Nvidia tops off its Turing range with the Titan RTX

Nvidia's Turing graphics cards may cost a pretty penny, but few people are going to argue with the performance of their massive arrays of tensor, RT, and CUDA cores. The green team thinks you may want even more, and it's just let loose its big gun: the Nvidia Titan RTX.

This massive card is fitted with a fully-enabled TU102 GPU coupled with a whopping 24 GB of GDDR6 RAM. The chip itself is earmarked with a nominal 1770 MHz boost clock, while the memory comes set at 14 GT/s for a total 662 GB/s of bandwidth. The fully-armed TU102 GPU and high core clock make the Titan RTX the fastest single gaming card around for those who can stomach its price, but gamers aren't really this card's raison d'être.

AI researchers have emerged as the real domain of Titan cards in recent years, and along with that 24 GB of RAM for massive data sets, those folks should find plenty to like in this card. The Titan RTX packs a complement of 576 tensor cores, up from 544 on the RTX 2080 Ti. According to Anandtech, the Titan RTX's ace in the hole for deep-learning folks is its support for full-rate tensor-crunching for operations with FP32 accumulate. GeForce cards can apparently only perform those operations at half of their potential maximum throughput. All told, that array of silicon goodies should let the Titan RTX make short work of most any task one can throw at it.

Nvidia's Titan cards are neither affordable nor meant for the average enthusiast, and the $2,499 price tag reflects that fact. Although that may look like a scary dollar figure, it's worth keeping in mind that people in the market for the kind of horsepower the Titan RTX can deliver will likely be thinking about their return on investment and not the upfront cost. For what it's worth, the Titan V still sells for $2,999 at Nvidia's site, so the Titan RTX nevertheless represents a haircut for Nvidia's work-and-play flagship.

Comments closed
    • Klimax
    • 10 months ago

    Just get finally support for Turing into Iray and other similar renderers. Then there will be sudden lack of available cards…

    • techguy
    • 11 months ago

    More evil deeds from my precious NGreedia! [url<]https://blogs.nvidia.com/blog/2018/12/03/physx-high-fidelity-open-source/[/url<] All part of my nefarious plot to take over the world, and destroy the usurper Lisa Su!

      • jihadjoe
      • 10 months ago

      Uncle Jensen just did that so he could poke fun at his little niece Lisa at the upcoming family dinner.

    • USAFTW
    • 11 months ago

    Nvidia Employee #1: Guys, our Turing GPUs aren’t flying off the shelves as fast as Pascal cards, what do we do?
    Nvidia Employee #2: Can we lower prices a touch to incentivize customers to buy these cards?
    Nvidia Employee #3: Now hear me out. What if take the $1200 2080 Ti, enable the extra SM that we’ve lopped off on it, call it a Titan, price it at more than double?
    Nvidia Employee #1 to #3: You’re a gosh darn genius!

      • techguy
      • 11 months ago

      Yeah, let’s just pretend like 24GB of GDDR6 doesn’t have a cost…

        • USAFTW
        • 11 months ago

        I don’t think the extra 13 GiB of GDDR6 would cost $1300.

          • techguy
          • 11 months ago

          Likely not, but since there’s nothing else on the market that competes, NV can price it however they want.

            • sweatshopking
            • 11 months ago

            That’s not really a great argument against price gouging.

            • techguy
            • 11 months ago

            It’s also kind of ridiculous to label it “price gouging” when the entire purpose of this product is to serve a niche market wherein professional users can save money compared to buying full-blown professional SKUs. And halo users can buy a halo product. Some people save money, some people don’t. Depends on the use case.

            • K-L-Waster
            • 10 months ago

            This is yet another tragic aspect of AMD’s nearly decade-long near death experience: we have a generation of techies who are convinced that selling hardware above cost == gouging.

            • Krogoth
            • 10 months ago

            It is capitalism 101 stuff.

            • Anonymous Coward
            • 10 months ago

            Times are ‘a changin. I doubt its got much to do with the competitive circumstances of AMD vs Intel however. It may have more to do with growing up in the modern world, in a time when consumerism appears that it might not deliver the goods quite as expected.

      • Krogoth
      • 11 months ago

      Actually, it is more like Nvidia is trying to sell off a small number of near-perfect/perfect yields that would normally be sold as Quadros but they either ate too much power for their vendor’s liking or just excessive stock.

      That’s how Titan brand was born. AMD RTG’s eventual answer would be Frontier brand.

    • techguy
    • 11 months ago

    Argh, it has 24GB VRAM. I figured NV would only release a 12GB Titan again this generation. Guess they’ve gone back to actually making Titan cards that are meaningfully superior to their x80 Ti brethren. Now I have to decide whether to keep my 2080 Ti or not…

      • BurntMyBacon
      • 11 months ago

      #FIRSTWORLDPROBLEMS

        • techguy
        • 11 months ago

        Yep. Not a bad problem to have, in the grand scheme of things. Notice that I’m not railing against Nvidia or throwing a temper tantrum, only lamenting the choice that now lays before me. Ultimately it is my choice to make.

          • chuckula
          • 11 months ago

          Your inability to heap scorn and hatred on any product that isn’t from AMD just shows you to be a biased shill!

            • techguy
            • 11 months ago

            Not only am I an Nvidia shill, I’m actually Jen-Hsun Huang! (not Jensen, you silly Americans)

      • NovusBogus
      • 11 months ago

      Turing in general seems to be an example of letting their engineering team run wild and free.

    • Mr Bill
    • 11 months ago

    As an example of industrial art, its very nice.

    • Chrispy_
    • 11 months ago

    I’m not sure that RTX is really the right name for this Titan. Maybe Titan T would have been better, since the 11 Gigarays/second performance of this card still isn’t going to get you anything better than 1080p60 for your $2499 outlay in games.

    If it’s all about the full-rate tensor cores, The 2080Ti-level of raytracing performance is a sideshow at best. Why name the damn thing (that’s clearly not for gaming) after the sideshow feature?

    Oh, wait. Milking rich kids who are stupid but like faux-gold covered tat?
    Sorry, reality caught up with me again.

      • stefem
      • 10 months ago

      People still mention 1080p at 60fps while even a $700 RTX 2080 can run BF V at 1440p at above 60fps…

    • PrincipalSkinner
    • 11 months ago

    Do any gamers here take Nvidia seriously anymore?

    edit: deanthropomorphises the games

      • chuckula
      • 11 months ago

      Games hate it when you anthropomorphise them.

      • K-L-Waster
      • 10 months ago

      Which part of “not aimed at gamers” got lost in translation?

      • Chrispy_
      • 10 months ago

      Why wouldn’t any gamer take Nvidia seriously?

      They are the driving force behind about half of the DirectX and OpenGL features. They pushed VRR into the market (AMD wouldn’t have made Freesync if it weren’t for Nvidia) and they have ALWAYS had a mainstream product capable of AAA gaming at around $200 and its the fierce competition at the $200 point that pushed GPU performance forward all these years.

      Nvidia don’t put a gun to your head and make you buy $1000+ graphics cards and $2000 monitors. You can definitely still have a great experience at 1080p60 on a mid-range Nvidia card and in this era that is witnessing a shift from desktop to mobile, Nvidia are winning on the performance/Watt charts by an almost laughable margin.

    • leor
    • 11 months ago

    Great, full product line, so let’s recap:

    [Edit] Price at release:

    480: $500
    580: $500
    680: $500
    780: $650
    780 Ti: $700
    980: $550
    980 Ti: $650
    ——————————————-
    1070 – $379
    1080 – $549
    1080 Ti – $699
    Titan XP – $1,200

    2070 – $499
    2080 – $799
    2080 Ti – $999 (really $1,199)
    Titan RTX $2,499

    I remember when the y first started charging 500 dollars. Scott in the review said “if you don’t mind spending half a thousand dollars on the fastest GPU out there,” to drive home just how expensive these things were. The sad thing is we’re not getting more value, the jumps from generation to generation are pretty much what they’ve always been.

      • jihadjoe
      • 10 months ago

      Wrong on Pascal prices

      1070 launched at $449
      1080 was $699 (FE) with MSRP initially at $599

      The prices you put in only happened after the 1080Ti launch.

      • thedosbox
      • 10 months ago

      Need to add the Titan V @ $2999 – it’s even listed in the story.

        • leor
        • 10 months ago

        Titan V was never the same class of product and sat right next to the Titan XP this whole time.

    • NovusBogus
    • 11 months ago

    This is what the RTX lineup should have led with.

    • Krogoth
    • 11 months ago

    This card is marketed towards game developers for future DX12 titles as the baseline for “High/Ultra” settings or professionals/prosumers that need something more than regular RTX SKUs but a Quadro/Tesla is outside the budget.

      • End User
      • 11 months ago

      If I listened to you I’d be playing with a ball of string.

        • Krogoth
        • 11 months ago

        It isn’t meant for gamers though. That’s what 2080Ti is for. If you are patient enough. You’ll get a faster and cooler version of “Titan RTX” with upcoming 7nm die-shrink that will be sold at a much lower MSRP (Probably Q3-Q4 2019).

    • NTMBK
    • 11 months ago

    Ridiculous price tag and faux-gold bling? Should have called it the Titan MAGA.

      • chuckula
      • 11 months ago

      MAGA: Make AMD Great Again.

        • USAFTW
        • 11 months ago

        FTFY: MAMDGA

          • Beahmont
          • 11 months ago

          Hmm… the pedant in me things Chuckula has it right. AMD isn’t an acronym any more it’s the actual name. So when putting the actual name in an acronym one would just use the first letter, A and not the complete AMD like you would if one is mixing acronyms.

    • LocalCitizen
    • 11 months ago

    10% fewer tensor cores than titan volta, but 21% higher frequency, AND double the memory for $500 less. this is tremendous value improvement over titan volta.

    double the memory not only means bigger data set, more importantly, it means a bigger model. data set can always be broken into smaller batches, but a model can’t.

    also, remember Quadro RTX 6000, same chip, same memory, is $6300

    i can also see the hollywood studios building ray tracing farms using these cards. does nvidia license still bans titans in data centers?

      • Krogoth
      • 11 months ago

      Titans do not have firmware and certificates for professional 3D graphical suites. You get Quadro SKUs for that. The licensing cost for said software suites are an order of a magnitude higher then hardware.

        • lycium
        • 11 months ago

        BTW, I’ve been working in professional 3D rendering for a long time now, and have yet to hear of a single person using these vaunted Quadro drivers. If there are 3 people left in the world using AutoCAD, I have still to meet them… everyone else is using Max, Maya, Blender, C4D, Modo, etc etc… and those all work absolutely perfectly fine with the normal drivers.

        Hell, anyone who thinks some Nvidia drivers is going to save 3ds from its abysmal performance has a terribly expensive surprise coming…

          • Chrispy_
          • 11 months ago

          We actually have hundreds of AutoCAD licenses (and Revit, and Maya, and Blender, and Rhino, and even old-fashioned 3DSMax!).

          Nobody uses a Quadro or Radeon Pro. Both AMD and Nvidia send me thousands of dollars of sample hardware to try out each year, and I plug them into the computers of willing engineer and architectural volunteers to be told time and time again that they want their Geforce or Radeon back.

          It’s not rocket science – 99% of the software on the market is designed to be sold to the widest possible audience in order to make the most money. That means it has to run on Intel IGPs, which means it will definitely run just fine on a Geforce or a Radeon, which usually has more shaders/cores enabled and costs only 20-40% of the overpriced pro-level card.

          Anyway. The marketing guff that both AMD and Nvidia sales people spout at me is that you get rapid replacement and dedicated support. I just keep a stockpile of replacement Geforce and Radeon cards on hand and replace failed cards in under five minutes. A ten-year-old kid could do it. And here’s the thing – by spending €400/workstation on consumer graphics cards rather than €2000/workstation on Quadros and Radeon Pros, we have easily saved €1m over the last half decade, and I’ve definitely not had to buy €1m of replacement cards. I budget a 10% failure rate over 3 years for the asset register and my actual spend is considerably less than half of that.

            • techguy
            • 11 months ago

            I like your approach.

            Though there are a few markets that still need the pro-level hardware due to artificial segmentation, graphics-accelerated VDI comes to mind, I wish the graphics IHVs would stop this practice and instead price cards purely according to performance. If that means they need to release Teslas with 256GB then so be it!

            • Krogoth
            • 11 months ago

            Not going to happen until freeware suites provide identical functionality as the current crop of professional software suites.

            Welcome to the world of commercial software and hardware vendors.

            • MarkHughes
            • 10 months ago

            Indeed, I’m a coder for Tekla Structures and we have also moved more toward gaming rigs than professional workstations (Also moved to directx) It’s made getting the right hardware easier and a lot cheaper and I am sure our customers appreciate it too. Especially if you are a small one or two person setup you can just buy a couple of gaming boxes and be in business.

            • Chrispy_
            • 10 months ago

            Good to know. When you say [i<]"moved more towards gaming rigs"[/i<], what API and hardware were you focusing on beforehand? Back in the days of Quadro wireframe antialiasing, all the major CAD packages took advantage of the pro-level features but even entry-level GPUs have been powerful enough to render complex models with full lighting, shading, and transparencies for several years now.

          • Krogoth
          • 11 months ago

          Quadros are used in major 3D graphics studios. They aren’t meant for amateurs, students and solo operations . The drivers for them are coded for accuracy in renderings not performance. That’s why Quadro/FirePro drivers have inferior gaming performance (~5-10% slower) then their Geforce/Radeon counterparts despite the fact underlying hardware between both platforms are identical.

          Certifications/validations and firmware are just artificial locks. They designed to prevent said 3D major studios from buying up customer-tier hardware to do the job. Nvidia/AMD RTG aren’t going to be cannibalizing their lucrative professional graphics market.

          You used be able to softmod regular Geforces/Radeon back in the day into Quadro/FireGLs and they would work with their drivers. Nvidia and AMD RTG started to locked it down with firmware around Geforce 8xxx/HD 2xxx era.

          Some of the advanced functions (like 10bit color output) for professional 3D/2D software suites are locked behind Quadro/FirePro firmware/drivers. It is part of the partnership between software vendors and Nvidia/AMD RTG. They are selling a service package to professionals.

          It is unlikely the landscape will change anytime unless there’s a freeware solution that replicate all of the functionality without the artificial locks.

            • sweatshopking
            • 11 months ago

            Been a few years, but I know the oil and gas company I worked for used quadros for their work.

            • NovusBogus
            • 11 months ago

            Our mechanical engineering groups use Quadro. No idea what the artsy-fartsies use, not my department.

            • Krogoth
            • 10 months ago

            Major 3D graphical studios use them for professional-tier work and productions.

            Pixar, Disney, Dreamworks etc

            • GrimDanfango
            • 10 months ago

            Quadros are used in major VFX studios because Dell, etc, package all their prebuilt “workstations” exclusively with low-end Quadros when they bulk-order their equipment, and VFX studios don’t care what they get, because they still render most of their work on CPU-based renderfarms.

            VFX studios don’t care about “accuracy”, they care about flexibility, combined with drumming out as much material as they can.

            Quadros might well have their place in scientific/research applications, but there’s absolutely no point to them in the VFX industry. The only reason they use them at all is because nobody can be bothered to tell vendors to stop sending them, when *any* basic 3D card will work just fine displaying a grey-shaded OpenGL preview window in the corner of your screen.

      • Waco
      • 11 months ago

      Without ECC nobody will be using these professionally, that’s the realm of Quadros. Saving money on hardware is not in the cards when the software driving them costs far in excess of the card value.

        • psuedonymous
        • 11 months ago

        ECC is valuable for simulation and GPGPU. For your render farm it’s not worth the increased cost, and path tracing acceleration is going to eat everybody’s lunch in the cinematic render market.

        • kuraegomon
        • 11 months ago

        See Chrispy_’s response above – I promise you that his usage more than qualifies as “professional”.

          • Chrispy_
          • 10 months ago

          Whoa whoa whoa, hold on there. Just because we design and construct airports, stadiums, transit hubs, and skyscrapers doesn’t mean we’re professional. I’m wearing shorts at work and browsing TR, for a start!

        • Krogoth
        • 11 months ago

        Exactly, the professional graphical world is a completely different animal then gaming world. Aside from computing power and minimization. It hasn’t really changed that much since SGI/Cray days.

      • Ninjitsu
      • 11 months ago

      Tremendous value improvement over something that was a poor value halo product? A gosh darn miracle, I say!

    • chuckula
    • 11 months ago

    Good article but it needs some editing:

    [quote<]The fully-armed [u<]and operational[/u<] TU102 GPU and high core clock make the Titan RTX the fastest single gaming card around for those who can stomach its price, but gamers aren't really this card's [i<][s<] raison d'être[/s<][/i<] [u<]target to join the dark side[/u<].[/quote<]

    • Waco
    • 11 months ago

    I really don’t like the switch from blower-style coolers to these axial fans. It used to be the reason to buy stock cards.

    Not that it matters for a Titan card over 2 grand.

      • kuraegomon
      • 11 months ago

      You may wish to consider just much heat needs to be dissipated from that there King-of-the-Monsters (and its 24 GB of RAM!) before you wish a blower-style cooler upon it.

        • Waco
        • 11 months ago

        There’s nothing inherent in the design of blowers that precludes them from cooling these cards.

          • kuraegomon
          • 11 months ago

          [i<]Can[/i<] you use a blower-style cooler for this card? Absolutely - after all, [b<][i<]Fermi[/i<][/b<] shipped with a blower. [i<]Should[/i<] you do it? Probably not - consider fan speed and noise profile, for two factors that suggest otherwise. Of course, this is all speculation - we don't know what the Titan RTX's power draw is yet. I'm just going to bet on a number somewhere between "a lot" and "holy cr@p".

            • Waco
            • 10 months ago

            Even if it’s “a lot” at 300+ watts, blowers *can* be designed that aren’t loud to handle that kind of heat load.

      • DavidC1
      • 11 months ago

      Blower cooler cards may have an advantage in extremely cramped cases, but other than that the axial fans are better.

      You don’t have to worry much about axial fans spreading hot air through the case. Just attach a case fan to expel it.

      • zqw
      • 10 months ago

      These fans also preserve segmentation vs Quadro RTX with blower fans, and their intended professional audience that packs multiple cards into a case.

        • Waco
        • 10 months ago

        Nvidia is learning from Intel. We’re doomed.

    • thedosbox
    • 11 months ago

    This thing makes a 2080 Ti look like a bargain

    /s

      • K-L-Waster
      • 11 months ago

      Kinda like how a Koenigsegg makes a McLaren 720S look like a bargain.

      (Cue complaints about automotive analogies…)

        • jihadjoe
        • 11 months ago

        They’re not even from the same manufacturer!

        How about the Senna makes a 720S look like a bargain? Same brand, same engine, just slightly more uncorked with the special edition, and 3x the price (Nvidia, take a hint!).

          • BurntMyBacon
          • 11 months ago

          [quote=”jihadjoe”<] (Nvidia, take a hint!).[/quote<] I think they already did.

Pin It on Pinterest

Share This