Spitballing Nvidia’s purported GeForce RTX 2080 and RTX 2080 Ti

This week at SIGGRAPH, Nvidia introduced its Turing microarchitecture, the next major advance for the green team since Pascal made its debut over two years ago. If you're not already familiar, Turing includes RT cores that accelerate certain operations related to ray-tracing and tensor cores for accelerating certain AI operations to render the results of those ray-tracing calculations usable for real-time rendering, among other benefits that we're likely unaware of.

A Turing GPU.

Based on information that Nvidia revealed at SIGGRAPH, some back-of-the-napkin calculations, and a waterfall of leaks today, I wanted to see how the rumored GeForce RTX 2080 and GeForce RTX 2080 Ti will stack up against today's Pascal products—at least hypothetically.

Before we begin, I want to be clear that this article is mostly speculative and something I'm doing for fun. That speculation is based on my prior knowledge of Nvidia's organization of its graphics processors and the associated resource counts at each level of the chip hierarchy. It's entirely possible that my estimates and guesstimates are wildly off. Until Nvidia reveals an architectural white paper or briefs the press on Turing, we will not know just how correct any of these estimates are, if they are correct at all. I've marked figures I'm unsure of or produced using educated guesses with question marks.

The Volta SM. Source: Nvidia

My biggest leap of faith about Turing is that its basic streaming multiprocessor (or SM) design is not fundamentally that different from those in the Volta V100 GPU of June 2017. Nvidia will almost certainly drop the FP64 capabilities of Volta from Turing to save on die area, power, and cost, since those compute-focused ALUs have practically no relevance to real-time rendering. The company needs to make room for those RT cores, among other, better things it might be doing with the die area.

Past that, though, Nvidia has already said that Turing will maintain the independent parallel floating-point and integer execution paths of Volta. Furthermore, the number of tensor cores on the most powerful Turing card revealed so far, combined with some simple GPU math, suggests the Turing SM will maintain the same number of tensor cores as that of Volta. Those signs suggest we can fairly safely speculate about Turing using the organization of the Volta SM. That leap of faith is necessary here because Nvidia hasn't revealed the texturing power of Turing yet. Volta uses four texturing units per SM, so that's the fundamental assumption I'll work with for Turing, as well.

I also believe, without confirmation, that Nvidia will be releasing two Turing GPUs. One, which I'll call “bigger Turing,” should power the Quadro RTX 6000 and Quadro RTX 8000, as well as the purported RTX 2080 Ti. That 754-mm² chip has a 384-bit memory bus and as many as 4608 CUDA cores, and I'm guessing it's organized into 72 SMs and six Graphics Processing Clusters (or GPCs).

The “smaller Turing” apparently has a 256-bit memory bus, and it likely powers the Quadro RTX 5000 and the purported RTX 2080. That card likely has 48 SMs, organized into four GPCs. Judging by today's leaks, Nvidia seems to be using slightly cut-down chips in GeForce RTX products (likely as a result of yields). Fully-active Turing chips seem to be reserved for Quadro RTX cards.

  Boost

clock

(MHz)

ROP pixels/

clock

INT8/FP16

textures/clock

Shader

processors

Memory

path (bits)

Memory

bandwidth

Memory

size

Radeon RX Vega 56 1471 64 224/112 3584 2048 410 GB/s 8 GB
GTX 1070 1683 64 108/108 1920 256 259 GB/s 8 GB
GTX 1080 1733 64 160/160 2560 256 320 GB/s 8 GB
Radeon RX Vega 64 1546 64 256/128 4096 2048 484 GB/s 8 GB
RTX 2080? ~1800? 64? 184/184? 2944? 256 448 GB/s 8 GB?
GTX 1080 Ti 1582 88 224/224? 3584 352 484 GB/s 11 GB
RTX 2080 Ti? ~1740? 88? 272/272? 4352? 352? 616 GB/s? 11 GB?
Titan Xp 1582 96 240/240 3840 384 547 GB/s 12 GB
Quadro RTX 8000 ~1740? 96? 288/288? 4608 384 672 GB/s 24 GB
Titan V 1455 96 320/320 5120 3072 653 GB/s 12 GB

This first chart primarily shows how the move from 8 GT/s GDDR5, 10 GT/s GDDR5X, and 11 GT/s GDDR5X to 14 GT/s GDDR6 will affect our contenders, as well as their basic (estimated) resource counts. We know that Nvidia claims a 16 TFLOPS FP32 math rate for the Quadro RTX 6000's GPU, so that means a roughly 1740-MHz boost clock range. The potential RTX 2080's clock speed, on the other hand, is a total guess from the gut.

  Peak

pixel

fill

rate

(Gpixels/s)

Peak

bilinear

filtering

INT8/FP16

(Gtexels/s)

Peak

rasterization

rate

(Gtris/s)

Peak

FP32

shader

arithmetic

rate

(TFLOPS)

RX Vega 56 94 330/165 5.9 10.5
GTX 1070 108 202/202 5.0 7.0
GTX 1080 111 277/277 6.9 8.9
RX Vega 64 99 396/198 6.2 12.7
RTX 2080? 115? 331/331? 7.2? 10.6?
GTX 1080 Ti 139 354/354 9.5 11.3
RTX 2080 Ti? 153? 473/473? 10.4? 15.1?
Titan Xp 152 380/380 9.5 12.1
Quadro RTX 6000 167? 501/501? 10.4? 16.0?
Titan V 140 466/466 8.7 16.0

This second set of theoretical measurements shows that unlike the transition from the GTX 980 Ti to the GTX 1080, the RTX 2080 is unlikely to eclipse the GTX 1080 Ti in measures of traditional rasterization performance (likely how most users will first experience the card's power, as software adapts to the hybrid-rendering future that Turing promises). The 2080 could certainly come close to the 1080 Ti in texturing power and shader-math rates, but its pixel fill rate and peak rasterization rates aren't much changed from its Pascal predecessor (at least, if my guesses are right).

My guesstimates about the RTX 2080 Ti, on the other hand, suggest a real leap in performance for a Ti-class “bigger” GeForce. The texturing power of the purported 2080 Ti is quite a bit higher than even that of the Titan V's by my estimate, and its triangle throughput, peak pixel fill rate, and peak FLOPS are basically chart-topping for consumer Nvidia graphics processors. That should lead to some truly impressive performance figures, even before we consider the possibilities opened up by the card's ray-tracing acceleration hardware.

Nvidia will be holding a Gamescom event this Monday, August 20, where we expect to learn all about these purported Turing GeForces. We won't be on the ground in Cologne, Germany for the event, but we will be monitoring the live stream and will bring you all the details we can as we learn them. Stay tuned.

Comments closed
    • jihadjoe
    • 1 year ago

    The positioning of the 2080Ti in the tables bother me. Since the back of the napkin maths position the 2080Ti above the Titan Xp in guesstimated performance, why not list it above (below?) the Titan Xp?

    • karma77police
    • 1 year ago

    That is all nice but Nvidia/Partners wants $900/$1000 for 2080 ti model, no thanks.

      • K-L-Waster
      • 1 year ago

      The darn thing isn’t even announced yet and you already know it’s over-priced? *And* what that price is?

      EDIT: ‘k, my bad, didn’t see the PNY article…

        • chuckula
        • 1 year ago

        Seriously, let’s give Nvidia a chance to overprice the cards for [b<]real[/b<] first!

    • Chrispy_
    • 1 year ago

    Let’s just say I don’t want to invest in raytracing this generation, because I know that game developers aren’t going to bother when the 99.9% of the 2018 poential customers and 99.5% of the 2019 potential customers don’t have an RTX;

    Can I get a GTX without the raytracing shenanigans?

    All the best games we’re expecting in 2019 are not ray-tracing compatible. Maybe in 2020 there will be a game or two that is good enough to warrant buying dedicated raytracing hardware. This will, of course, be a PC exclusive because Microsoft, Sony, and Nintendo aren’t making new consoles with raytracing hardware anytime soon.

      • Hattig
      • 1 year ago

      Raytracing is only in these cards in order to get hardware out there today so game devs can start working on the technology for their 2020+ games.

      There will likely be some short-term demos, benchmarks and indie games, and a racing game is inevitably going to use it, and maybe some limited use will be made as optional settings in other games, but in the main it won’t be used in this generation.

      • ColeLT1
      • 1 year ago

      [quote<]Can I get a GTX without the raytracing shenanigans?[/quote<] I've got a GTX 1080ti on the cheap for ya, $500, sounds like what you are looking for.

    • liquid_mage
    • 1 year ago

    I know these are educated guesses, but I really struggle to believe the 2080 will not be faster than the 1080TI. The last few bumps the XX70 or X70 would be very close to the X80TI from the previous gen cards. But if your speculation is true it may explain whey they waited to 6+ months past their normal cycle to release the cards maybe tuning performance(well that and the money from the miners). I will be very disappointed if I waited 8 months for a 2070 that is basically the same as a 1080 at a similar price point.

      • thecoldanddarkone
      • 1 year ago

      It’s not really a die shrink. It has sizable amount of space dedicated to the tesla cores. I won’t even be slightly surprised if the above is the case.

    • Bauxite
    • 1 year ago

    Could’ve sworn Titan V has 5120 shaders, from the volta blog you linked:

    Tesla Product Tesla K40 Tesla M40 Tesla P100 Tesla V100
    FP32 Cores / GPU 2880 3072 3584 5120

      • thecoldanddarkone
      • 1 year ago

      It does. Someone filled out the chart wrong.

      • Jeff Kampman
      • 1 year ago

      Right number on the back of my napkin, wrong number in the chart. Fixed.

    • ronch
    • 1 year ago

    Hey this thing looks faster than a GeForce2 GTS.

      • DoomGuy64
      • 1 year ago

      Strange, the Geforce4 MX model performs shockingly similar. I think you might actually need the Geforce 4 RTX model to get the newer architecture, can’t explain why the new MX (GTX) model doesn’t use it thought. Might have something to do with the marketing dept….

    • yeeeeman
    • 1 year ago

    Don’t you already have both graphics cards ready for review?

      • techguy
      • 1 year ago

      If TR does, they can’t say. NDAs don’t allow it, and you don’t get products prior to release without signing an NDA.

        • Firestarter
        • 1 year ago

        buffalo

          • Krogoth
          • 1 year ago

          Spicy buffalo wings. 😀

        • MrDweezil
        • 1 year ago

        I would be really surprised if TR would publish a “speculative/educated guess specs” piece if they had actual hardware in-house. Seems way too close to an NDA violation even if the author isn’t a person with access to the cards.

      • K-L-Waster
      • 1 year ago

      Impending announce /= impending launch.

    • ronch
    • 1 year ago

    No need to spitball these things. We all know what they’ll do to Radeon.

    Good grief, AMD. Try to catch up. Pour that R&D money into Radeon!!

      • Krogoth
      • 1 year ago

      AMD RTG isn’t going to attempt to do that at this stage of the game. If Tahiti (GCN 1.0) couldn’t wrestle away marketshare from Kepler. There’s no chance that high-end Navi would pull it off (assuming it turns out to be a Pascal/Turing killer which in itself is extremely unlikely)

      AMD RTG is going back to ATI’s roots as being a major provider of iGPUs and semi-intergrated solutions.

        • techguy
        • 1 year ago

        “extremely unlikely” is a bit of an understatement, if you ask me. RTG won’t be competing with NV at the high end for several more years. Navi has basically zero chance of moving the bar for top-end performance. They might be able to raise the bar for mid-range performance and offer more bang for the buck, but that’s been the story of the company. I want more from AMD.

    • HERETIC
    • 1 year ago

    Some pretty pictures-
    [url<]https://www.fudzilla.com/news/graphics/46985-msi-and-palit-geforce-rtx-2080-2080-ti-smile-for-the-camera[/url<]

    • Krogoth
    • 1 year ago

    It sounds about right for the most part, but I suspect the max boost speeds on some SKUs will be closer to 2Ghz range.

    • mad_one
    • 1 year ago

    It does look as if they are going with a slightly modified Volta SM, while they used different SM organisations for GP100 and the gaming chips.

    So 4×16 shaders instead of 4×32 and unless they halved the register file from GV100, twice as many registers per shader. Add to that the seperate int and fp execution units (I kind of assume you can still issue at most 2 instructions per cycle, with the int units being added to the selection of fp32, load/store and SFU). I think they can also do double rate FP16 now. It may also have the optimisations Volta added for diverging branches.

    All of this makes the SM a lot larger per shader compared to Pascal, but the gains will depend a lot on the code. I don’t expect game shaders will make quite as much use of int instructions for addressing as HPC code. We know from Vega that fast FP16 won’t change the world for now. Games probably avoid diverging branches pretty well. Twice as much register space would mean a lot more batches in flight and thus better latency hiding.

    So it seems perf for the 2080 will minimum only be ~20% better, but might be more like 30-40% typical. Memory bandwidth should help as well, the more common 10Gbps versions of the 1080 seemed to suffer a bit and I hope the dual channel architecture will make GDDR6 more efficient than GDDR5X (the 10Gbps on the 1080 seems to buy very little over the 8Gbps GDDR5 on the 1070 Ti).

      • Leader952
      • 1 year ago

      Isn’t the cache size also larger ?

        • mad_one
        • 1 year ago

        We already know the L2 has doubled and I think there are significant changes to the Volta L1 cache and shared memory arrangement, but I don’t remember those offhand.

          • Leader952
          • 1 year ago

          Wouldn’t the increase in L2 cache (and L1/shared memory) increase the performance more in your above analysis?

            • mad_one
            • 1 year ago

            Honestly, I am not qualified to do a precise analysis. I’m trying to list the known changes, any predictions as to their effect should be taken with an unhealthy amount of salt 😀 .

            Cache doesn’t work all that well on graphics workloads, which is why those gigantic GPUs only feature very small (but extremely high bandwidth) caches. In fact AMD hasn’t had a unified L2 before Vega. The ROPs had their own small cache for write combining, but transfers of data from the ROPs to the shaders would go through memory.

    • torquer
    • 1 year ago

    I vote we call the 2080 Ti “Grand Turing”

      • Neutronbeam
      • 1 year ago

      And the benchmarking will be known as the “Turing Test”.

        • K-L-Waster
        • 1 year ago

        I dunno, the results would be something of an Enigma…

      • Mr Bill
      • 1 year ago

      Good for [url=https://getyarn.io/yarn-clip/325086bb-bf7e-458a-9eb7-5e88863bf092#Hk_w7sgr87.copy<] relieving my... tensors[/url<]

    • techguy
    • 1 year ago

    I think your boost clocks are way too low. RTX 2080 may not even beat 1080 Ti with those specs, let alone Titan V as some rumors indicate. Functional unit counts seem likely though, given everything that’s come out over the past few days.

      • techguy
      • 1 year ago

      Well, looks like I was wrong.

      [url<]http://www.pny.com/RTX-2080-Ti-Overclocked-XLR8-Edition?sku=VCG2080T11TFMPB-O[/url<] 1350MHz Base/1545MHz boost. Should be about 8% faster than 1080 Ti. All for a mere $1000! I'm sitting out this gen.

        • Krogoth
        • 1 year ago

        Somebody at PNY is going to get canned for violating NDA.

        Based on those leaked specs, 2080Ti should be least 15-30% faster than 1080Ti with the same power consumption at current generation stuff. If you are adventurous enough it should scale better with overclocking then 1080Ti due to GDDR6 having more bandwidth.

        $999 MSRP sounds about right. People were still buying 1080ti when they were at around $1,000 during the crypto-currency craze. Nvidia marketing is confident enough that 2080Ti will still move at that price point.

        • Leader952
        • 1 year ago

        $1000 is probably a place holder.

          • Krogoth
          • 1 year ago

          Nah, it is likely going to be around the intended price. Nvidia is going to shoot for $999 MSRP with 2080Ti and will mostly certainly move at that price point.

    • Freon
    • 1 year ago

    Those seem like reasonable estimates. The timing also makes me believe the Geforce RTX will share dies with the Quadro cards, with (roughly) all the RT and tensor capabilities. Stepping back a bit, it surprises me that the tensor and ray tracing capabilities are not being cut down, leaving us with rather significant die space used for things that may have little impact on gaming in the near future.

    Raw specs should be sufficient to beat previous generation parts, but I worry will (edit: NOT) offer significant increase in value over the previous generation Pascal parts. We should reasonably guess the 2080 will beat the 1080 Ti with its increase in clock and FLOPS, assuming memory bandwidth is not a bottleneck, but not by a large margin. NV seems to surprise, though, and Pascal was a fair bit stronger than I expected.

    It seems NV may be taking the absence of a credible threat from AMD RTG to push tensor and RT features. Getting an installed base started now may make strategic sense for them–to create a market for features AMD and Intel may never be able to catch at this point. Vega dropped like water logged bread loaf.

    As a software engineer I’m excited to see these features in consumer products, but as a gamer I feel this generation will not offer a ton of value.

      • dpaus
      • 1 year ago

      Agreed. I can see us using these cards in some of our real-time ‘deep analysis’ routines, where pattern-matching offers us far better insights than linear logic. But for gaming? Especially given the expected price point? Nope.

        • JustAnEngineer
        • 1 year ago

        Maybe NVidia will drop the ugly G-Sync tax and allow this new generation of graphics processors to work with attractively-priced VESA adaptive sync displays.

          • Krogoth
          • 1 year ago

          Not going to happen.

          Gsync 1.0 will continue to endure as long as Nvidia has the high-end and mid-range discrete GPU market firmly in their hands.

          • derFunkenstein
          • 1 year ago

          LOL

          Whenever a company that has the vast majority of its market already locked up has the opportunity to do something customer-friendly, expect the opposite.

          • Freon
          • 1 year ago

          I wish they would, but it seems unlikely given the leverage they have on the market.

          • Voldenuit
          • 1 year ago

          FPGAs are expensive. I recall watching a teardown of one of the fancy 4K 120 Hz G-sync monitors, and the FPGA unit in it was estimated to be upwards of $200 BOM.

          Now nvidia may have over-engineered their approach to VRR with the G-Sync module, since AMD was able to use existing panel display components in Freesync, but we also had a lot of teething problems with low-quality early Freesync monitors (and still do, to some extent). Whereas almost all G-Sync monitors have been of decent quality.

          Is panel display tech now good enough that we don’t need an FPGA with a big physical buffer to display VRR? Most likely. Is nvidia going to abandon its sunk costs in engineering, personnel, business partners and OEMs so that the consumer can save a bit of money? Realistically, it would take AMD being competitive for that to happen.

          • K-L-Waster
          • 1 year ago

          They will only change course on G-sync if it’s costing them money.

          Given the quarterly results they just posted, I wouldn’t hold your breath.

      • caconym
      • 1 year ago

      I feel like they’re not going to leave much of the raytracing hardware intact on the GeForce cards, for market segmentation reasons. Quadro seems like it needs to offer more than just a bigger VRAM pool and ISV certification.

      Would love to be wrong though!

        • mad_one
        • 1 year ago

        So far Quadro cards rarely had a lot of features over the consumer cards (they used to have some OpenGL features like additional clipping planes, no idea if that’s still important). The driver differences and guaranteed support seem to be enough and so. Also the Quadro cards have much more memory.

        I’d be more worried about the high AI performance if I were Nvidia.

        • Krogoth
        • 1 year ago

        They are going to likely gimp ray-tracing/tensor performance by disabling half of the units. Enough to handle “ray-tracing” acceleration at gaming but not enough to risk cannibalizing their Quadro line.

        • Freon
        • 1 year ago

        Wild speculation, perhaps memory footprint may be enough to keep RT from having the professional rendering prowess and there won’t be so much RT gimping in the driver.

Pin It on Pinterest

Share This