{"video_id":"q0Kn2hL9ZJQ","title":"10 YEARS of NVIDIA Video Cards Compared!","channel":"Linus Tech Tips","show":"Linus Tech Tips","published_at":"2018-05-06T14:53:29Z","duration_s":863,"segments":[{"start_s":0.08,"end_s":7.68,"text":"Now, as most of you probably know, NVIDIA acquired our company back in","speaker":null,"is_sponsor":0},{"start_s":4.319,"end_s":10.0,"text":"early Q2 of 2016. So, you can imagine","speaker":null,"is_sponsor":0},{"start_s":7.68,"end_s":16.24,"text":"how upset our Emerald overlords were when we decided to test 10 years worth","speaker":null,"is_sponsor":0},{"start_s":13.04,"end_s":19.92,"text":"of AMD video cards before giving the","speaker":null,"is_sponsor":0},{"start_s":16.24,"end_s":22.48,"text":"green team the same treatment. Well, we","speaker":null,"is_sponsor":0},{"start_s":19.92,"end_s":28.56,"text":"finally put out the fire in our server room and just in time to bring you the","speaker":null,"is_sponsor":0},{"start_s":25.439,"end_s":32.079,"text":"other side of the GPU performance coin.","speaker":null,"is_sponsor":0},{"start_s":28.56,"end_s":34.1,"text":"Welcome to NVIDIA video cards through","speaker":null,"is_sponsor":0},{"start_s":32.079,"end_s":39.069,"text":"the","speaker":null,"is_sponsor":0},{"start_s":42.12,"end_s":50.239,"text":"ages. Rockat's Plus Force FX features","speaker":null,"is_sponsor":0},{"start_s":46.239,"end_s":52.32,"text":"pressure sensitive QWES as keys which","speaker":null,"is_sponsor":0},{"start_s":50.239,"end_s":57.36,"text":"can be configured in three different modes for multiple functions. Check it","speaker":null,"is_sponsor":0},{"start_s":54.64,"end_s":62.559,"text":"out now at the link below. This is going to be a long video, so let's get right","speaker":null,"is_sponsor":0},{"start_s":59.52,"end_s":65.6,"text":"into it. We used a 5960X test bench with","speaker":null,"is_sponsor":0},{"start_s":62.559,"end_s":68.479,"text":"64 gigs of DDR4 RAM to isolate the","speaker":null,"is_sponsor":0},{"start_s":65.6,"end_s":74.88,"text":"performance of the graphics cards. And we used the same benchmarks as AMD GPUs","speaker":null,"is_sponsor":0},{"start_s":71.92,"end_s":81.84,"text":"through the ages for consistency. First up is NVIDIA's ancient flagship, the","speaker":null,"is_sponsor":0},{"start_s":78.08,"end_s":84.56,"text":"8800 GTX. This card was freaking","speaker":null,"is_sponsor":0},{"start_s":81.84,"end_s":92.28,"text":"revolutionary. It brought us an all new architecture called Tesla in the biggest","speaker":null,"is_sponsor":0},{"start_s":87.759,"end_s":95.92,"text":"chip we had ever seen in a GPU code name","speaker":null,"is_sponsor":0},{"start_s":92.28,"end_s":99.759,"text":"G80. It was built on the same 90mm","speaker":null,"is_sponsor":0},{"start_s":95.92,"end_s":104.0,"text":"process node as its predecessor G71, but","speaker":null,"is_sponsor":0},{"start_s":99.759,"end_s":107.2,"text":"was over twice as big with almost 2 and","speaker":null,"is_sponsor":0},{"start_s":104.0,"end_s":110.24,"text":"1/2 times as many transistors devoted to","speaker":null,"is_sponsor":0},{"start_s":107.2,"end_s":112.56,"text":"gaming performance, enabling new DirectX","speaker":null,"is_sponsor":0},{"start_s":110.24,"end_s":117.52,"text":"10 features like SM 4.0 0 and a reduction in API overhead. CUDA, which","speaker":null,"is_sponsor":0},{"start_s":115.68,"end_s":122.719,"text":"is still very much in use today and enabled developers to run generalpurpose","speaker":null,"is_sponsor":0},{"start_s":119.84,"end_s":128.479,"text":"code on the graphics chip and the ability to run three cards at once in","speaker":null,"is_sponsor":0},{"start_s":125.96,"end_s":134.56,"text":"SLI, which isn't to say that its standalone performance wasn't impressive","speaker":null,"is_sponsor":0},{"start_s":130.64,"end_s":139.12,"text":"in its own right. G80 with its 768 megs","speaker":null,"is_sponsor":0},{"start_s":134.56,"end_s":142.319,"text":"of GDDR3 RAM on a wide 384-bit bus","speaker":null,"is_sponsor":0},{"start_s":139.12,"end_s":145.36,"text":"smacked around even two of NVIDIA's","speaker":null,"is_sponsor":0},{"start_s":142.319,"end_s":148.08,"text":"previous flagships in SLI. And all of","speaker":null,"is_sponsor":0},{"start_s":145.36,"end_s":152.56,"text":"that for a measly $100 more, which almost makes me feel better about the","speaker":null,"is_sponsor":0},{"start_s":149.84,"end_s":157.76,"text":"cash grab 8800 Ultra that they released 6 months later for $230 more that was","speaker":null,"is_sponsor":0},{"start_s":155.519,"end_s":165.36,"text":"basically the same thing with a fancy plastic shroud on it. The GeForce 8800","speaker":null,"is_sponsor":0},{"start_s":161.8,"end_s":168.56,"text":"GT wasn't strictly speaking a new","speaker":null,"is_sponsor":0},{"start_s":165.36,"end_s":171.519,"text":"flagship, but rather a weird Tesla","speaker":null,"is_sponsor":0},{"start_s":168.56,"end_s":178.16,"text":"architecture refresh. It was the first built by NVIDIA on TSMC's shiny new 65","speaker":null,"is_sponsor":0},{"start_s":175.2,"end_s":184.64,"text":"nanometer process. This node shrink allowed NVIDIA to drop power consumption","speaker":null,"is_sponsor":0},{"start_s":180.48,"end_s":187.36,"text":"by 60 watts, reduce the die size, and","speaker":null,"is_sponsor":0},{"start_s":184.64,"end_s":192.64,"text":"actually bump up the transistor count at the same time. a modest increase that","speaker":null,"is_sponsor":0},{"start_s":190.72,"end_s":200.959,"text":"you guessed it. No, actually you probably didn't. The 8800 GT was on par","speaker":null,"is_sponsor":0},{"start_s":196.8,"end_s":203.44,"text":"or slower than its predecessor almost","speaker":null,"is_sponsor":0},{"start_s":200.959,"end_s":210.239,"text":"across the board due to its narrower memory bus and 16 fewer shader units.","speaker":null,"is_sponsor":0},{"start_s":207.04,"end_s":214.159,"text":"Where people got excited was the price.","speaker":null,"is_sponsor":0},{"start_s":210.239,"end_s":218.4,"text":"8800 GT was less than half the price of","speaker":null,"is_sponsor":0},{"start_s":214.159,"end_s":221.28,"text":"8800 GTX. It was single slot, enabling","speaker":null,"is_sponsor":0},{"start_s":218.4,"end_s":226.64,"text":"MATX users to run to an SLI with additional expansion cards, and it also","speaker":null,"is_sponsor":0},{"start_s":223.599,"end_s":229.44,"text":"featured an onchip display engine.","speaker":null,"is_sponsor":0},{"start_s":226.64,"end_s":235.76,"text":"Finally, it was the first consumer GPU with a PCI Express 2.0 interface, which","speaker":null,"is_sponsor":0},{"start_s":232.64,"end_s":238.319,"text":"still to this day is barely saturated by","speaker":null,"is_sponsor":0},{"start_s":235.76,"end_s":243.12,"text":"flagship graphics cards. The last flagship silicon before NVIDIA","speaker":null,"is_sponsor":0},{"start_s":240.48,"end_s":248.56,"text":"transitioned to the GTX insert three numbers and maybe a couple letters here","speaker":null,"is_sponsor":0},{"start_s":245.0,"end_s":252.159,"text":"nomenclature was G92, the beating heart","speaker":null,"is_sponsor":0},{"start_s":248.56,"end_s":255.04,"text":"of the 9800 GTX Plus, a skew that","speaker":null,"is_sponsor":0},{"start_s":252.159,"end_s":262.479,"text":"lowered NVIDIA's cost and bumped up performance over the architecturally","speaker":null,"is_sponsor":0},{"start_s":257.68,"end_s":265.919,"text":"identical 9800 GTX by shrinking from 65","speaker":null,"is_sponsor":0},{"start_s":262.479,"end_s":270.56,"text":"to 55 nm and boosting up the clock","speaker":null,"is_sponsor":0},{"start_s":265.919,"end_s":272.8,"text":"speed. Now we get to the GTX 280. Yes,","speaker":null,"is_sponsor":0},{"start_s":270.56,"end_s":276.639,"text":"we know it actually came out before that last one I was talking about, but deal","speaker":null,"is_sponsor":0},{"start_s":274.88,"end_s":285.24,"text":"with it. So, forget everything we said earlier about 55 nanmter. The 280 built","speaker":null,"is_sponsor":0},{"start_s":280.16,"end_s":290.4,"text":"on the mature 65nm process had a massive","speaker":null,"is_sponsor":0},{"start_s":285.24,"end_s":293.68,"text":"576 mm die with 1.4 billion transistors.","speaker":null,"is_sponsor":0},{"start_s":290.4,"end_s":297.12,"text":"And to put that insanity in context,","speaker":null,"is_sponsor":0},{"start_s":293.68,"end_s":298.36,"text":"AMD's closest competitor, the 4870 had a","speaker":null,"is_sponsor":0},{"start_s":297.12,"end_s":304.24,"text":"measly 959 million. And further driving up","speaker":null,"is_sponsor":0},{"start_s":301.52,"end_s":311.52,"text":"complexity, it shipped with a minimum of 1 Gigabyte of graphics memory on a huge","speaker":null,"is_sponsor":0},{"start_s":308.44,"end_s":315.28,"text":"512bit bus, something NVIDIA hasn't done","speaker":null,"is_sponsor":0},{"start_s":311.52,"end_s":317.919,"text":"before or since on a consumer card. The","speaker":null,"is_sponsor":0},{"start_s":315.28,"end_s":322.479,"text":"good news is that it absolutely crushed their previous generation efforts, even","speaker":null,"is_sponsor":0},{"start_s":320.4,"end_s":328.16,"text":"delivering ample performance for stereoscopic 3D gaming, a big push for","speaker":null,"is_sponsor":0},{"start_s":325.36,"end_s":332.479,"text":"NVIDIA at that time. We'll skip over the 300 series because all of these were","speaker":null,"is_sponsor":0},{"start_s":330.24,"end_s":339.44,"text":"just crappy rebrands for notebooks and Best Buy PCs, meaning that the next real","speaker":null,"is_sponsor":0},{"start_s":335.199,"end_s":342.88,"text":"GPU in line is the GTX 480 based on the","speaker":null,"is_sponsor":0},{"start_s":339.44,"end_s":346.0,"text":"then all-new and now infamous Fermy","speaker":null,"is_sponsor":0},{"start_s":342.88,"end_s":348.72,"text":"architecture. This was NVIDIA's first","speaker":null,"is_sponsor":0},{"start_s":346.0,"end_s":353.44,"text":"swing at DirectX11 support, which is still widely used today. And this is","speaker":null,"is_sponsor":0},{"start_s":351.68,"end_s":359.44,"text":"becoming a bit of a common theme. But while the new 40nm process allowed it to","speaker":null,"is_sponsor":0},{"start_s":356.479,"end_s":368.8,"text":"be smaller than the GTX 280, it was still huge with a 529 millimeter die, a","speaker":null,"is_sponsor":0},{"start_s":364.759,"end_s":372.56,"text":"wamping 3 billion transistors, and","speaker":null,"is_sponsor":0},{"start_s":368.8,"end_s":375.759,"text":"support for DDDR5 memory, which again is","speaker":null,"is_sponsor":0},{"start_s":372.56,"end_s":378.16,"text":"still widely used today. Interesting","speaker":null,"is_sponsor":0},{"start_s":375.759,"end_s":384.319,"text":"fact, NVIDIA apparently thought it performed so well not just in gaming but","speaker":null,"is_sponsor":0},{"start_s":381.6,"end_s":390.88,"text":"also in compute which was a big focus for Fermy that the largest configuration","speaker":null,"is_sponsor":0},{"start_s":387.44,"end_s":393.68,"text":"of GF100 silicon never actually made its","speaker":null,"is_sponsor":0},{"start_s":390.88,"end_s":399.84,"text":"way into a shipping product. Or maybe that was due to power constraints. Cuz","speaker":null,"is_sponsor":0},{"start_s":396.12,"end_s":402.0,"text":"uh yeah, even with the theoretical","speaker":null,"is_sponsor":0},{"start_s":399.84,"end_s":407.36,"text":"efficiency advantage of a smaller node, the 480 was such a barbecue that NVIDIA","speaker":null,"is_sponsor":0},{"start_s":405.12,"end_s":414.0,"text":"provided a convenient grill on the face of the card to cook your eggs. Our next","speaker":null,"is_sponsor":0},{"start_s":410.88,"end_s":416.96,"text":"contestant is the GTX 580, and it","speaker":null,"is_sponsor":0},{"start_s":414.0,"end_s":423.68,"text":"featured an optimized and fully enabled Fermy dubbed GF 110. We've got most of","speaker":null,"is_sponsor":0},{"start_s":420.96,"end_s":429.28,"text":"the usual improvements here. more VRAMm, higher clock speeds, and a slightly","speaker":null,"is_sponsor":0},{"start_s":426.319,"end_s":434.24,"text":"smaller die, but nothing as exciting as video card cooking, which we mean in a","speaker":null,"is_sponsor":0},{"start_s":431.599,"end_s":439.639,"text":"good way. 580 performed noticeably better than 480 while also managing","speaker":null,"is_sponsor":0},{"start_s":436.72,"end_s":445.599,"text":"lower temperatures, noise, and power consumption. It was kind of like the GTX","speaker":null,"is_sponsor":0},{"start_s":443.0,"end_s":449.919,"text":"480S, if that makes sense to the Apple fans out there. In the seventh corner of","speaker":null,"is_sponsor":0},{"start_s":448.319,"end_s":457.44,"text":"our showdown, this must be a weird looking ring. We have the GTX 680. Ah,","speaker":null,"is_sponsor":0},{"start_s":453.759,"end_s":460.8,"text":"yes. Goodbye Fermy. Hello Kepler. A lot","speaker":null,"is_sponsor":0},{"start_s":457.44,"end_s":465.039,"text":"of things changed here. 4K 60 Hz output","speaker":null,"is_sponsor":0},{"start_s":460.8,"end_s":467.759,"text":"was enabled via DisplayPort 1.2. PCI","speaker":null,"is_sponsor":0},{"start_s":465.039,"end_s":476.479,"text":"Express 3.0 burst onto the scene with double the bandwidth of version two. The","speaker":null,"is_sponsor":0},{"start_s":470.4,"end_s":480.479,"text":"die size shrunk way down to 294 mm. And","speaker":null,"is_sponsor":0},{"start_s":476.479,"end_s":482.879,"text":"the GDDR5 got kicked up a notch to 6 GHz","speaker":null,"is_sponsor":0},{"start_s":480.479,"end_s":490.12,"text":"thanks to a newly redesigned memory controller. Finally, we got it. Wait a","speaker":null,"is_sponsor":0},{"start_s":486.16,"end_s":493.4,"text":"minute. We dropped down to a 256 bit","speaker":null,"is_sponsor":0},{"start_s":490.12,"end_s":497.039,"text":"bus. Hold on a sec. Look at that code","speaker":null,"is_sponsor":0},{"start_s":493.4,"end_s":500.16,"text":"name. NVIDIA top tier chips usually end","speaker":null,"is_sponsor":0},{"start_s":497.039,"end_s":503.68,"text":"in a zero. Well, what they did was they","speaker":null,"is_sponsor":0},{"start_s":500.16,"end_s":506.4,"text":"used AMD's complacency and a new 28nmter","speaker":null,"is_sponsor":0},{"start_s":503.68,"end_s":513.279,"text":"process along with some dynamic power tricks called GPU Boost to crank up","speaker":null,"is_sponsor":0},{"start_s":509.759,"end_s":516.56,"text":"performance so much that they gave us a","speaker":null,"is_sponsor":0},{"start_s":513.279,"end_s":519.519,"text":"stepdown chip as a replacement for a top","speaker":null,"is_sponsor":0},{"start_s":516.56,"end_s":524.159,"text":"tier one to save a buck and still manage to make it look like an upgrade. Not","speaker":null,"is_sponsor":0},{"start_s":522.24,"end_s":530.0,"text":"that they passed any of those savings to the consumer. It was priced at $500,","speaker":null,"is_sponsor":0},{"start_s":527.279,"end_s":536.16,"text":"same as the GTX 580. Though, I guess at least we got hardware H.264 encoding for","speaker":null,"is_sponsor":0},{"start_s":533.36,"end_s":541.839,"text":"game stream and Shadow Play, G-Sync variable refresh rate technology, and","speaker":null,"is_sponsor":0},{"start_s":538.64,"end_s":544.0,"text":"support for the Vulcan API. Now, the","speaker":null,"is_sponsor":0},{"start_s":541.839,"end_s":551.839,"text":"next logical stop in our journey might seem like GTX 780, but in February 2013,","speaker":null,"is_sponsor":0},{"start_s":548.959,"end_s":560.24,"text":"NVIDIA kicked convention and logic in the head with a Titan card based on the","speaker":null,"is_sponsor":0},{"start_s":556.399,"end_s":562.959,"text":"chip that might have been GTX 680 in a","speaker":null,"is_sponsor":0},{"start_s":560.24,"end_s":570.8,"text":"parallel universe where AMD had been keeping up, but instead it turned into","speaker":null,"is_sponsor":0},{"start_s":566.56,"end_s":572.64,"text":"780 and then 780 Ti. So, we're going to","speaker":null,"is_sponsor":0},{"start_s":570.8,"end_s":580.8,"text":"go with that since it was kind of the final form of the GK 110. Now, not much","speaker":null,"is_sponsor":0},{"start_s":577.36,"end_s":584.32,"text":"changed product or feature-wise here.","speaker":null,"is_sponsor":0},{"start_s":580.8,"end_s":588.24,"text":"Big Kepler was a much bigger, much","speaker":null,"is_sponsor":0},{"start_s":584.32,"end_s":590.08,"text":"better 680 with a 90% bigger die, more","speaker":null,"is_sponsor":0},{"start_s":588.24,"end_s":597.279,"text":"than double the transistors at a whopping 7 billion, and a $699 price tag","speaker":null,"is_sponsor":0},{"start_s":593.92,"end_s":599.76,"text":"to go with it. Ouch. At least though,","speaker":null,"is_sponsor":0},{"start_s":597.279,"end_s":607.04,"text":"this amounted to a sizable bump in performance over last generation's 680","speaker":null,"is_sponsor":0},{"start_s":604.56,"end_s":612.399,"text":"in basically every application. In another unconventional move, NVIDIA","speaker":null,"is_sponsor":0},{"start_s":609.519,"end_s":618.48,"text":"completely skipped over 800 series and went straight to 900. So, even though","speaker":null,"is_sponsor":0},{"start_s":615.2,"end_s":620.079,"text":"the 980 and the 980 Ti were 9 months","speaker":null,"is_sponsor":0},{"start_s":618.48,"end_s":625.04,"text":"apart, we're going to bunch them together cuz apparently that's what","speaker":null,"is_sponsor":0},{"start_s":622.48,"end_s":631.44,"text":"everybody's doing. So, the 900 series was actually based on the same 28nm","speaker":null,"is_sponsor":0},{"start_s":627.839,"end_s":634.56,"text":"process as Kepler. Shrinking silicon got","speaker":null,"is_sponsor":0},{"start_s":631.44,"end_s":636.959,"text":"a lot harder in the early 2010s, but 2","speaker":null,"is_sponsor":0},{"start_s":634.56,"end_s":643.68,"text":"years later, we were due for some kind of performance improvement. So, NVIDIA's","speaker":null,"is_sponsor":0},{"start_s":639.68,"end_s":646.48,"text":"engineers brought us Maxwell. These GPUs","speaker":null,"is_sponsor":0},{"start_s":643.68,"end_s":650.48,"text":"feature an integrated ARM CPU, which according to NVIDIA provides more","speaker":null,"is_sponsor":0},{"start_s":648.48,"end_s":658.079,"text":"independence from a given systems primary CPU. The 980, again a step down","speaker":null,"is_sponsor":0},{"start_s":655.36,"end_s":664.0,"text":"chip in big kid clothes, performs a little better than the Kepler GTX 780","speaker":null,"is_sponsor":0},{"start_s":660.72,"end_s":668.32,"text":"Ti, which is an impressive feat. But the","speaker":null,"is_sponsor":0},{"start_s":664.0,"end_s":672.079,"text":"big story is efficiency. It boasts an 85","speaker":null,"is_sponsor":0},{"start_s":668.32,"end_s":674.32,"text":"watt lower TDP. The 980 Ti was a bit of","speaker":null,"is_sponsor":0},{"start_s":672.079,"end_s":679.839,"text":"a different story, though. It rocked 8 billion transistors, an unheard of 601","speaker":null,"is_sponsor":0},{"start_s":677.519,"end_s":686.64,"text":"square millimeter die, NVIDIA's biggest ever, 6 gigs of VRAM, and all the API","speaker":null,"is_sponsor":0},{"start_s":683.92,"end_s":692.24,"text":"support you could ever want, except maybe DirectX12, depending on who you","speaker":null,"is_sponsor":0},{"start_s":689.12,"end_s":694.48,"text":"ask. While Maxwell was marketed as fully","speaker":null,"is_sponsor":0},{"start_s":692.24,"end_s":699.279,"text":"DirectX12 capable, the developer of Ashes of the Singularity, Oxide Games,","speaker":null,"is_sponsor":0},{"start_s":697.12,"end_s":704.0,"text":"found that Maxwell cards performed very poorly with Async compute enabled, which","speaker":null,"is_sponsor":0},{"start_s":701.839,"end_s":709.04,"text":"is kind of a big deal. So, all of this amounted to an astonishing performance","speaker":null,"is_sponsor":0},{"start_s":706.079,"end_s":714.32,"text":"improvement considering the lack of process node change, and HDMI 2.0","speaker":null,"is_sponsor":0},{"start_s":711.76,"end_s":720.959,"text":"support was a nice touch. Finally, we arrive at the modern day.","speaker":null,"is_sponsor":0},{"start_s":717.04,"end_s":724.8,"text":"The GTX 1080 and 1080 Ti based on the","speaker":null,"is_sponsor":0},{"start_s":720.959,"end_s":726.8,"text":"Pascal architecture. 10 series GPUs have","speaker":null,"is_sponsor":0},{"start_s":724.8,"end_s":734.399,"text":"brought a lot to the table, including GDDR 5X at the top end, which increased","speaker":null,"is_sponsor":0},{"start_s":730.56,"end_s":738.48,"text":"memory speeds to up to 10 GHz. Multiple","speaker":null,"is_sponsor":0},{"start_s":734.399,"end_s":741.519,"text":"DisplayPort 1.4 four ports for 8K and","speaker":null,"is_sponsor":0},{"start_s":738.48,"end_s":744.399,"text":"high refresh 4K monitor support, high","speaker":null,"is_sponsor":0},{"start_s":741.519,"end_s":750.959,"text":"bandwidth SLI bridges, not to mention the elimination of 3-way and 4-way SLI","speaker":null,"is_sponsor":0},{"start_s":748.8,"end_s":758.8,"text":"for gaming, and the smallest manufacturing node yet, 16 nanmter. 1080","speaker":null,"is_sponsor":0},{"start_s":755.2,"end_s":760.959,"text":"launched in May of 2016 with 7.2 billion","speaker":null,"is_sponsor":0},{"start_s":758.8,"end_s":765.519,"text":"transistors. And this is becoming a pattern now, a small performance","speaker":null,"is_sponsor":0},{"start_s":762.48,"end_s":768.56,"text":"improvement over the 980 Ti. Then 1080","speaker":null,"is_sponsor":0},{"start_s":765.519,"end_s":770.959,"text":"Ti followed with a much more substantial","speaker":null,"is_sponsor":0},{"start_s":768.56,"end_s":777.279,"text":"bump. One of the biggest stories here though was really compute. Pascal was","speaker":null,"is_sponsor":0},{"start_s":774.079,"end_s":781.0,"text":"designed to be a compute powerhouse with","speaker":null,"is_sponsor":0},{"start_s":777.279,"end_s":785.839,"text":"the top tier GP100 chip going actually","speaker":null,"is_sponsor":0},{"start_s":781.0,"end_s":789.279,"text":"unused in gaming products. So 1080 Ti","speaker":null,"is_sponsor":0},{"start_s":785.839,"end_s":795.6,"text":"and its Titan analog were instead based","speaker":null,"is_sponsor":0},{"start_s":789.279,"end_s":795.6,"text":"on GP 102, a new in between tier.","speaker":null,"is_sponsor":0},{"start_s":796.88,"end_s":803.279,"text":"The ingenious EWS 360AP is an enterprise","speaker":null,"is_sponsor":0},{"start_s":800.72,"end_s":808.32,"text":"class dualband access point that's ideal for large offices, schools, and","speaker":null,"is_sponsor":0},{"start_s":805.68,"end_s":814.16,"text":"hospitality guest Wi-Fi access. It features AC speeds with business class","speaker":null,"is_sponsor":0},{"start_s":811.2,"end_s":819.44,"text":"features and operates in autonomous or managed modes. It's got a 3x3 antenna","speaker":null,"is_sponsor":0},{"start_s":817.12,"end_s":823.519,"text":"design and a high-owered radio and comes with a free management software called","speaker":null,"is_sponsor":0},{"start_s":821.519,"end_s":830.16,"text":"Easy Master. You can manage anywhere from one access point to thousands. And","speaker":null,"is_sponsor":0},{"start_s":826.56,"end_s":832.399,"text":"there are no per device license fees. So","speaker":null,"is_sponsor":0},{"start_s":830.16,"end_s":837.36,"text":"that means no annual subscription, which is great for keeping the total cost of","speaker":null,"is_sponsor":0},{"start_s":834.8,"end_s":842.68,"text":"ownership down. Check it and other ingenious products out over at their","speaker":null,"is_sponsor":0},{"start_s":839.12,"end_s":845.12,"text":"website through the link in the video","speaker":null,"is_sponsor":0},{"start_s":842.68,"end_s":848.88,"text":"description. Thanks for watching NVIDIA GPUs through the ages. If this video","speaker":null,"is_sponsor":0},{"start_s":847.279,"end_s":852.88,"text":"sucked, you know what to do. But if it was awesome, get subscribed, hit the like button, or check out the link to","speaker":null,"is_sponsor":0},{"start_s":851.199,"end_s":856.16,"text":"where to buy the stuff we featured in the video description. Also linked down","speaker":null,"is_sponsor":0},{"start_s":854.8,"end_s":863.399,"text":"there is our merch store, which has cool shirts like this one, and our community","speaker":null,"is_sponsor":0},{"start_s":858.399,"end_s":863.399,"text":"forum, which you should totally join.","speaker":null,"is_sponsor":0}],"full_text":"Now, as most of you probably know, NVIDIA acquired our company back in early Q2 of 2016. So, you can imagine how upset our Emerald overlords were when we decided to test 10 years worth of AMD video cards before giving the green team the same treatment. Well, we finally put out the fire in our server room and just in time to bring you the other side of the GPU performance coin. Welcome to NVIDIA video cards through the ages. Rockat's Plus Force FX features pressure sensitive QWES as keys which can be configured in three different modes for multiple functions. Check it out now at the link below. This is going to be a long video, so let's get right into it. We used a 5960X test bench with 64 gigs of DDR4 RAM to isolate the performance of the graphics cards. And we used the same benchmarks as AMD GPUs through the ages for consistency. First up is NVIDIA's ancient flagship, the 8800 GTX. This card was freaking revolutionary. It brought us an all new architecture called Tesla in the biggest chip we had ever seen in a GPU code name G80. It was built on the same 90mm process node as its predecessor G71, but was over twice as big with almost 2 and 1/2 times as many transistors devoted to gaming performance, enabling new DirectX 10 features like SM 4.0 0 and a reduction in API overhead. CUDA, which is still very much in use today and enabled developers to run generalpurpose code on the graphics chip and the ability to run three cards at once in SLI, which isn't to say that its standalone performance wasn't impressive in its own right. G80 with its 768 megs of GDDR3 RAM on a wide 384-bit bus smacked around even two of NVIDIA's previous flagships in SLI. And all of that for a measly $100 more, which almost makes me feel better about the cash grab 8800 Ultra that they released 6 months later for $230 more that was basically the same thing with a fancy plastic shroud on it. The GeForce 8800 GT wasn't strictly speaking a new flagship, but rather a weird Tesla architecture refresh. It was the first built by NVIDIA on TSMC's shiny new 65 nanometer process. This node shrink allowed NVIDIA to drop power consumption by 60 watts, reduce the die size, and actually bump up the transistor count at the same time. a modest increase that you guessed it. No, actually you probably didn't. The 8800 GT was on par or slower than its predecessor almost across the board due to its narrower memory bus and 16 fewer shader units. Where people got excited was the price. 8800 GT was less than half the price of 8800 GTX. It was single slot, enabling MATX users to run to an SLI with additional expansion cards, and it also featured an onchip display engine. Finally, it was the first consumer GPU with a PCI Express 2.0 interface, which still to this day is barely saturated by flagship graphics cards. The last flagship silicon before NVIDIA transitioned to the GTX insert three numbers and maybe a couple letters here nomenclature was G92, the beating heart of the 9800 GTX Plus, a skew that lowered NVIDIA's cost and bumped up performance over the architecturally identical 9800 GTX by shrinking from 65 to 55 nm and boosting up the clock speed. Now we get to the GTX 280. Yes, we know it actually came out before that last one I was talking about, but deal with it. So, forget everything we said earlier about 55 nanmter. The 280 built on the mature 65nm process had a massive 576 mm die with 1.4 billion transistors. And to put that insanity in context, AMD's closest competitor, the 4870 had a measly 959 million. And further driving up complexity, it shipped with a minimum of 1 Gigabyte of graphics memory on a huge 512bit bus, something NVIDIA hasn't done before or since on a consumer card. The good news is that it absolutely crushed their previous generation efforts, even delivering ample performance for stereoscopic 3D gaming, a big push for NVIDIA at that time. We'll skip over the 300 series because all of these were just crappy rebrands for notebooks and Best Buy PCs, meaning that the next real GPU in line is the GTX 480 based on the then all-new and now infamous Fermy architecture. This was NVIDIA's first swing at DirectX11 support, which is still widely used today. And this is becoming a bit of a common theme. But while the new 40nm process allowed it to be smaller than the GTX 280, it was still huge with a 529 millimeter die, a wamping 3 billion transistors, and support for DDDR5 memory, which again is still widely used today. Interesting fact, NVIDIA apparently thought it performed so well not just in gaming but also in compute which was a big focus for Fermy that the largest configuration of GF100 silicon never actually made its way into a shipping product. Or maybe that was due to power constraints. Cuz uh yeah, even with the theoretical efficiency advantage of a smaller node, the 480 was such a barbecue that NVIDIA provided a convenient grill on the face of the card to cook your eggs. Our next contestant is the GTX 580, and it featured an optimized and fully enabled Fermy dubbed GF 110. We've got most of the usual improvements here. more VRAMm, higher clock speeds, and a slightly smaller die, but nothing as exciting as video card cooking, which we mean in a good way. 580 performed noticeably better than 480 while also managing lower temperatures, noise, and power consumption. It was kind of like the GTX 480S, if that makes sense to the Apple fans out there. In the seventh corner of our showdown, this must be a weird looking ring. We have the GTX 680. Ah, yes. Goodbye Fermy. Hello Kepler. A lot of things changed here. 4K 60 Hz output was enabled via DisplayPort 1.2. PCI Express 3.0 burst onto the scene with double the bandwidth of version two. The die size shrunk way down to 294 mm. And the GDDR5 got kicked up a notch to 6 GHz thanks to a newly redesigned memory controller. Finally, we got it. Wait a minute. We dropped down to a 256 bit bus. Hold on a sec. Look at that code name. NVIDIA top tier chips usually end in a zero. Well, what they did was they used AMD's complacency and a new 28nmter process along with some dynamic power tricks called GPU Boost to crank up performance so much that they gave us a stepdown chip as a replacement for a top tier one to save a buck and still manage to make it look like an upgrade. Not that they passed any of those savings to the consumer. It was priced at $500, same as the GTX 580. Though, I guess at least we got hardware H.264 encoding for game stream and Shadow Play, G-Sync variable refresh rate technology, and support for the Vulcan API. Now, the next logical stop in our journey might seem like GTX 780, but in February 2013, NVIDIA kicked convention and logic in the head with a Titan card based on the chip that might have been GTX 680 in a parallel universe where AMD had been keeping up, but instead it turned into 780 and then 780 Ti. So, we're going to go with that since it was kind of the final form of the GK 110. Now, not much changed product or feature-wise here. Big Kepler was a much bigger, much better 680 with a 90% bigger die, more than double the transistors at a whopping 7 billion, and a $699 price tag to go with it. Ouch. At least though, this amounted to a sizable bump in performance over last generation's 680 in basically every application. In another unconventional move, NVIDIA completely skipped over 800 series and went straight to 900. So, even though the 980 and the 980 Ti were 9 months apart, we're going to bunch them together cuz apparently that's what everybody's doing. So, the 900 series was actually based on the same 28nm process as Kepler. Shrinking silicon got a lot harder in the early 2010s, but 2 years later, we were due for some kind of performance improvement. So, NVIDIA's engineers brought us Maxwell. These GPUs feature an integrated ARM CPU, which according to NVIDIA provides more independence from a given systems primary CPU. The 980, again a step down chip in big kid clothes, performs a little better than the Kepler GTX 780 Ti, which is an impressive feat. But the big story is efficiency. It boasts an 85 watt lower TDP. The 980 Ti was a bit of a different story, though. It rocked 8 billion transistors, an unheard of 601 square millimeter die, NVIDIA's biggest ever, 6 gigs of VRAM, and all the API support you could ever want, except maybe DirectX12, depending on who you ask. While Maxwell was marketed as fully DirectX12 capable, the developer of Ashes of the Singularity, Oxide Games, found that Maxwell cards performed very poorly with Async compute enabled, which is kind of a big deal. So, all of this amounted to an astonishing performance improvement considering the lack of process node change, and HDMI 2.0 support was a nice touch. Finally, we arrive at the modern day. The GTX 1080 and 1080 Ti based on the Pascal architecture. 10 series GPUs have brought a lot to the table, including GDDR 5X at the top end, which increased memory speeds to up to 10 GHz. Multiple DisplayPort 1.4 four ports for 8K and high refresh 4K monitor support, high bandwidth SLI bridges, not to mention the elimination of 3-way and 4-way SLI for gaming, and the smallest manufacturing node yet, 16 nanmter. 1080 launched in May of 2016 with 7.2 billion transistors. And this is becoming a pattern now, a small performance improvement over the 980 Ti. Then 1080 Ti followed with a much more substantial bump. One of the biggest stories here though was really compute. Pascal was designed to be a compute powerhouse with the top tier GP100 chip going actually unused in gaming products. So 1080 Ti and its Titan analog were instead based on GP 102, a new in between tier. The ingenious EWS 360AP is an enterprise class dualband access point that's ideal for large offices, schools, and hospitality guest Wi-Fi access. It features AC speeds with business class features and operates in autonomous or managed modes. It's got a 3x3 antenna design and a high-owered radio and comes with a free management software called Easy Master. You can manage anywhere from one access point to thousands. And there are no per device license fees. So that means no annual subscription, which is great for keeping the total cost of ownership down. Check it and other ingenious products out over at their website through the link in the video description. Thanks for watching NVIDIA GPUs through the ages. If this video sucked, you know what to do. But if it was awesome, get subscribed, hit the like button, or check out the link to where to buy the stuff we featured in the video description. Also linked down there is our merch store, which has cool shirts like this one, and our community forum, which you should totally join."}