Gadgetory

today we're reviewing the GTX 1660 ti whose name is going to trip us up for the entirety of its existence the GTX 1660 Ti is Nvidia's mid step between Pascal and taurine keeping most of the Turing architectural changes to the SMS in the memory subsystem but dropping the official RT act support and RT core is in favor of a lower price the EVGA gtx 1660 TI that we're reviewing today the XC model should have a list price of a $280 sticking it between the $350 base line of the RT x 2060 and the rough $200 price point of modern gtx 1060 s although sometimes that's higher for further reference at vega 56 now sells closer to $280 to 300 with RX 590 is still around the $260 range today we're looking at the 1660 TI XC model and we'll have a teardown up separately if you want that before that this video is brought to you by the gigabyte z 390 ARS master motherboard which comes equipped with one of the more powerful z3 90 V RMS for heavier overclocks on the new 9th Gen Intel CPUs the ARS Master is also one of the few motherboards with a really heatsink this generation featuring a mix of high surface area fins and looks oriented hover blocks oh and it's also got updated RGV illumination learn more at the link below the last Nvidia card that we reviewed was the RT X 2060 of course it's the next most recent one and overall we were fairly positive on it it was the best showcase from Nvidia to date on its RT X series the price point made more sense the performance for the price made more sense in the scaling in terms of the percent spent over a predecessor versus that percent before one - gained felt good it really felt pretty reasonable compared to the 2080 for example which we simply couldn't recommend that launch especially the 1080i still out so Nvidia as the SKUs have come down in the to the lower end has improved it's sort of value proposition we'll see if that continues with the 16 60 TI today especially with and these more aggressive price cuts on some of its existing competitors the first question though is why the name and so Nvidia said the quote here is as far as naming goes under why 16 Series instead of just using 11 quite simply we felt that from an overall architecture and performance perspective tu 1:16 is closer to the other tu 10x parts than it is to the prior generation GP 10x T 116 has most of the Torian architectural features including the new dedicated cores for in 32 and FP 16 operations and it also has all the nutrients 18 features including variable rate shading and mesh shading and as you see performance it's closer to the g4s RT x 2060 than it is to the 10 60 in fact like you said at performance closest to the 1070 beating it in some games and losing and some others so we ultimately settle on 16 60 TI instead of 11 60 TI confirming that 16 is in fact numerically closer to 20 so fair enough fair enough on the answer to that that's why it was named that though if you were curious we obviously are not offended by its name but a lot of people were making fun of it rather fairly so we thought we'd just ask why that's the answer but there's more important stuff to talk about today so some differences in architecture versus Pascal quite a few it's actually as I said much closer to taurine if you were to just look at the spec sheet for this or put it up on the screen you would see that the 1660 ti has two times the SM count over the 1060 that's not a linear comparison because the SM layout change significantly from Pascal so just like the rest of the touring cards the SMS are now cut down to 64 CUDA cores per SM rather than 128 and that's why you see that difference of a two times increase in s times but not a 2 times increase in CUDA cores and there's still other differences that's not to downplay anything just to illustrate that that's a non linear number do not compare just SM count doesn't mean anything if you do that so to highlight some of the other differences this SM block diagram better illustrates the changes to the assembly out and taurine as a refresh we didn't see it before we now see FP 32 lanes alongside in 32 and FP 16 and 32 processing can now be done in parallel to FP 32 which has proven beneficial in some games with heavier compute workloads like Wolfenstein 2 or Sniper Elite 4 or shadow of war as opposed to by the way the 10 series and its preference to instead interrupt the pipeline with each instruction turning can now do them in parallel FP 16 also get some more focus here at P 16 has been really interesting since Andy started making that push for its rapid packed math discussion on Vega because for the most part we haven't really seen a lot about P 16 usage in games it's primarily useful for things like learning a machine learning where you can really throw away a lot of your data accuracy because you have so much data that it just doesn't matter if you have billions of images for example but FP 16 in games it's if you don't know about P 16 can be referred to as half precision it is less precise than FP 32 but FP 16 is still useful for speeding up the pipeline for things like for example apparently according to Nvidia water effects and water simulation in Far Cry 5 we haven't validated exactly the water simulation performance or behavior in Far Cry 5 but according to Nvidia that is one area where FP 16s gettin usage in the gaming world and we'll see if that sort of starts to spread more because it would uplift performance significantly when you can get away with it you could not get away with it for things that need accuracy like accurate physics simulation or hitboxes things like that so the 16 60 TI ends up with turns architectural changes the integer pipeline changes are in there that we discussed previously in an interview at CES with Nvidia but also it acts as the RT cores and the tensor core so price is offset and it's a non r-tx product and then finally a memory subsystem moves to the significantly faster GD d r 6 platform from gddr5 5x and so forth and the baseline native speed is 12 gigabits per second on the t1 16 and well with the memory that's used with t1 16 rather so other improvements also move over an ven see improvements for example but memories the big one the memory speed on most the card should be about 12 gigabits per second you can overclock it to about 14 plus as we did here today so let's get into the testing we'll start with probably Apex legends and look through the rest of our benchmarks Talk thermals overclock and stuff like that first up is Apex legends and we'll put some gameplay footage on the screen if you're not familiar with the title but of course note that this is not direct benchmark capture so apex legends is a highly populated branded DirectX 11 title for comparison we just posted a large benchmark study on this one the show performance scaling across multiplayer and different zones of the game as the map is at large and view distances are pretty long so that does impact performance heavily we ultimately decided that multiplayer testing is consistent enough for test patterns that we could get a realistic look at performance without having to use any sort of synthetic test the down is that we don't have as many cards tested in this game as we do for the others coming up so you'll have to look at our other game benchmarks for cards like the gtx 1070 but let's look anyway at how apex legends runs on the 1660 ti at 1080p and with all settings at high the gtx 1660 TI XE ends up at 92 FPS average with lows spaced reasonably behind this lands the 1660 ti as between the RX 590 fatboys 89 FPS average and Vega 56 is 106 FPS average the 1660 ti ends up about three percent better than the rx 590 which is way outside of our zero point seven FPS average standard deviation from the mean so this isn't far from GPU silicon quality boosting range of variants and they're really functionally about the same when it comes to gameplay the difference between the 590 and the 1660 ti is insignificant - not noticeable to the player frame times are functionally equally spaced on each with differences of landing within the larger standard deviation of our 0.1% lows and 1% lows notably versus the gtx 1066 gigabyte that the 1660 TI sort of replaces it ends up about 29% ahead as you'll see in our other charts this is comparable to a 1070 in performance the 1660 ti that is versus the new RT x 2060 however the one that preceded the 1660 TI m launched the 2060 leads the 1660 TI by 22% for average FPS overclocked in the 1660 TI with a + 125 megahertz core offset and + 11 20 megahertz memory offset gets it to nearing equivalence with the vega 56 card on the charge and narrowing the stock 2060s lead to about 10 point 6 percent before of course you could overclock it higher here's a frame time performance chart to better illustrate the frame the frame behavior of each of the tested devices lower frame times are better but more consistent frame times are best the 1660 TI ends up plotting at about 12 milliseconds per frame on average with variants frames of frame well under 4 milliseconds for perspective users do not typically notice frame to frame interval variants less than 8 to 12 milliseconds from the previous frame in this regard the 1660 TI does well for testing overall and plotting the overclocked 16 60 TI next we can see that performance again improves and does so without introducing frame time in stability this shows that the overclock is stable you would see a lot more spikes in there if not the rx 590 ends up similar in performance to the previous two lines although averages higher in frame times and therefore lower in framerate 1440p positions the gtx 1660 TI closer to Vega then 1080p does with this higher resolution the 1660 TI approves capable of frame rates equal to or north of 60 fps with high settings although obviously FPS can go up with the settings reduction we're at 67 FPS average one stock ranking the car dies about 9% faster than the XFX rx 590 or fury X although the fury X struggles fiercely in frame time consistency and as an objectively worse performance than either the 590 or the 1660 TI the 1660 TI also outperforms that you want kin 980ti by 13% but is led by Vega 56 by 5% as for the gap between this $280 1660 TI and $350 RT x 2060 that's a 25% price increase which amounts to about a 23 percent performance increase not bad really especially considering that we thought the 2060 was Nvidia strongest offering to date for the RT X series f1 2018 is next giving us another DirectX 11 game before getting on to Sniper Elite 4 for f1 2018 at 1080p the 16 60 TI ends up at 104 FPS average which is exactly tied with the Vega 56 reference card although a good bit behind our more recently tested Vega 56 red dragon lows remain advantaged on the AMD cards in this game the 16 60 TI outperforms the 980 TI marginally and out stretches the 1070 SC by 2.9 percent compared to the RX 5 nineties 82 FPS average the 16 60 TI runs about 28% higher in frame rates at 1440p the 16 60 TI ends up running at 80 FPS average landing it slightly ahead of the 1070 although functionally tied and just below the 980 TI again functionally tied Vega 56 reference is in the vicinity as well with the over clocks at 16 60 TI ending up below the red dragon kart scaling is about the same here as in the previous chart with the 16 60 TI ending up about even with the gtx 1070 better than the 90 and average frame rate and worse than the 56 in frame time consistency and if comparing against the better partner model 56 also the 56 red dragon does better in average FPS Sniper Elite for ease is direct x12 unlike the previous two games and also has asynchronous compute support for this one we run tests at 4k and at 1080p the game runs well at both resolutions on most cards but for the 1660 TI 4k acts more like a synthetic test to see how it stacks up against other cards we've tested at that 4k resolution the 1660 TI stock card ends up at 46 FPS average putting it just ahead of the rx 590 fat boy with functionally equivalent lows compared to the 1070 the GTX 1660 ti ends up marginally worse although overclocking puts it ahead of the 1070 se stock card and below the 980 TI hybrid our 1080p chart is much more limited for Sniper Elite as we only recently added it but the 1660 ti XC ends up at 126 FPS average stock ranking it as about tied with the overclocked rx 590 but behind the RT x 2060 overclocking the 1660 di pushes it to nearing 140 FPS average allowing the 2060 stock card a 13% lead although you could obviously overclock the 2060 for more gains GTA 5 gives us an old popular DirectX 11 title that's mostly getting content updates at this point and not seen much by way of driver or game optimizations at 1080p the 16 60 TI stock card runs at 95 FPS average placing it just between the Vega 56 reference and Vega 56 red dragon cards when both are running at stock settings lows are also nearby the 16 60 TI stock cards 95 FPS average outperforms the stock MSI GTX 1066 gigabyte gaming X is 83 FPS average by about 15 percent further outperforming the rx 598 gigabyte fatboys 76 FPS average by about 26% which is a stark contrast from our apex legends benchmark note that the GTX 1060 has always done better relative to the rx 5 ATM GTA 5 specifically sort of seeing this continue in the 16 60 TI is really no surprise it's just this particular game does favour the 10 series and now I guess the 16 series as odd I mean is the 1660 t-i-is outperformed by the stock 1070 SC in this title which runs at 106 FPS average and leads by a sort of an impressive 11% regarding overclocking the 1660 TI overclock gets us to 102 FPS average leading the 1660 TI XE stock card by 6.4 percent and nearing the vega 64 performance in GTA 5 at 1440p the gt-r 1660 TI XE stock card operates at 69 FPS average ranking it again between both Vega 56 models that we've recently tested and about nineteen point five percent ahead of the GTX 1066 gigabyte stock cards 57.9 FPS average overclocking the 1660 TI XE planted at 73 FPS average with the low as reasonably spaced for this title this allows the 16 60 TI to approach the Strix Vega 64 outpaced the 980 TI hybrid 73 FPS average and approach the 1070 SCS 76 FPS average although the 1070 holds a technical lead here the experiential difference is functionally zero to the player far cry 5 as we always say is valuable for its heavy reliance on long draw distances and screen space reflections at 1080p the 1660 TI XC operates at 94 FPS average ranking it as about 7% fashion the 980 TI hybrid and about 21% faster than the ex of x RX 590 stock card 78 FPS average although overclocked 590 performance isn't all that different the gtx 1070 runs roughly tied with a 16 60 TI with its 96 FPS average art ranking the 16 60 TI by adjust 2.5% ahead of this way gives the six reference and power colors Vega 56 rests just beyond reach of these 16 60 TI performance with these 16 60 TI overclocked Landing between the two at 99 FPS average the RT x 20 60 meanwhile leads the stock 16 60 TI by 14% although the overclocked closes that gap at 1440p the gtx 660ti operates at 66 FPS average for far cry 5 planting it has functionally tied with the GTX 970 se stock card and not too distant from Vega 56 the overclocked 16 60 TI ties the reference Vega 36 card 69 FPS average although is outpaced by the red dragon Vega 46 card 74 FPS average the 16 60 TI does reasonably well at 1440p with elevated grab settings and some drops to settings would allow obviously a higher framerate shadow of the Tomb Raider and it's modified crystal engine come up next for this one running at 1080p high we measured the 1660 ixc at 83 FPS average putting it as functionally tied with the gtx 1070 lows are within margin of error between the two the card ends up about 15% ahead of the RX 590 Fatboy 72 FPS average and even further ahead of the gtx 1066 gigabyte 62 FPS average wooden stock overclocking the 16 60 TI gets it 288 FPS average ranking the card close to the 1070 TI but still permitting the 10 so DCI elite of about 5% at 1440p the gtx 1660 TI XE proves that a minor settings adjustment would easily permit a 60 FPS average baseline although our standardized tests have the card at 57 FPS average with lows very close behind illustrating consistent frame time performance the GTX 970 SC ends up at 55 FPS average here so these 16 60 TI takes a slight lead the stock 16 60 TI ends up between the stock red dragon 56 and the two rx 590 benchmarks that we have on this chart power consumption is up next for this we're using total system power consumption as logs from the wall so this is not individual component consumption the test system is heavily controlled including all voltages in BIOS so the only thing changing is the GPU power consumption for our standardized ashes of the singularity power consumption test we see the gtx 1660 TI stock system pulling about 260 watts at peak with average power consumption in the range of 230 to 250 watts for total system draw we won't be plotting the overclocked 16 60 TI here because our XC model does not change it has a fixed power target of 100% and no higher the power consumption of the card peaks at about 130 watts and that's for gpu-z numbers not total system numbers and it stops there so just to be clear there's no a power offset for the 16 60 TI but we'll look at overclocking numbers next the RX 590 applauds after this showing us a higher ceiling at 325 watts total system power consumption the line remains study here as the boosting algorithms and power balancing are less aggressive than on the Nvidia cards the delta is between about 65 watts and 120 watts depending on which part of the test you're comparing for perspective against another Nvidia card.the r-tx 2060 total system draw ends up at about 262 290 watts so we'll just plot that line now as well there's not a massive difference between the 1660 TI and the 2060 when you're looking at the low end of one and the high end of the other so you have to be careful look at the overall average as the line does move based on what part of the test it's in overclocking the EVGA gtx 1660 TI x he proved straightforward seen as we had no control over increasing the power target that said EB J's card does run a 10 watt higher power target than the reference 1660 TI board designs and so there's some baked in overclocking Headroom it be about equivalent to a 108 to 109 percent power target on a reference board which has a power target of 120 watts whereas EVGA s does 130 watts an hour overclock stepping chart you can see that the first two rows illustrate no change aside from allowing the GPU to warm up the testing when it was just starting the test closer to 51 degrees Celsius the GPU boost into nineteen hundred five megahertz on average with the maximum boost frequency at nineteen thirty-five megahertz this was with no offsets of any kind as the card warmed up looking at the second row the average frequency dropped to eighteen seventy-five megahertz with a 62 degree GPU core temperature this is normal and is just how boost stepping behaves on NVIDIA but we like to illustrate it so that people understand the difference in cooler quality and what you actually get with a higher fan speed curve once we started really trying to overclock we hit about twenty fifty five megahertz peak frequency briefly but averaged about 20 25 megahertz with a two hundred megahertz offset to the core this proved unstable once we started to actually gaming and so we dropped the offset to 175 Bagger Hertz and hit an average frequency of about twenty ten megahertz core that's really not bad for a 16 60 TI with just a 130 watt power limit memory was stable with a an offset of 1120 megahertz resulting in an actual memory frequency of 1780 megahertz or about fourteen point two gigabits per second when looking at the effective speed if you look at the Watts column you'll see that we were at 130 watts for the entire test even when stock so this was just a matter of tuning the frequency to fit silicon quality we did not boost the fans to max speed for this test so of course is a bit more room there you can kind of look at the maximum frequency column to get an idea for where you would land with maybe chilled water for example thermals and frequency stability are up next starting with 3dmark firestrike extreme as representation of frequency over time we see that the gtx 1660 TI ends up at nineteen thirty five megahertz peak and only decays to eighteen ninety megahertz as the test drags on to the 30-minute mark this is good overall and is indicative primarily of a reasonable cooler and fan curve for the GPU core we sometimes see much sharper fall-off in frequency from peak with worst coolers for GPU temperature the card ramps to about sixty five degrees celsius and attempted to remain there for out-of-the-box thermals the gtx 1660 TI XC has a thermal target of about 65 to 66 degrees for the stock fan curve which is a reasonable temperature that still allows plenty of boost Headroom plotting fan RPM next we see that the fan ramps to about twenty-one ad rpm once at full load and steady-state and does not oscillate meaningfully from that baseline this fairly fixed fan speed reduces the more audible and noticeable pitch change that more variable fans would produce and so the fan curve works out well on this cart moving to the next chart our power virus torture workload places the sixteen sixty t IX C's GPU core temperature at once again about 64 degrees Celsius this remains stable throughout the test period with the fan speed ramping to about 2200 rpm during this load level let's just quickly swipe over to the next chart to show power consumption during the same test power consumption during the same power virus workload has the 1660 Ti pulling about 130 watts through the GPU core oscillating a bit depending on the load note that the power consumption is plotted against the right axis here not the left and this illustrates that this is the maximum power target for the EVGA card and is the same that you'd see with an overclock concluding that it's 280 dollar products well that's what we're told it is anyway it's before launch so it's not online yet but 280 bucks in theory for the AXI puts it between the well the most relevant products would be the our TX 2060 for which there is a pretty linear price to performance increase over the 16 60 Ti so that makes the math a bit easier there 25 percent 22 percent price increase versus performance increase in some of the games tested so not bad in terms of that scaling it is it is somewhat linear to the dollar spent you do lose the RT fee the RT acts features RT cores there aren't a lot of reasons to care about that in the gaming world yet there are far more reasons to care about that in the development world but in the development world you're probably buying higher-end anyway if you are maybe ultra budget indie developer then over 26 you'd probably make more sense if you do want the RT x features but that's entirely up to you and depending on what you're developing but on the gaming side you could probably drop the RT x stuff at least for now for today in favor of a lower price it's just you also lose other traditional rasterization performance because it's it's not a one-to-one otherwise you do lose some SMS and with the SM loss you is about a bit of everything else including CUDA course or the lanes on the vectors as it were so the bigger comparison is probably something like Vega 56 and these prices at time of us filming haven't fully taken effect yet but our understanding is that Andy's supposed to be dropping the price to about the same as the 16 60 TI we can't validate that today at time of filming so you'll have to go look but that's what it what should happen once this card launches so if that's the case then here's how we stack it up as we've said before Vega 56 and its entire existence is the easiest to argue for in the scenarios of enthusiasts overclocking if you just like messing up the hardware you like doing things like power play table mods to increase the power target you like to maybe change the V bios Vega is less restricted than the Nvidia counterparts so in those instances if you really just like messing around with the hardware from an enthusiast standpoint you don't care too much about performance that alone makes Vega 56 still worth buying so that's been our biggest argument for it it's just it's fun to mess around with and do tuning on so if that's what you're looking for it's still a good purchase from the performance standpoint the 16 60 TI does a whole lot to reduce power consumption in a pretty big way whether or not this is relevant will depend on where you live what power costs things that everyone has it seems different opinions on how much they care about our utilization so we'll just list that strictly as a fact and then you can decide what to do with that information but power consumption is pretty low on the 1660 Ti comparatively this cards got 130 watt power target in testing looking at the power from the drop the wall that looks to be about accurate GPZ validates further so it is a lower power target than a 2060 by a bit but 2060 is more performance to lower power consumption than the competing A and D cards does that matter well that depends on the price really how much cheaper is one than the other if you can go with higher power consumption if the price is lower it's a bit easier to justify as for performance the card is in a few different places but on average it's about a gtx 1070 so anywhere you see 1070 performance for the most part with few exceptions you could just say that's where this would fall to so if you're looking at older charts and you see a 1070 in a game that you want to play that's where the 1660 ti would be and hopefully that helps you make a decision on purchasing so this card $280 seems about right for NVIDIA did well - it's interesting because the 2080 was really not well received for its price position we we more or less condemned it and said buy anything I was like the 1080 TI at the time which no longer exists the 2080 ti you kind of justified because well best and it's hard to argue with best but it was still obviously very expensive the 2060 came out then and the 520 70s and suddenly looks like kind of heroes in the archaic landscape because the predecessors were so willfully overpriced that you end up in a scenario where and video looks like the hero to its own villain and it's a very interesting play from Nvidia which is largely uncontested let's rephrase that which is entirely uncontested in the high-end market and has some competition primarily in this market today where this card exists or the 2060 exists but not really at the high end Radeon seven sort of to some extent competes at this point but not up until recently so a very interesting launch strategy but we would say that it's working out well for the 2016 the 16 60 TI in terms of pricing because the pricing feels about right now and videos done well to to sort of recuperate its price position there and its value proposition performance wise it's fine it's a 1070 the 1070 does well in games you play 1440p no problem you play 1080 no problem 4k is a bit of a stretch but with sort of lower graphics settings lower fidelity games no geometric complexity you could do it for example pac-man but other than that 1440p 1080p it should be a really no problem so yeah just keep an eye on the prices for Vega 56 it's what its gonna come down to if you like tuning stuff 56 is a good choice and if you like the lower power target lower power consumption I should say and the competitive performance overall this is a good choice but 56 does outperform it in a few games and sometimes noteworthy ways so you'll just have to look game the game and and then check the prices online because I don't know what those are today so it makes it really hard for me to give you a firm yester now on 56 over 1660 ti when today at time of filming I don't have the numbers so that's it for this one subscribe for more as always go to patreon.com/scishow his next hops out directly or we go to store that carries excess net to pick up a mod matte like the medium mod matte on the table here that we just made and it sold that right away but we have them on backorder thanks for watching I'll see you all next time