review nvidias rtx super and amds radeon rx 5700 series graphics cards reviewed

Nvidia’s RTX Super and AMD’s Radeon RX 5700 series graphics cards reviewed

It’s been a while since our last GPU review, and a great many graphics cards have come out since then. A lot has changed around here, too. Because of all the commotion, we haven’t been able to take time out to look at the latest video cards—until now. Thanks to generous support from EVGA, Gigabyte, AMD, and a few friends of the site, we’ve been able to procure a partial set of the latest graphics processors for your perusal. So saying, let’s pause a moment to ponder what the GPU market is like in August, 2019.

Over on the green team, Nvidia refreshed its extant GeForce RTX series graphics cards with a new line-up affixed with “Super.” I use the term “new” in the most specific sense, because the GeForce RTX Super cards aren’t really all that different from the original recipe. In fact, there are no technology changes whatsoever between the original GeForce RTX series and their “Super” descendants—each card gets a spec bump, and then out the door it goes. Those spec bumps can be pretty significant though, as you can see in the chart above.

EVGA GeForce RTX 2080 Super OC Ultra

The RTX 2080 Super is the least-revised of these chips. It is based on a fully-enabled TU104 processor, but the original RTX 2080 was pretty close to that already, so it only gains two Streaming Multiprocessors (SMs). The biggest change in the RTX 2080 Super is arguably the bump in memory clock, from 14 MT/sec to 15.5 MT/sec. That gives it the fastest RAM on any GeForce to date. Thanks to EVGA for supplying our GeForce RTX 2080 Super XC Ultra card, which has a 30-MHz improved boost clock over the base model.

GIGABYTE GeForce RTX 2070 Super Windforce OC

For its part, the RTX 2070’s “Super” variant steps up from its original TU106 form to that of a cut-down TU104 chip. This change isn’t as drastic as you might expect. Both chips use a 256-bit GDDR6 memory interface, and both top out at 64 render output units (ROPs). The RTX 2070 Super gains 4 SMs, sure, but the biggest change is that the RTX 2070 Super gained 195 MHz of base clock and 150 MHz of boost. In combination with the expanded compute resources, that makes the RTX 2070 Super a solid step up from its predecessor. Thanks to Gigabyte for providing us with our GeForce RTX 2070 Super Windforce OC card. It runs 15 MHz quicker than the reference card, but we think it should be perfectly representative.

GIGABYTE GeForce RTX 2060 Super Windforce 8G

The RTX 2060 Super also gains 4 SMs, putting it just two shy of the original RTX 2070 (34 vs. 36). Furthermore, it gains 105 MHz of base clock, but then loses 30 MHz of boost compared to the original card. That’s not going to hurt it much, considering the extra compute resources it has on hand. Its onboard memory also expands from a meager 6GB to a more-standard 8GB, and thanks to that change, its memory bus expands to 256-bit width. Altogether, the RTX 2060 Super is the most-improved of the three models, and that’s why it’s the only one that got a price bump.

AMD Radeon RX 5700 XT

AMD knocked it out of the park with its Zen 2-based third-generation Ryzen CPUs last month, but its Radeon team hasn’t been asleep this whole time. The “Navi” GPU core and its “RDNA” graphics architecture actually launched the same day as the third-gen Ryzens—over a month ago—and in that time, we’ve spent quite a bit of ours with the Radeon RX 5700 and RX 5700 XT. These cards are a fair bit different from their predecessors architecturally, so keep that in mind as you compare them on this chart:

AMD went into great detail on the architectural changes in RDNA as compared to the venerable GCN architecture that preceded it, and unfortunately, not being a graphics programmer, a lot of it went over my head. Also unfortunate is that the company claims a 1.5x improvement in performance-per-watt compared to its Vega 64 cards, but I don’t have a Vega card here to test that claim. Still, we can talk a bit about how AMD came to achieve that number.

Naturally, most of the power-efficiency improvements in Navi come down to the move to 7-nm manufacturing. That’s not the whole story, though. AMD says it took the lessons it learned from Zen and applied them to GCN to create RDNA. So saying, RDNA was designed with an emphasis on efficiency and single-threaded performance. AMD specifically mentions a focus on instructions-per-clock (IPC) several times in its briefing, which isn’t a term we usually hear in the context of GPUs.

AMD Radeon RX 5700

The short version is that RDNA works on fewer things simultaneously, but in larger chunks. This ends up giving it almost exactly the same theoretical compute throughput as a GCN GPU of similar specifications. As we all know, theory and reality rarely match up, and GCN chips often struggled to even get into the ballpark of their theoretical performance. In practice, the new architecture is 50% faster at the same power budget, according to AMD. RDNA is also fundamentally compatible with GCN, meaning developers don’t have to re-learn how to write applications for the new cards, even if they’re using low-level APIs like Vulkan.

With all of that said, you’re here to hear about how these cards actually perform, so let’s get this show on the road.

Page listing:

  1. Introduction
  2. Testing setup and methodology
  3. Doom (Vulkan)
  4. Far Cry 5 (DX11)
  5. Grand Theft Auto V (DX11)
  6. Hellblade (DX11)
  7. Mirror’s Edge Catalyst (DX11)
  8. Monster Hunter World (DX11)
  9. Warframe (DX11)
  10. Witcher 3 (DX11)
  11. The Division 2 (DX12)
  12. Devil May Cry V (DX12)
  13. Deus Ex: Mankind Divided (DX12)
  14. Hitman 2 (DX12)
  15. Another look at Nvidia’s DLSS
  16. Radeon Image Sharpening
  17. Radeon Anti-Lag
  18. Further RTX discussion; power & efficiency
  19. Conclusions

55 responses to “Nvidia’s RTX Super and AMD’s Radeon RX 5700 series graphics cards reviewed

  1. What this review tells me is that my 1080Ti held up really, really well. I got 2070 Super performance for two extra years at a $150 premium.

    Sure no ray tracing, but I’d rather buy into that tech when it becomes better developed.

  2. I have a Matrox G650. Its a fabulous office card for multi-monitor but far too slow for gaming.

  3. I should add, this is not meant to denigrate the AMD’s Navi. Just AMD’s reference cooler design.

    I think the Sapphire cards (specifically, the Pulse 5700XT or Pulse 5700) for ten bucks more than the reference design, is competitive, and worth looking into. But they may be hard to find at the moment.

  4. That is in fact what the reference shroud looks like.

    (As if peeps didn’t have enough reasons to hold out for 3rd party coolers…)

  5. Zak, excellent review. You always have interesting observations for each game and how the cards deal with the game. That in-the-second commentary is what brings the nerds to the yard.

  6. I usually set mine to 2 globally in the Nvidia control panel – enough for double-buffered vsync if I want it, and much less latency in game engines that pre-render like 5 frames by default.

  7. So sad that great content like this has to get fakked by this horrible horrible generic template.

  8. thank you SOOOO much for this review! Time Spent Beyond x ms / average frametime now is the first thing i look at with reviews

  9. Vega 56 is a bit of a wildcard because it is highly depend on how well you can undervolt the unit. Unlike the 5700 and 5700XT which can operate at their performance levels without too much hassle. Vega 64 only pulls ahead if you are brave enough to overclock/undervolt to its limits and are willing to tolerate the power consumption.

    Vega units are decisively better if you care more about general compute performance.

  10. Yep, the Radeon VII is a much better general compute/content creation card then a gaming card. There’s nothing close to it in its price point. You have to spend a lot more if you want to get performance in either market.

    It was a steal for general compute hobbyist like Kepler-based Titans back in the day.

  11. “If you want a ballpark figure just take 5700XT results and reduce them by like ~2-10% to get Vega 64 stock/tuned Vega 56 results.”

    I think you’re over-simplifying the comparison. I still think seeing the Vega GPUs on the ‘scatter charts’ would be valuable data points.

    In addition, there is the price dimension that needs to accounted for. Right now, an ASRock Phantom Gaming X Radeon RX Vega 56 can be found for US$270, while a ASRock Radeon RX 5700 goes for US$330 @ TheEgg. Saving ~20% of GPU cost for a difference of “~2-10%” in performance is worth considering. Running the benches on the Vegas would have firmed-up the comparison.

  12. My guess is AMD doesn’t really want to make any more of the VII than is necessary. It’s relatively cheap for something that uses such expensive components, and built on an expensive process.

    90% of those chips they rather go into the Mi50 accelerators.

  13. One note I don’t think I recall being brought up (the lack of index drop-down in the pages listed makes me loath to go back and check):

    Everything I’ve seen says that if Navi is your card, do not buy a reference design. Some reference blowers are quite good (nVidia Pascal was; I was very happy with the noise level on two 1070 Founders Edition cards at full load, and a single one was whisper quiet).

    AMD’s blower -is not. It is both loud, and a poor cooler; Sapphire’s own dual-fan Pulse 5700XT runs twenty degrees cooler. Twenty degrees [i]Celsius[/i] cooler. That’s an incredible temperature drop and could make all the difference in your GPU throttling. Its cooler is far better designed, and in addition to two fans, pays special attention to how the VRMs are cooled too. And it’s much quieter.

    Buying an AMD reference design this time around looks to be a bad choice; I’d look for the aftermarket models. Sapphire’s design currently beats MSI’s two-fan as well (better heatsink).

  14. Yup. Radeon VII owners are sadly being left high and dry.

    It was a lousy buy even for the most die-hard AMD fans, and its short market time is pretty disappointing for anyone who bought one.

  15. Completely agree, especially since the model used was the 4GB version. One if these things is not like the other.

  16. “For gamers like myself who use myriad monitors—I’m currently using five—”

    got a matrox card? 😉

    when are the $235 card coming out?!?!

  17. If those cards were not benchmarked on the same rigs as used for this review, the comparison would not be univariate. Multivariate is what proper reviews try to avoid.

  18. Radeon VII was a stopgap solution until Navi was ready. It was a way to clear out excessive Vega 20 stock that ate too much power for ISV customers.

    Navi already bests Vega 20 at gaming performance when memory bandwidth isn’t a factor.

  19. I was wondering the same, that said a check on newegg shows that VII is very limited. Is AMD running into production issues with the VII?

  20. It is likely that Zak simply doesn’t have any Vega hardware on hand and his test system is different from previous Vega benches making an apples to apples comparison difficult at best.

    If you want a ballpark figure just take 5700XT results and reduce them by like ~2-10% to get Vega 64 stock/tuned Vega 56 results.

  21. I feel like a lot more cards should have been included in the conclusions graph (given that was already charted on previous reviews) so we could see where a larger range of cards would fall (even if the dots were marked as previous reviews or something).

  22. This might be a strange request and its not a traditional game but Warthunder can really tax a graphics card well. Its a F2P game with a high user base and may be a good consideration to add to the testing suite which would cost $0 to add.

  23. Vega will continue on as general compute solutions while RDNA will focus more on graphical prowess.

  24. Vega will continue on as general compute solutions while RDNA will focus more on graphical prowess.

  25. Also of note:
    “in DX12 and Vulkan titles, the game decides when to queue the frame”
    would be nice to include in your review write-up.

  26. Nvidia’s new Ultra Low Latency setting description may shed some light on AMD’s implementation:

    By reducing maximum pre-rendered frames to 1, you go to a just-in-time frame scheduling….Sure, that would improve latency, but it would also leave you susceptible to frame-time spikes if a frame takes a little longer than expected. I suspect that using VRR can reduce this effect, but still would be interesting to test. How many pre-rendered frames is optimal?

  27. So… Are the Radeon Vega cards an evolutionary dead end?

    I was disappointed to see the Nvidia 1080 ti card in testing, but not the Radeon RX Vega 64. If you’re making a generational comparison including the Nvidia 10-series and its follow-ons, I’d think you should include the Radeon Vega-series in comparison to the RX 5700-series. The RX 580 card shown in the comparison isn’t in the same price/performance range as the newer cards being benched.

  28. Oh Zak, you’ve done it again mate.

    I know any of our staff who were tasked with carrying the site’s major drawcard articles would give it every bit of justice you had – as you all do with your own posts. But to see what we all crave seeing on the site hold the same format, same prose, same detail as it always has – means an awful lot to a sentimental fella like me.

    [Site] Formatting and [staff] introductory niggles around the ownership change aside, I’m heartened to see the stalwart content keep coming (Fish, Bruno, Ben, Josh, Nath) and the variety growing. Erik and Aaron, you’re both excellent additions to Staff and thankyou very much for your pieces, which complement the site’s tradition well.

    From a long-time lurker to each of you – thanks for holding the torch and keeping the dream alive. It might not be reflected in every comment section (…), but it’s truly appreciated and you’re doing a damn good job. Keep up the top work.

    Justin (nah, I’m not going all Adi – just this once!)

  29. The “all-white” background is going to take a lot more getting used to. On my desktop, sorry say, but it looks really amateurish. Since there is no drop-down menu with the relevant page titles, it would help us a lot if you put the “Page listing:” at the bottom of each page. So we can skip to pages that interest us. What I do now is, I open the page with the “Page listing:” in one tab and open the pages I like in another tab.

    The reviews itself is great as usual and thanks a lot for that.

  30. Just for the record, Metro Exodus is also available on Windows Store – some people had issues with this version but after few updates it does work fine. Windows Store is also a home for Forza Horizon 4, which is a good game to benchmark as an example of driving sim genre of games, and as DX12 title it is much more optimized than something like Deus Ex Mankind Divided with patched in support for DX12 which is largely useless.

    As for FFXIV – just FYI, it has unlimited trial mode where you can play any character up to level 35, this does not cost any money. This is high enough to visit city like Limsa Lominsa, which is one of the most populated areas on most servers and has largest amount of people at evening hours. Running around main Aetheryte plaza with dozens of people nearby is a good enough test for this game, especially zoomed out. Of course, the official benchmark also outputs FPS results and I understand you don’t like using such benchmarks but I believe this would still be useful for some people.

  31. Metro Exodus wasn’t an option because I don’t use the Epic Games store.

    I attempted to benchmark Black Desert Online, actually. It is… just absolutely awful in a technical sense, to say the least. The options menu alone is a nightmare of repeated and inconsistent settings. I actually spent about 2 whole working days on it trying to get it to cooperate, but between the weird FPS caps (sometimes 30, sometimes 60?), vsync-while-vsync-is-disabled problems, inconsistent and hitchy performance on every machine and graphics card I have, plus the general problems of benchmarking a live online game with player interruptions, it just wasn’t worth my time. I agree it would have been interesting, but it’s just not possible.

    FFXIV, on the other hand, is a subscription game that I don’t care for. I own it, but I wasn’t about to re-up my subscription just to benchmark it, heh. Besides which I haven’t played since “A Realm Reborn” came out, so none of my characters are at a point where they could get to the interesting areas to benchmark, anyway. Historically we don’t run canned benchmarks but I suppose I could have banged out some numbers on the official Shadowbringers benchmark, since I do have that downloaded.

    Anyway, thanks for the comment.

  32. TR left out the RTX2060 which will continue at $350 and is in direct competition with AMD 5700.
    The other “SUPER” cards supersede the previous models (RTX2070 SUPER completely replaces the original 2070), but the original 2060 will continue.
    It seems TR was not sourced one and hence did not include it in the graph.

  33. On some pages you mention about “Jeff’s write-up” and you link to another article, but that article has a different author name, without “Jeff” anywhere. You should correct it to not confuse the readers (yes, I know Jeff wanted his name removed from TR).
    Also, it is disappointing you did not use some other DX12 games like Metro Exodus. It is a perfect game to test performance with RTX enabled and also useful for comparing GPUs without RTX effects. It is also disappointing that you did not select some popular MMORPGs for benchmark, something like Black Desert Online or FFXIV. I know that they run well even on older hardware but a lot of players are still curious to see how the new card will perform when they start considering upgrading their GPU. Plus most other hardware sites do not use those games for benchmarks so it would be a unique content in this regard.

  34. Thank You ! Tech Report Staff for making this possible.
    Many of us I am sure have waited for this review. And it was smart of the Tech Report to wait for all the Geforce Super cards and the new Radeons to arrive, and showcase them against each other. Being the most productive choice.
    I like how all the new driver features of the products were showcased.
    Was this Zaks first major video card review? It was well done.

  35. Die sizes are irrelevant since AMD is clearly paying a fortune for 7nm silicon or else they would have launched these chips for a small fraction of the price of the RTX parts to grab market share. Furthermore, TSMC’s “12nm” is the 16nm process with a couple of tweaks and a marketing name change.

    As for the data center you should have paid attention to Nvidia’s most recent earnings beat where the data center was actually down a good bit but overall results beat the street and –unlike AMD — Nvidia just raised it’s outlooks for the rest of 2019.

  36. It makes me wonder just how much of Nvidia is currently propped up by datacenter sales – their die sizes compared to AMD are monstrous.

    Also, Nvidia is at 12 nm, not 16.

  37. Adding a page title listing to the content of the first page would help a lot, since we lost the convenient dropdown box after the refresh. Some people just want to see benchmarks for a certain game and don’t want to click through 14 pages to find that.

  38. Thanks for the review!

    As for the product well, let’s say that 7nm has allowed AMD to avoid some of the worst issues with previous cards essentially being overclocked out of the box. But given how much guff Nvidia has gotten for dedicating silicon to RTX, it’s also pretty telling that their 16nm parts (2060 super in particular) are still competitive even when you never turn RTX on and even when you look at power consumption.

    I think this calls for some market disruption by a third party (and of course I mean S3!)

  39. It’s good to see essentially price-parity on the XT / 2060 Super. That’s good news for anyone in the $400 and below market. The standard 5700 looks to stand on its own pretty handily between the 1660 Ti and 2060 Super, it’s slightly better value than either if you look at 99th% FPS / dollar.

Leave a Reply

Your email address will not be published.