaning the new RT 00:09:06.079 --> 00:09:12.959 cores are providing some benefit but it's pretty small compared to the impact 00:09:10.279 --> 00:09:16.720 of NVIDIA just plunking in more of them this is obviously a downer compared to 00:09:14.760 --> 00:09:20.640 the good old days when we used to get yearly GPU refreshes with dramatic 00:09:18.839 --> 00:09:24.320 improvements to Performance per dollar but it's clear that unless cuttingedge 00:09:22.200 --> 00:09:30.440 semiconductor manufacturing miraculously gets cheaper those days are never coming 00:09:26.920 --> 00:09:32.880 back so if we compare this more to say 00:09:30.440 --> 00:09:38.399 adding a second card an SLI a feature NVIDIA no longer supports then the glass 00:09:35.360 --> 00:09:40.200 half full take is hey at least it costs 00:09:38.399 --> 00:09:45.839 less than 2490 but NVIDIA still has some tricks up 00:09:43.320 --> 00:09:49.600 their sleeve holy heck that's one tight leather jacket how the heck did you fit 00:09:47.240 --> 00:09:53.480 that stuff in those sleeves with dlss 4 and multiframe gen by making use of flip 00:09:52.079 --> 00:09:59.480 metering and swapping them from convolution neural networks to Transformer based 00:09:56.720 --> 00:10:04.680 models there lot of words again lots of words again let's break it down dlss 4 00:10:02.480 --> 00:10:09.279 is NVIDIA's latest Suite of AI enhancements and it's the biggest change 00:10:06.680 --> 00:10:14.680 in years previous versions of dlss included a convolutional neural network 00:10:11.680 --> 00:10:17.640 or CNN a CNN can be thought of as a 00:10:14.680 --> 00:10:22.120 series of filters that look for specific details when used for image processing 00:10:20.360 --> 00:10:27.880 one layer could be looking for vertical edges one for horizontal edges and one 00:10:24.839 --> 00:10:29.320 for contrast Etc the neural network then 00:10:27.880 --> 00:10:33.760 observes the results from the fil filters and can use that information to 00:10:31.720 --> 00:10:39.680 identify things like if an image contains a dog or a stop sign seems 00:10:36.160 --> 00:10:41.639 convoluted well it literally is on 4,000 00:10:39.680 --> 00:10:45.600 Series gpus this information was combined with an optical flow 00:10:43.560 --> 00:10:51.160 accelerator that interpreted the motion in the scene to upscale or generate 00:10:47.839 --> 00:10:54.120 frames so why the switch scaling each 00:10:51.160 --> 00:10:58.800 filter can only scan and compute a small number of pixels at a time when we have 00:10:56.519 --> 00:11:04.160 millions of pixels dozens of times per second increasing performance can be 00:11:01.040 --> 00:11:06.760 tough so dlss uses a new Transformer 00:11:04.160 --> 00:11:11.800 model which as NVIDIA explains allows them to evaluate the relative importance 00:11:08.760 --> 00:11:14.839 of each pixel across an entire frame and 00:11:11.800 --> 00:11:16.639 over multiple frames to achieve a deeper 00:11:14.839 --> 00:11:21.040 understanding of the scenes that offers greater stability reduced ghosting 00:11:18.600 --> 00:11:25.800 higher detail in motion and smoother edges they also scale better which is 00:11:23.720 --> 00:11:30.959 part of why they have become so heavily used in things like large language 00:11:27.440 --> 00:11:33.519 models Transformer is the T in chat GPT 00:11:30.959 --> 00:11:38.440 so well a CNN can see this picture and say there is a cat there is a product 00:11:35.800 --> 00:11:44.480 from LTT Store.com a Transformer might say there is a cat enjoying the premium 00:11:41.079 --> 00:11:46.920 CRT themed pet cave from LTT 00:11:44.480 --> 00:11:52.120 store.com however while a Transformer can process complete images faster it 00:11:49.639 --> 00:11:56.200 does require more data for training and honestly with the side by-side 00:11:53.680 --> 00:11:59.880 comparisons it is tough to tell the difference in image quality between the 00:11:58.320 --> 00:12:05.079 two models like really tough there are clear benefits in 00:12:02.680 --> 00:12:09.760 specific areas that NVIDIA points out things like fences power lines and 00:12:07.000 --> 00:12:14.079 Barbed wires but there's still obvious artifacts when dealing with 00:12:11.240 --> 00:12:19.839 semi-transparent objects or just very busy scenes a lot of the artifacts are 00:12:16.880 --> 00:12:24.480 different than dlss 3 but are still present on the bright side at least on 00:12:22.320 --> 00:12:29.360 high-end cards the Transformer models don't show a substantial performance hit 00:12:27.399 --> 00:12:34.000 enough Theory I want to talk about m effing G multiframe gen is perhaps the 00:12:32.240 --> 00:12:37.720 most game-changing tech Landing with these new cards like the previous 00:12:35.920 --> 00:12:41.920 version of frame gen it uses AI to generate in between frames based on 00:12:39.800 --> 00:12:46.680 optical flow data and rendered frames but multiframe gen now allows users to 00:12:44.320 --> 00:12:52.279 generate up to three inet rather than just one boosting frame rates to up to 00:12:48.800 --> 00:12:55.519 four times native does it work well 00:12:52.279 --> 00:12:57.920 according to our charts yes very yes the 00:12:55.519 --> 00:13:03.040 numbers double triple and quadruple and make the 590 look absolutely ridiculous 00:13:00.920 --> 00:13:09.120 at least in the charts but as big as that bar is the real frames haven't 00:13:05.440 --> 00:13:11.600 changed so what's the deal well MFG is 00:13:09.120 --> 00:13:16.360 pretty wild dlss 3 frame required specific Optical flow accelerating 00:13:13.639 --> 00:13:19.600 Hardware on gpus combined that with the game data like depth and motion vectors 00:13:17.880 --> 00:13:25.360 to generate in between frames and it was an okay solution but you had to have two 00:13:22.639 --> 00:13:29.399 bits of Hardware processing each frame and that's just inefficient and could 00:13:27.320 --> 00:13:33.120 even cause the GPU to throttle resulting in a lower base frame rate to multiply 00:13:31.680 --> 00:13:37.480 off of that's why you didn't just see a straight doubling of frame rate when you 00:13:34.720 --> 00:13:42.040 turn frame gen on the 590 and multiframe gen is Shue A's Optical flow accelerator 00:13:40.079 --> 00:13:47.120 and instead utilize tightly integrated tensor and Cuda cores in Blackwell to 00:13:44.160 --> 00:13:51.440 run a lightweight AI Optical flow model not Hardware accelerated it's just an AI 00:13:49.279 --> 00:13:56.839 model this means that single frame gen should now run 40% faster while using 00:13:53.800 --> 00:13:59.639 30% less vram we know it works but how 00:13:56.839 --> 00:14:03.519 does it look well it depends on who you ask if you want to see your FPS number 00:14:01.720 --> 00:14:07.560 much higher then it works great nothing else sort of hacking your FPS counter 00:14:05.560 --> 00:14:12.320 will let you get nearly 600 FPS in cyberpunk but visually it's not perfect 00:14:10.079 --> 00:14:15.759 and frame gen weirdness still persists look at the combing on these crosswalks 00:14:14.079 --> 00:14:19.199 in cyberpunk or this bottle phasing in and out in the Benchmark or the doubling 00:14:17.320 --> 00:14:23.160 of the fan blades in these large HVAC units in Allen W 2 there's obvious 00:14:21.440 --> 00:14:26.480 artifacting around the player model and around the edge of your flashlight which 00:14:24.519 --> 00:14:30.920 is sadly exactly where you will be looking 100% of the time and it's worth 00:14:29.440 --> 00:14:34.800 noting that these artifacts are not present when we're just using dlss for 00:14:32.800 --> 00:14:39.160 upscaling curiously while cyberpunk and Allen Wick both updated with explicit 00:14:36.680 --> 00:14:43.959 support for multiframe Gen the feature can be forced in any dlss 3 single frame 00:14:42.000 --> 00:14:48.199 gen supported games through the NVIDIA driver and the game that fared the best 00:14:46.240 --> 00:14:52.079 was Dragon Age Veil guard the world's full of magic so who's to say exactly 00:14:50.000 --> 00:14:56.519 what a Vortex of Shadow is supposed to look like and since it already ran at a 00:14:54.279 --> 00:15:00.920 solid 70 FPS with all the settings cranked no frame gen it kept input put 00:14:59.199 --> 00:15:05.600 latency manageable which is really important because if your base FPS is 00:15:03.040 --> 00:15:10.560 only 30 frames well frame gen will make it look smooth but you'll observe many 00:15:07.639 --> 00:15:15.360 visual anomalies and latency is still dictated by your true frame rate meaning 00:15:12.759 --> 00:15:19.240 that the game feels far less responsive than it looks like it should be the good 00:15:17.279 --> 00:15:23.199 news says NVIDIA is that MFG at least isn't adding any latency compared to the 00:15:21.560 --> 00:15:27.680 base frame rate but we felt like it would be a good idea to verify that and 00:15:25.160 --> 00:15:31.920 verify we did using our trusty lat our click to Photon test result showed that 00:15:29.560 --> 00:15:36.680 frame gen does not increase latency over native with Reflex on in fact it seems 00:15:34.480 --> 00:15:39.600 to actually lower the latency slightly it doesn't really make sense to us so 00:15:38.199 --> 00:15:43.519 we're going to Chu that up to sampling error but if there is any effect on 00:15:41.399 --> 00:15:47.720 latency it's so minor that it's not noticeable compared to our total system 00:15:45.240 --> 00:15:51.959 latency and that is mighty impressive but the issue remains you can't beat 00:15:50.040 --> 00:15:56.360 your base frame rat's latency and when it's disconnected from the motion you 00:15:53.360 --> 00:15:57.839 see on screen it almost makes it worse 00:15:56.360 --> 00:16:01.440 perhaps the situation will improve with reflex 2 but that's not here yet so we 00:16:00.000 --> 00:16:05.240 aren't going to dwell on it even if the tech is really cool our take on 00:16:03.440 --> 00:16:10.440 multiframe gen right now is that it has the same key flaws before it's a win 00:16:07.880 --> 00:16:14.839 more feature that works the best when it makes the least sense to use which means 00:16:12.839 --> 00:16:18.959 it is definitely not the Silver Bullet that NVIDIA's graphs make it out to be 00:16:17.160 --> 00:16:22.360 with all this power it would make a lot more sense to use the 590 to make some 00:16:20.480 --> 00:16:26.800 money right so let's talk productivity hey you might not know me 00:16:24.639 --> 00:16:30.880 but I do this kind of thing and this kind of thing and NVIDIA's new 00:16:28.920 --> 00:16:34.639 architecture has benefits for me too the encoder and decoders provide support for 00:16:32.759 --> 00:16:38.279 422 chroma sampling which will make working with high-end video files much 00:16:36.480 --> 00:16:42.759 faster especially for multi- camera video edits the encoders also provide 00:16:40.560 --> 00:16:46.800 better quality at smaller file sizes sadly we can't verify that for you today 00:16:44.720 --> 00:16:50.519 as we are currently re-evaluating our encoding benchmarks but the New Media 00:16:48.720 --> 00:16:55.199 engine is almost certainly playing a role in Puget bench where we see a nice 00:16:52.560 --> 00:17:00.720 9% bump in Premier Pro performance and an even nicer nearly 20% Improvement to 00:16:58.120 --> 00:17:04.799 Vinci resolve when compared to the 490 in blender NVIDIA has us considering 00:17:02.600 --> 00:17:08.600 finding a new Benchmark as the 590 has brought Barber Shop render times to less 00:17:06.959 --> 00:17:14.760 than half a minute more than double the speed of the 3090 TI nice overw editing 00:17:12.400 --> 00:17:20.360 transition here double the 3090 TI you say AI nerds Rejoice if you're like me 00:17:17.959 --> 00:17:24.439 and for your sake I hope you're not uh you've been dying for NVIDIA to release 00:17:22.039 --> 00:17:29.160 a new 32 gig consumer card so with all the bragging and vide been doing about 00:17:26.000 --> 00:17:31.480 AI tops I'm expecting some big numbers 00:17:29.160 --> 00:17:37.080 and in the proon textt benchmarks what the number is not big sure the 590 is 00:17:35.080 --> 00:17:41.240 still the best card on the charts but I was expecting more than this we'll see 00:17:39.200 --> 00:17:47.520 roughly 20 to 30% improvement over the 4090 depending on the Benchmark and 60 00:17:44.039 --> 00:17:49.960 to 70% over the 390 TI I can see why 00:17:47.520 --> 00:17:55.080 they keep talking about AI tops and not specific performance in ml perf the 00:17:52.880 --> 00:17:59.480 story remains largely the same in the time to First token and the token 00:17:56.960 --> 00:18:04.120 generation rate benchmarks for image generation Our preferred proon Benchmark 00:18:01.559 --> 00:18:08.720 doesn't support the 59 yet so we tested using the Benchmark provided by NVIDIA 00:18:06.200 --> 00:18:13.840 prepare your salt grains for the taking in the proen flux fp8 image generation 00:18:11.760 --> 00:18:18.080 the 590 leads by a margin in line with the rest of our benchmarks but when we 00:18:15.760 --> 00:18:22.120 switch to fp4 Precision the 590 shows how powerful the native Hardware support 00:18:20.039 --> 00:18:26.240 can be taking less than one quarter of the time to generate images compared to 00:18:23.919 --> 00:18:30.440 the 490 we'd love to see how this fares against the 390 TI but this NVIDIA 00:18:28.480 --> 00:18:34.520 provides Benchmark doesn't support older cards I was expecting AI to be the place 00:18:32.840 --> 00:18:41.120 where this card really shines but I guess potential buyers will have to settle for just having the best consumer 00:18:38.880 --> 00:18:45.400 friendly grade card for AI we'll call it the Nifty 00:18:42.640 --> 00:18:49.400 FY somehow there is still more review to go good thing this is so digestible and 00:18:47.200 --> 00:18:52.799 uncomplicated so far right Blackwell's main efficiency improvements over Ada 00:18:51.120 --> 00:18:57.679 seem to come from what they're calling Max Q functionality it boils down to a 00:18:55.039 --> 00:19:01.880 few small but significant changes improvements to power thanks to an 00:18:59.480 --> 00:19:06.120 additional power Rail and improved logic allows more of the GPU to switch to a 00:19:03.840 --> 00:19:09.720 low power State more rapidly in CPU bound or frame cap scenarios this could 00:19:08.120 --> 00:19:14.600 help save some power especially on mobile chips but in GPU bound full load 00:19:12.600 --> 00:19:22.400 scenarios this monster will absolutely draw its fully rated 575 Watts including 00:19:18.400 --> 00:19:23.840 transient spikes of as high as 637 watts 00:19:22.400 --> 00:19:29.159 and you can see even in real world gaming it will pull space heater levels 00:19:25.760 --> 00:19:31.720 of Power with an average of 554 w an 00:19:29.159 --> 00:19:35.480 f124 and to manage all that power they had to make a unique cooler design while 00:19:34.120 --> 00:19:39.960 everyone else was making four slot behemoths NVIDIA built something very 00:19:37.240 --> 00:19:44.440 different and man does it look good not only is it classy it's also Innovative 00:19:42.159 --> 00:19:47.520 over ambitious even the main board of the video card is just the middle 00:19:45.799 --> 00:19:50.440 section and the outputs in PCIe connector are on daughterboards 00:19:48.799 --> 00:19:54.600 connected via what they call a flexible PCB it's like a stiff ribbon cable I 00:19:52.480 --> 00:19:59.240 guess this allows for a double flowthrough design with fans blowing 00:19:56.480 --> 00:20:04.039 through dense heat sinks and odly the fans run quiet too quiet even so like 00:20:02.240 --> 00:20:07.520 the 4090 most of the time the loudest thing about the card will be its coil 00:20:05.679 --> 00:20:12.039 wine it's not the worst we've heard but it's noticeable and if you expected it 00:20:09.320 --> 00:20:15.880 to run cooler than the 4090 Founders uh it doesn't but given how much smaller it 00:20:14.360 --> 00:20:19.760 is not to mention the enormous thermal load it's dealing with I'd say it's 00:20:17.559 --> 00:20:23.200 doing a great job but the new cooler style brings new build considerations we 00:20:21.840 --> 00:20:26.360 know that flowr coolers can have a noticeable impact on CPU temps 00:20:25.080 --> 00:20:30.760 especially for those using tower heat sinks so is a double flow through double 00:20:28.240 --> 00:20:35.400 bad we took the 590 and the 490 Fe and put them in a corser 4000d and even with 00:20:33.360 --> 00:20:40.919 both running at 450 watts to control the experiment our poor noctua nhd15 saw CPU 00:20:38.520 --> 00:20:45.240 temps that are roughly 3° higher in both synthetic and gaming workloads a CPU 00:20:43.400 --> 00:20:49.720 Cooler upgrade perhaps an intake mounted radiator could be in order for some 00:20:46.960 --> 00:20:53.919 folks wow that was a lot to talk about so what's our conclusion well if you're 00:20:51.679 --> 00:20:58.039 professional game developer or any other professional or basically anyone who can 00:20:56.200 --> 00:21:02.880 use the new performance and especially the new features to make money it's a 00:21:00.559 --> 00:21:09.200 no-brainer as for the gamers well if the 490 was stupid stupid price but stupid 00:21:06.120 --> 00:21:11.559 performance the 590 is stupider it 00:21:09.200 --> 00:21:16.840 provides 30ish per more performance by using 33% more hardware and 30ish per 00:21:14.200 --> 00:21:21.960 more power and it gets a roughly 25% price increase it's a 4090 plus plus on 00:21:20.559 --> 00:21:26.640 the one hand you could look at this and say wow this doesn't look good for the 00:21:24.760 --> 00:21:30.919 rest of the 50 Series lineup but it's also worth considering that the 49 was 00:21:29.120 --> 00:21:34.760 kind of an outlier for 40 series offering a huge boost over its 00:21:32.960 --> 00:21:38.880 predecessor with the rest offering smaller upgrades so if you're on 30 00:21:36.720 --> 00:21:43.679 series and still not in the mega mega baller income bracket I guess all we can 00:21:41.080 --> 00:21:49.400 do is wait and see see if lus Has the strength to do this segue who 00:21:46.679 --> 00:21:54.080 our Squarespace building a userfriendly visually captivating website for your 00:21:51.440 --> 00:21:58.039 business is not particularly easy is what I used to say but with Squarespace 00:21:56.480 --> 00:22:02.360 you're set up with a blueprint for success their new guided design system 00:22:00.240 --> 00:22:05.679 aptly named Squarespace blueprint helps you build your own unique online 00:22:04.000 --> 00:22:09.159 presence from the ground up with professionally curated layouts and 00:22:07.159 --> 00:22:12.600 styling options then you can dive into the details with their fluid engine 00:22:10.720 --> 00:22:15.760 editor which features intuitive drag and drop building blocks once you're happy 00:22:14.279 --> 00:22:19.000 with your marvelous creation you can then use their comprehensive email 00:22:17.360 --> 00:22:23.080 marketing tools to engage with your audience generate leads and make sales 00:22:21.480 --> 00:22:26.640 then it's just time to watch the D roll in or you could go deeper with their 00:22:24.880 --> 00:22:29.960 Advanced analytics tools that let you see what's working and what you need to 00:22:28.200 --> 00:22:35.840 tweak for your next campaign so don't wait go visit squarespace.com 00:22:32.320 --> 00:22:38.440 to get a free trial and to receive 10% 00:22:35.840 --> 00:22:45.159 off your first purchase of a website or domain if you like this video check out 00:22:42.279 --> 00:22:51.840 the one get on Hing lot of big nuggets in 00:22:48.480 --> 00:22:54.840 there I wish I could eat a nugget right 00:22:51.840 --> 00:22:54.840 now