How Nvidia Won and AMD Lost the GPU War 

If what AMD has been doing for the past 3 years has been 'focusing on AI' then I don't really have a lot of hope for a better future when they're 'focusing on gaming' because they didn't really make any footholds at all in the ai world until the last 3 months.... Despite Vega being better at compute than Pascal, and even how good Polaris was compare to the 900 series. There was no library support until second half 2018!! AMD has had better compute hardware since the 7970 but the ONLY way to leverage it has been writing in GCN assembly. Ain't nobody got time for dat except crypto miners. Because all they have to do is write one program! All serious research takes place in higher level languages. CUDA has been dominating there since forever. They should know by now it doesn't matter how good your hardware is if there is no software to leverage that advantage.

AMD can make a comeback, not through innovation, or even crafty marketing. No, they just need to let nVidia's biggest enemy come in first, beat up the king, and then AMD with good timing can just "show up to the fight." NVidia's biggest enemy is of course, itself.

Can we just appreciate how AMD has been fighting with both Intel and Nvidia in two different fields with far less cash that either one of them and dealing with all of their bullshit? Madness

Seems like I will be going to bed in 37 minutes.

it's funny when you look at Tahiti now, still a perfectly usable mid/high 1080p gpu, often trading blows with the later released 780.... where as GTX 680s and their younger 770 competition are now all in the bin.

Even if Nvidia is winning this round, I'll switch to a full AMD system for my next build, Intel/Nvidia combo isn't worth it anymore.

📂 NVIDIA └📁RTX 2018 └📁Gaming Value └⚠️ This folder is empty

I have always said AMD looks to the future more then it should. While it is good to be aware of the future and the direction things will go. You have to master the moment as well. AMD produced 6 and 8 core gaming chips long before games would use all those cores effectively. Intel mastered the moment and their quad cores ruled. The plan to introduce a new architecture every year does not impress me. As stated in this video, "immature drivers" etc... Master the moment while keeping an eye out for the future. Polaris 20 should have been the original Polaris launch. Polaris 30 should be on 7nm, which would justify the launch. Still I prefer AMD GPU's because the graphics quality is better and I like the software more. I got rid of my 1070 SC2 and got a Vega 56 Pulse for those very reasons. Very informative video. Thank you.

Hopefully AMD makes the "Zen" of GPUs and catches up

As a former employee I would say the years passed 2006 was under poor leadership. I think with the graphics and the CPU Road map as well, like Zen AMD will create a GPU Architecture that will come from left field. Don't forget this was the same AMD leadership that sold off their mobile CPU GPU business to Qualcomm during the years that you are speaking about. I think we will see a new AMD

Said it many times but will say it again anyway "I love how you entertain while you educate" you have a great talent my friend :)

it still doesn't make any sense to me that a company making the clearly superior product line lost to a company that couldn't compete. What a world.

Amazing video, as usual. I love watching the historical breakdown and at the end you even gave us a tiny bit of hope that maybe AMD will get it together, lol. I also think ATI's brand destruction didn't help anything. AMD had become known for slow, crappy CPUs and that toxic mindshare bled all over their GPU products, at least for me it certainly did. I didn't want to buy a GPU from a company that looked like it was about to go out of business, regardless of how fast it was. Zen fixed that, for now at least. Let's see if AMD capitalizes on that Zenergy to breathe some life back into their GPU's.

This video was on point. Sad but true. Prepare your wallets people, the following 3 years are going to be hell for an average consumer.

I had a discussion with a guy on the internet where we discussed the possibilities that AMD has going forward. Looking at their CPU range, they have had tremendous success with their chiplet designs in the HEDT and server market with Rome looking like it's going to be absolutely insane. I saw a leak of a 12 600 score on Cinebench with the upcoming Rome 64 core. Intel's 28 core 5 GHz managed 7700 in the same test. Sure, take it with a grain of salt, but Rome will be insane thanks to a great architecture, but more importantly, thanks to chiplets. This made me think, what about GPU chiplets? Could the reason for no Navi for highend be that there will be no monolithic successor to Vega ever and that the successor will come after the initial Navi release when AMD has found a way to fuse together 4 Navi dies using their chiplet technology from the desktop? Would that even be technologically possible? Why, why not? As it stands with AMD and GPU atm, I don't see how they could ever dethrone Nvidia without something magical like chiplets. I'd be very interested to hear your thoughts on GPU chiplets :)

This is some great tech journalism. Thanks for the great content! I remember pretty much all these cards. I've owned a lot of them! I don't have them anymore, as I sold a lot of them - but fond memories. Particularly my 5970. I crushed many a LAN party with that :D

Nvidia has won the graphics card war, but we'll see who wins the GPU war. If Nvidia is free to increase prices they could completely destroy the AiB market and could leave AMD to swoop in with cheaper APU based solutions for the average gamer. I still think it's kinda open, but I doubt AMD will even care about AiB for gamers. I still encounter PC gamers who don't know who AMD is. Most people are sheep (I know it sounds edgy, but when it comes to tech it's so true)

It's so sad to see even more clearly how AMD had a technological lead for long but never got rewarded for it. I think GCN was a huge mistake. AMD should've sticked with TeraScale for a few more years. But they bet everything on compute in an early stage. On one hand it's principally right, since they need to take risks to get more reward. On the other hand, just like the TSMC 32nm process, there were signs a rather long time before that things would not materialize as expected. From what I've heard Microsoft was working on a low-level DX12-like API already in 2010, and this is what Fermi and GCN was designed towards. But that API was scrapped for some reason. Nvidia quickly changed course, but AMD stayed. GPGPU and compute was all the rave in 2009-2011. Everybody talked about it. It would take over everything. All major tasks would be done on the graphics chip. The hype was huge. This was the next big step for computers. And I guess AMD caught on to this. But beyond the visionary talk there were very few ideas on how to practically do this. I remember that Grand Central Dispatch (GCD) was one of the big advertised features of Mac OS 10.6 Snow Leopard. If I remember correctly GPGPU was part of it. In the next release it was just another half-forgotten framework. The hype's last stand of compute and GPGPU before it died off completely, as well as the biggest practical implementation of it, were in the PS4 and Xbone. But that was out of necessity since the CPU's in them sucked. I don't know if it ever has been used. The story is similar for tablets. It would take over everything. Mobile was the final destination. Even desktop computers, those that would survive, would change to accommodate this. Sure, tablets came and added some colours and hills to the landscape, but the landscape is overall much like before. Laptop computers are still more common even if both Microsoft and Apple tried to brake them with a "mobile" redesign of their OSs. As I see it GCN is the Bulldozer of graphics. Way too early, badly implemented, and - in either case - best suited for workstations and datacenters. Unfortunately for AMD, because of said Bulldozer that came out the same year, they got into even worse financial trouble instead of flourishing after the letdown of the K10/Phenom architecture. And since CPUs are really the only thing that's made AMD money, it got priority, and all development on the Radeon side was put to an absolute minimum and they were stuck with an ill-suited architecture and no optimised software. That is perhaps the only thing I will give Nvidia credit for; they realised early, and are still to this day the only ones who's truly understood the importance of software tools and drivers for graphics. Not even 3Dfx grasped this despite they had a complete closed ecosystem. Things _could_ change though, now that AMD is finally making money again. AMD have showed signs of learning their lesson and taken a few steps in the right direction with what little they've had. GCN will have one last parade before we can welcome a fresh contender and potentially break new ground. A whole decade to hell. I cross my fingers for salvation with the new one. (grammar note: is a company referred to as 'they' or 'it' in English? I've never learned that)

Oh boy...those halcyon days when a 69% performance increase was frowned upon! Superb video bud :)

Honestly, I'd put more of a emphasis on the driver issues with the 7970 as in the modern world many people are still very comfortably using the card which on current drivers is out way ahead of the gtx680. Had AMDs drivers been properly optimised at the time, I'm sure it would have been a much bigger success.

37 minutes thankyou lord of tech analysis.

Awesome video Jim. Informative as always. I'm curious as to what is going through Wang's mind now with RTX being out now and so forth. I'd love if you had the chance to interview him one day. As well as Nvidia's CEO. Looking forward to the next video as always!

Jim coming in with the knowledge bombs!

AdoredTV posted a new video!? Click on video, pause video, SMASH THAT LIKE BOTTON! Then play video. (Edit) Just finished watching it and It was amazing as always, Keep up the good work Jim! Your by far the best Tech-tuber out there!

1 Am. Time to go bed. Thanks for the bed time story Jim.

What an excellent video. For sure one of your best yet.

Considering how massive Turing is, it seems like Nvidia is going for the big die strategy, only this time against themselves, and they seem to be failing.

Im watching this vid while waiting in line at Disneyland. Great to fill in the time with Adored Tv.

AMD: "Call an ambulance! BUT NOT FOR ME."

In Navi we trust...

Oh happy day. A video by the man himself on my very own birthday?

Dude, it's midnigth why you post a video right before i go to bed, i will watch your video tomorrow i need to sleep OK... The title makes it really hard tho, i hope i won't watch it in bed on the phone :D

What a video!What a "history" lesson!Congrats Jim!

Damn you make good videos Sir

Dual posting this to give other people a chance to read about AMD's HSA technology: (Something I will bang on about forever) What's really sad is the amount of work AMD has put into GPGPU compute over the last decade. en.wikipedia.org/wiki/Heterogeneous_System_Architecture Originally called AMD Fusion, this slowly crept towards full support in AMD's platforms. The first useful result of this was the Xbox 360 fantastic memory architecture that gave developers a lot of freedom and allowed it to keep up with the rest of the gaming world. The nVidia chip in the PS3 couldn't, with two separate lumps of RAM. It's an initiative to solve the problem of the GPU being almost a seperate computer inside your computer. With separate memory banks. One of the limiting factors in using GPU cores to perform general computing tasks is the god awful latency caused by having to copy data from main system RAM to GPU RAM. So often there is no speed increase due to that latency wiping it out. So AMD fixed the problem. Take their APUs (and almost all the consoles they've built). en.wikipedia.org/wiki/AMD_Accelerated_Processing_Unit#AMD_Heterogeneous_System_Architecture It's not "on-board graphics" anymore. It's a 4C/8T/704S heterogeneous APU. All the cores have access to all the RAM and fantastic abstracted dispatch methods. They've even extended support to HSA compatible discrete GPUs with shared virtual memory. This is why I'm so excited to see if AMD pull it all together with multi small GPU + Infinity Fabric cards with DX12 mGPU and HSA. Or the OpenGL equiv which used to be Mantle, whatever that's called. And with AMD wins in laptop and desktop computing with APUs, maybe... MAYBE we'll FINALLY see some acceleration of ordinary desktop tasks (Excel? Photo editing? Video Editing? ETCETCETC) using HSA enabled OpenCL2. And it's now in all of your smartphones too. Because really who uses desktop PCs anymore? (;-) Here's the list of companies that now implement HSA support in their respective products: AMD, ARM, Imagination Technologies, MediaTek, Qualcomm, Samsung Electronics and Texas Instruments So now Android can take advantage of HSA. All that FPU power available to more than just 3D graphics! So, Jim, can you give us a video on how AMD gave us HSA and that it doesn't really make them any money and just makes computers better? :-) A Different Jim. (Next year I'm going to save up for a Raven Ridge laptop and start writing some APU software) (Anybody got any ideas?)

Well, at least 2019 can be less of a disaster than 2014-2018. Sure, AMD recovered market share with Polaris, but Fiji and especially Vega still flopped. Not even because Vega it that terrible. No because the RTG shot itself in the foot big time with Vega. I got a reference RX Vega 56 at launch for MSRP which I undervolted to 910mv. Still hits 1400Mhz+ clocks and outperforms the stock profiles. Neither does the card throttle or does the fan spin up to the point where the card tries to take off and fly away with my old-fashioned 10kg+ case. Most Vega chips can be run with 1000mv and hit 1500MHz ish clocks. You don't even need to put the fan on "take off mode" to keep that cool enough with the terrible reference cooler. With such clocks, Vega isn't terrible against GP104. Not as good from a technical point of view, but not terrible either. It is a lot better than the stock profile which tends to throttle to values below 1200MHz, makes a lot of noise, and pulls more power... The chance for AMD is that Nvidia has a pretty bad Pascal replacement in Turing. Huge chips, little traditional performance increase and questionable ray tracing support for now. Support that will most likely not be enough for serious ray tracing, and will be superseded in a few years by much better support, thus rendering Turing obsolete. It opens up a opportunity for AMD to continue with there sweet spot strategy for one more year before they hopefully ditch GCN for something far better. I don't expect anything better from Navi than a Vega performance level chip on 7nm with GDDR6 instead of HBM2. That might actually be enough against TU-106/RTX2070. AMD could very well sell the top SKU Navi for $400, cut down for $300 and position/rebrand Polaris 10/20 below it. Nvidia can't really introduce something substantially better in mainstream than the GTX1060 because it would cannibalise the RTX2070. That opens the door for a rare AMD win. Then the rumours about a lot of funding going to the RTG are not that hidden. ;) Apparently quite a few of the Zen people already have been moved to the RTG and the overal R&D budget of AMD has been rising fast since early 2017. We can only hope that they will preform magic there again. Not only with a new gaming oriented architecture, but also making a serious push into AI. GCN is pretty good for that, but the software stacks have been an utter joke for years. Nobody will write custom software for AMD hardware to save a few bucks. They rather go to Nvidia, pay the premium because it is faster and cheaper for the company or scientific institute to get the hardware into use. Recently it seems to have gotten a bit better for AMD, but it is far to late. GCN could have broken into that market when Nvidia was buzzy with Kepler and Maxwell. GCN always was better for compute than what Nvidia had in 2013-2016, but once again they squandered there advantage...

The video helps those of us who hadn't caught on to this because, although we know the history we haven't gone into the detail to see just how bad the value is now. I have been an Nvidia buyer since 3dfx went. This video might be the thing that makes me stop buying a new card every other year (generation) and give AMD a chance when they have something new. Cheers, your videos are amazing.

Great video Jim, always enjoy the history lessons, as without history we would have no future. It's quite sad to see the state of the GPU market, prices are astronomically high, and looking at past the mid-range market it's pretty much all green. I think AMD might have learned from their past mistake though, hopefully, as many were surprised to see that they wouldn't be delivering a 7nm Vega for the pc gaming market. While it was great for HPC, there were better alternatives in the gaming space. Zen has been a huge success for them, and I hope that gives their RTG division more headroom to work with than they had before. So that they could hopefully design a gaming architecture, rather than just one and think of gamers as some...side quest. There have been some rumors circulating around that NAVI was designed with the PS5 in mind. And I'm sure Microsoft has their finger in their too somewhere. Therefore being a chip with gaming in mind primarily. Not sure how much truth there is to that, but it gives me a little bit of hope. The other thing is that they also have to figure out how to take back the market share. As having the fastest GPU doesn't necessarily mean you'll be successful. Ray tracing, RTX, Gameworks, these are tools Nvidia has used besides their GPUs to embed in the minds of gamers. I mean take a look at all those huge streamers on twitch, they've all got Nvidia GPUs in their rigs and I highly doubt its because they watched a youtube video or read in article from say, anandtech that told them to buy Nvidia.

4 years later RDNA 2 matched Nvidia, and RDNA 3 is poised to take the performance crown, going through the adoredtv backlog is always interesting

The trouble for the 7970 was that being launched on the wrong process due to the lack of 32nm availability. This means that the chip had to be cut down to fit on an older process and the power consumption was heightened for the same reason. If it had been released on 32nm it would most likely have been clocked much higher and had features that were cut off due to the manufacturing issues. I have a Radeon 7870 which lasted me quite well for several years. The mess with the drivers initially ended up with performance being gained through driver improvements as time went along. To me this made that card extremely good value. The same happened with my Radeon 3870 which I later upgraded from to a GTX 460 and had trouble noticing any real performance improvements.

2020 amd won. you can keep pining and liking rtx fanbois, but my amd fanboism will never die. ryzen and radeon will live FOREVER!

great look at the other side of history - the little-knowns that had huge ramifications. i know "what if"s are not your thing, thankfully, but... imagine AMD manages to nearly guarantee 100% multi-GPU scaling & eliminate micro-stutter through XGMI. they then just turn out 7nm Vega/Navi as efficient midrange $220 GPUs where the bulk of buyers sit, while allowing an easy/dependable upgrade path for enthusiasts via adding cards on XGMI. it'd take a lot of "teaching the consumer" through good marketing that multi-GPU was fixed/solid, but imagine the yields & volume!

I bought 3 GTX 470s around launch, ~1,320 USD. Water blocks brought that to ~1,650 USD and that setup lasted me until this June when I bought a 1070 Ti on sale for ~$475, plus a ~$90 water block. The kicker is that the only thng really holding the 470s back was the staggeringly small amount of Vram. A shame, as I have little hope that the 1070 Ti will last as long. Then again, change is the only constant, even if I utterly resent the lack of 3-way SLI/Crossfire support moving forward.

I'm liking that a few people are thinking of moving to AMD. From their drivers to how ryzen is holding. It's a small bit of mind share but it's nice to see it. Hope navi competes at some level. Love the vids Jim

So informative, great journalism. Thanks again for another awesome video.

Are you sure? Maybe high end GPU for pc. But in terms of mainstream and lower end, AMD is killing it with dedicated graphics chips... I mean, how many consoles and Mac computers has sony/Apple/Microsoft sold with AMD stuff inside?

Your perspective is immaculate. Thank you, Jim.

To everyone who says nvidia drivers are "good", "stable", etc. I switched one option ("Optimal power" to "Maximum power") and boom my GTX760 died right there. I will try Radeon now.

Wow another awesome vid - it should be impossible to put out such quality content time after time; never a bad or boring one! Do you need assistance with research or anything else? Honestly brother I'm beginning to think with a small bit of capital you could (and should) be running your own anandtech-style site and getting a lot more visibility. Your content doesn't really have a peer that I'm aware of and it's needed.

Really interesting information. Thanks and keep it up.

Hey, Jim. I don't remember which video it was that cited nVidia drivers causing the majority of Vista crashes, here's the info from another perspective, a former MS developer. He said that in his practice of debugging bluescreens, he personally never encountered a kernel bug, but that these errors are most often caused by device drivers. 6:45 - 7:10 "Why are bluescreens blue" by "Dave's Garage"

This video made me a bit sad.... Did AMD ever had the momentum though? I worked at a big online IT retailer at the time of the 4870 and 5870, and still Nvidia cards sold so much. It feels like AMD just... never managed to exploit the times they were the fastest and never became THE brand to buy. Sad.....

Personally, I'll always stick with AMD on moral grounds.

Which company has more success? Ferrari or Toyota? Navi is slated to offer 20 TFLOPS, more than the RTX 2080. All the options are within AMDs reach.

Gaming Laptop had a ATI HD 5870 in it and Desktop had two ATI HD 6870's running in crossfire until the capacitors died early this year. It lasted a long time.

O'm so glad AMD is out of the highest tier GPU market. People get what they deserve. They wanted "AMD to die" in GPU's, they got it. Nvidia performed absolutely disgraceful business processes to kill the market, knowing long term, they would win. Regardless of some slap on the wrists from various bodies. Stupid fan boys. Fan boys of any guise that want competition to die, get what they deserve. Now people can pay $1200 for a GPU so they can play games on. Good job. It's better that AMD focuses on mid range and entry level and Data Centres.

I watch for your videos every day

hey, AMD/RTG is still close 2nd, noone's even close, intel GFX make even the shittiest tier radeons look like monsters (although radeon is very good in the low-end midrange, and starts fallling back in the high end) i know, it's not any better for the consumer, but it does kinda mean nothing is off the table for Radeon yet. now let's watch.

Great video as always. A bit of a nitpick on the price chart near the end, for the GTX 1080 you gave the price of the Founder's Edition, not the MSRP, which at the very least should have been specified. It also would have been more fair to adjust prices for inflation. It makes little difference to the final point as far as mid-range GPUs go (high-end GPUs are another story - inflation-adjusted, the GeForce chips have mostly released within the same narrow price range), it would just be nice to see that extra bit of due diligence.

To be fairly honest sitting on a GTX 1060 6GB, I'm more interested in that clock bumped 12nm Polaris respin with +15% performance uplift and a getting a new FreeSync display, rather than upgrading to a GTX 2060 at higher then GTX 1070 prices...

Jim i feel sorry for your haters, I think i am going to dislike this video just because the likes to dislike ratio looks so sad. :( .... Unfortunately i like your vids before i actually look at them.

This video hasn't aged well

I still have a Sapphire HD 5850 sat somewhere. I loved that card and it was beautiful. Who remembers the Sapphire artwork on the blower shrouds?

I find it ironic now. AMD planned for the future by building graphics cards that scale better than NVIDIAs graphics cards. The recent Radeon VII proves the power that AMD has had. They only lacked good support and driver improvement. I believe that AMD will most likely win the war, even if NVIDIA has been winning the battle.

I'm not giving up on AMD, something is coming...they crushed intel...they will crush Nvidia in time.

CEO of Nvidia is the uncle of AMDs CEO. True fact. This is how they won.

Just discovered your channel yesterday and I've been binge-watching. I was always looking for a channel similar to Gaming Historian and found yours, which I find to be surprisingly even better.

To be honest, it's not really about AMD turning things around but them delivering enough performance bump over what I have and I normally upgrade with 70% to 100% performance boost over what I have at mid level pricing, so around £200 to £250. I'm on the 280x at the moment which is a rebrand of the 7970 and this have been the longest I've not upgraded my PC all my life as being a PC gamer since the mid 90's. So for me it's all about the numbers at mid range level, I don't really care what goes on at the higher end level because games don't really target them and it just becomes a bottomless pit of wasted money chashing that. For me, I'll likely upgrade to AMD's next mid range gpu not the rebrand that they seem to be coming out with anytime now but the real mid range gpu maybe next year, that over what I have should give me at least a good 70% plus performance boost over what I have and I'm fine with that. As for AMD getting back into the game at the high end, that wont happen for quite some time if ever and if it does, it wont happen till 2020 at the earliest. If AMD keeps doing well with Ryzen and takes enough market share from Intel over the next few years, that will help AMD out a lot with the gpu devision by giving them a lot more money to invest in it, problem is that it will take years before we see anything of it so Nvidia have the high end market to it's self, I also got the impression that if Ryzen wasn't a success, AMD would scale back with gaming gpu's with the intention of leaving that market, now I'm not so sure. With that said, it's hard to buy into Nvidia hardware with some of the tactics the company does over the years so over the last few years, I buy going off performance per cost so in some sense, it makes the companies compete with themselves if they want my money.

first

I like these silent coolers are not mentioned enough in RU-vid video's.

When i see the notification that AdoredTV has posted a new video, i get excited. When i see it's another tech history lesson, i can hardly contain that excitement. Thanks again Jim. Great stuff as usual.

Another gem of a content, had to watch whole video in bed. Keep it up.

Great video! Another enjoyable watch. Just curious but could you consider making a video on the history of DirectX? Why did it stop at 12?

Eh. Actually if you look through the history of most businesses, it's being second that is best. Being first means you're taking all the risks, and have no idea what works and what doesn't. Being second means you can learn from someone else's mistakes without having to pay the price of making those mistakes yourself. Being 3rd or later tends to mean you're too far behind to compete effectively. Often you see the second company in a given field becomes a bohemoth, the first goes bankrupt, and everyone else remains small an innefective.

I'm still hoping beyond hopes that a Polaris refresh basically increases everything, including memory bus width by 50%. Then they have a 6gb and 12gb option for a high end GPU with 50% more streaming processors. They will also have 50% more memory bandwidth. Heck, maybe even if they just increased the SM count by 30-35% while providing the 50% more memory and memory bandwidth so they could push more power thru those SMs and not be bound as much by memory bandwidth for those SMs. I really think something like this on a cheaper proven node (as opposed to 7nm) could provide competition near Nvidia at half the price and really put some mindshare back towards AMD. I mean if you could get 80% the performance of a 1080ti/2080 for only 300-350 bucks brand new with warranty, it would be a very compelling option.

One thing i think is interesting is how the Nvidia card always uses quite a bit less bandwidth for the same performance. How is this a thing. It seams to only work in gaming. When working scientific workload the higher physical bandwidth of AMD cards kills it, but in games, they don´t. I can´t stop thinking that Nvidia is cheating somehow. Like using a inline texture compression right in the memory buss or something like that.

One of AMD/ATI major issues was driver/game support. There's a reason for the whole "finewine" meme. They have had good hardware but always lagged behind drivers to support the hardware. Hence even if at launch the ATI/AMD Card was keeping up, they fell behind due to drivers.

Me: Well I think it's time for bed. Adored: *Oh do you now?*

I really love the fact that you read each and every comment. I would always stick with the best value for money product in the market unless I really need the more power or features for any electronic item. I try not to be company bias. So my aim is to get the best of my money. I hope the best for every company and every consumer

Have you noticed that the RTX 2070 uses a die codenamed TU106? How low will nVidia's high-end go?

Nice timing mate, I was about to go to bed. Oh well, I guess it's *that* time again.

Amd can still turn things around, if Nvidia keeps its prices high I think a good midrange chip that is much more affordable can give them some mindshare back. Ultimately I really don't know if Nvidia's current implementation of hybrid rendering for ray tracing is going anywhere it is still in the very early stages. It does however take a Zen like shift to get back into the game.

A different, probably more insightful perspective than 'The GPU War is Over' from a few years back. Great video and I think an excellent historical reference piece. PS: With a bit of context, RTX will be remembered as the low point for enthusiasts gpus and consumers should exercise restraint i.e. don't buy RTX cards if you can help it.

At first I was sceptical about Koduri leavivng and his succession but this Wang guy, he seems to understand what needs to happen for RTG. I trust him.

One question: how would the GTX 780 slot into the graphs at the end of the video? The one with the x80-type card summary? I noticed it was absent. I seem to recall the Maxwell launch being reasonable favourable, whereas I couldn't see the point in the 600 series when it came out, IIRC at the time I had two GTX 460 SLI which was still competitive with a 680, so I skipped sideways and snapped up used 560 Tis which gave a nice boost, then later two used 580s (I sold those on later for a profit to a company that wanted them for low cost CUDA, which made the upgrade to a single 980 very affordable). What's also interesting about all this is what happened after Fermi with the "streaming processor" (SPs) design NVIDIA was using (now called CUDA cores by most people). A few years ago I talked to Chris Angelini at toms about this, trying to work out why the 580 was so ridiculously strong for CUDA (it beats the entire 600 series and the only 700s that beat it are the 780 Ti and Titan) when its SP count was so low compared to the 680: www.tomshardware.co.uk/geforce-gtx-780-ti-review-benchmarks,review-32820-14.html Eventually the info that came back was that NVIDIA completely changed the SP design after Fermi in order to be far more power efficient and raise clocks, but in order to do so they needed a lot more SPs to achieve the same performance in gaming, but this didn't make up the gap in compute (while of course FP64 was crippled anyway). Thus, the 680 is (a bit) faster than the 580 in gaming but is much slower for compute, sometimes abysmally so. Having watched this video though, now it's clear why, it's because the 680 isn't really a 680 at all, it's what should have been the 660, as that way a "real" 680 would have had compute performance that actually made sense. This of course meant that the 780 wasn't that much of a step up either (for compute) so even that was still slower than the 580 in most cases. Indeed, I know someone with a TItan who saw slower results than I had with two 580s (he was very surprised); it's the 780 Ti or Titan Black which almost perfectly match two 580s for CUDA. Before I finally understood why, I just couldn't figure how my system with four 580s (total 2048 SPs) could outperform a friend's system with two TItans (total 5376 SPs): www.sgidepot.co.uk/misc/3930K_quad580_13.jpg It sure made a nonsense of some of the PR which boasted about SP count. Anyway, just a little aside, but interesting, shows how a major design shift can turn some aspects of performance upside down.

Came back after big Navi presentation, it ain’t over yet

And man... It's sad to look back seeing 59% being an underwhelming generational improvement. Especially with today's prices.

Thanks Jim, this one probably took a lot of time to make... As long as amd cards are good enough to compete with mid range in price performance, I am fine with it. Sadly that wont bring them any R&D money, although most of current r&d money for gpu is from their semi custom business. Maybe that next gen stuff might get closer to high end, since navi looks mid range to me, but nvidia will bulldoze all their efforts in time they make it. In any way I will stick to AMD cards simply because it looks that they at least put some effort into Linux drivers which means a lot to me.

What war? It ain't over until it's over and it ain't over.

Strange, why did the last consoles (PS5, PS4, Xbox One, Xbox Series X/S) only used AMD architecture as graphic cards? We're they cheaper than Nvidia's architecture?

i hope the Navi will be the new Radeon HD 5870, small chip, cheap less power hungry and much faster than Vega 64

Another fantastic video... Every single time... Thanks

Great video Jim! It felt like 3minutes:p great job putting this all toghether in such an engaging way mate! Regards ^^

Uni in the morning and course work due. This is a fantastic video to listen to as I work.

You ever wonder why Turing seems so half-hazardly rushed? They're terrified of David Wang, he's the Keller of AMD gpus if Keller literally invented CPUs. But of course, Nvidia has a strong lead, and I doubt we'll see AMD really catch up, or take off, until we move on passed Navi. But it's pretty clear that Nvidia is doing anything to try and stay a step ahead, and not to get blindsided like Intel did.We'll see.. Good watch as usual, thanks.

The war is perpetual and it's a series of battles, though right now AMD are fighting with muskets against machine guns. With Nvidia raising prices of GPU's to insane levels, we're seeing a mid range GPU (2070) cost what some years past flagship GPU's cost, is it any wonder that they launched with fake MSRP's this time around, a strategy that I believe will affect Nvidia's loyalty mindshare in the near future, especially in the mid range which has been held due to flagship dominance. People pay more for owning a little bit of the best, even if what they get is inferior in price performance or doesn't even win its tier (ref 1060 vs RX580). This mindset stays with historical performance not just current, so even if AMD brings out a far superior range of cards, it'll take a number of years to wipe that memory from the market. Here's where I think it'll get interesting in 2019, Navi is said to deliver 1080 levels of performance in that mid tier space and rumours have been that Navi is/was to be the RX580 successor. That's the $200-$300 space but with the RTX 2070 @ MSRP $499 but in reality $549-$599 up from it's $300-$400 price bracket, that gives Navi a huge space to play in and rather than replace the 580, it can now slot in a price bracket higher and enter the market @$399 and destroy the RTX 2070 in price performance while giving a massive boost to margins. Ray tracing isn't going to be a reason to pick RTX over traditional rendering especially at the mid level and DLSS has already been shown to be easily duplicated by simply upscaling from a lower resolution, so no technological wins for Nvidia with its new architecture. With rumours of a 580 refresh on 12nm out there, it's quite feasible that rather than retire the 580 in the $200-$300 space, if it's up against a GTX2060, it only needs to find another 15-20% to compete, this can be achieved with clock speed increases and memory bandwidth. People are voicing that Nvidia are ripping off consumers and many are ready to jump ship if AMD can deliver with Navi, that mid tier mindshare linked to flagship dominance has all but disappeared at the moment amongst enthusiasts and many normies who buy usually know an enthusiast that recommends.

This made me think of my GPU history lol, try my best to remember right as I've built 100's of PC's First PC when i was 12 y/o back in 1999 - P3 733MHz (overclocked to ~950GHz? lol) upgraded to a 1GHz P3 overclocked to 1.2GHz, 64MB ram upgraded to 768 over time, 15GB Hard Drive upgraded to 80GB. ATI Radeon 7200 LE 32MB Low Profile PCI, mehh overclock as it was in a slim micro case. Went as far as ducting fresh air with construction paper lol and worked very well. GeForce FX 5200 Low Profile PCI, good overclock in the same case after i cut a hole for direct fresh air as the fan stuck out further and was too close to the housing. But quickly found the PCI's bus limit when overclocking which was sad because the FX 5200 overclocked well. First PC i built, AMD Athlon 64 3000+ (quickly upgraded to Opteron 180 and overclocked to 2.8GHz) with 1GB DDR ram (upgraded to 2GB). Used IGP till i could afford a AGP card. GeForce 6600 GT (which i still have laying around) Quickly upgraded to 6800 XT and stuck with that for a little while GeForce 8600 GT New PC build, Core 2 duo E4300 overclocked to 3.2GHz, 2GB RAM etc. Got a second 8600 GT, hey! worked in some games.. kinda HD 4850, volt modding it with a large aftermarket cooler and pushed it to 1450MHz without making the fans get loud, had that for a while Computer was stolen out of my house (with upgraded Q9550@3.625GHz, 8GB 1100MHz CAS5 DDR2, SanDisk Extreme 120GB SSD etc) Used my laptop with a GeForce 210 that could play CS:CZ perfectly and CS:S at 60FPS dipping down to 45FPS here and there for half a year Built a temp PC with a FX-4100, 8GB RAM, forgot the rest and a HD 5450 to get things working (it was like $25). Stopped playing games for a while, i was semi-pro in CS before lol Built a proper PC (kinda, couldn't resist the bargain but wanted a 3770k) FX-8320 overclocked to 4.5GHz with an EVO 212, 8GB 1866 CAS7 upgraded to 16GB 2400 cas11 (runs 2166 cas9), 750 EVO 120GB upgraded to 850 EVO 250GB etc etc Used a HD 7850 1GB (mistake getting the 1GB) Got a second HD 7850 (used for cheap)for shits n gigs, lol 1GB VRAM HD 7950 3GB (bought it used, used for mining), After a year or two put Liquid Ultra on everything!!! clocked the GPU to 1650MHz stable in Furmark for 30+ minutes.. so i went a took me a dump :-) and.. pop (poop) went the card. Years later i took it apart and found the Liquid Ultra squirted out onto on of the resistors causing a short. R9 280x 3GB, but was butt hurt because then i got a 40" 4K monitor (Korean off brand one) R9 390 Strix (two+ years ago to now) lightly used for $160, the card ran insanely warm from the factory. Almost necessary to undervolt them. Fucking FX-8320@4.5GHz is still extremely stable and good enough for all modern applications and games, would have got a i7-3770k if i didn't find it open box with the motherboard for $180 all together.. fyucc lol. Delidded 3770k FTW.. still. Eyeballing a Vega 56, but will wait for Zen 2 in 2019 to upgrade that first. Then Vega 7nm or Navi 7nm idk what going on with that. Just want a cool running GPU is all.

Watching this video on a 7970 vapor X ^^ It plays black ops 4 very well. If youre surprised, count me in too - never thought it would be a smooth experience in 2018 but it is.

HD 3870 was my very first GPU I bought. Right now I am running on RX 580. Both purchases were worth it :).

Really disappointed that Navi was not announced. But that chip shown at CES sure does have room for one. Imagine a processor with I9 performance and able to do 1440\60 or better, all on the same chip? I am just confused as to how they got there with an 8 core and predicted R7 performance without two chips (12 core)? And with all this performance, AMD may have felt no need to announce the 16c just yet. Are they just going to kill cores and bin by frequency? I am confused.