Gadgetory

Intel's ri9 9900 k is most boasted feature in all marketing is its solder or s Tim or soldered Tim worse Tim so we decided to test their moles with the new soldered interface then D live the cpu and put the Roll paste back on it because apparently we didn't get the memo we'll be looking at soldered versus paste tests gaming benchmarks blender workloads overclocking and live streaming benchmarks in our review of the i9 9900 K before that this video is brought to you by the be quiet dark rock 4 and dark rock pro 4 CPU coolers these high-end coolers to focus on a smarter approach to air cooling by adding a mini fin stack on top of the direct contact cold plate adding small bumps to the fins for increased service area and by using silent Wayne's 135 millimeter fans custom-built for high performance cooling without too much noise the pro is a duel tower cooler rated for 250 watt TDP while the dark rock 4 is built for 200 watt CDP's learn more at the link of the description below very quick note before we get started we have a new limited edition shirt on the store it's this graph logo shirt that's up there on stored on Karen's ex's net once they're gone we're not buying more of them and it's a basically a quad foil shirt so check that out on the store so the most important pieces of information if you missed all the other stuff the new CPU of the 9900 k is an 8 core part it's got 16 threads it cost 500 30 bucks and it clocks all core to 4.7 gigahertz with single core about 5.0 plus or minus a bit you can overclock it fairly easily we pushed 5.2 without any effort really whatsoever and we'll be doing a live stream later which will put the date and time on that on the screen where we will be doing some more serious overclocking so those are the absolute basics intel is calling the cpu ninth gen they're calling the whole well the whole series nice gen and in reality it's eight gen it's coffee lake it's coffee lake architecture but it's been updated a bit so the silicon's new it's a new piece of silicon it's bit larger thigh size which factors into the thermals but it's not truly a new generation so keep that in mind as well other pricing for reference this is 530 bucks so the new CPU today you'll see that we've taken hours apart already and the 8700 k previous generation sort of is 370 bucks present the 8086 a bit over that the 2700 ex has recently fallen and is now $300 and the 2700 9x is $250 and you could fairly trivially overclock that to become a 2700 ex so basically you get a 2700 X equivalent for 250 bucks so AMD's most direct competitor is the eight core 16 thread 2,700 X or 2,700 overclocked CPU and that's what we looking at for gaming benchmarks primarily the first thing to start with here on this new CPU is definitely the solder which we've replaced with thermal paste we went backwards we did the opposite what we normally do so in tell us it's Tim we're soldered thermal interface material is the company's biggest bragging point for the 900k they've been pushing it hard and deservedly so this is what all of us in the enthusiast community I've been complaining about for years now so we deleted the 9900 k we removed all of the solder and replaced it with thermal grizzly hydro not they're all pasted to see how does a high-end thermal paste compared to the solder because not all solder is created equal and it might not even be better than the liquid metal approach we've been using on previous generation CPUs so basically we went backwards just to see what would happen and cue the clip of talking with Gordon from PC world we're in Gordon says we can't complain about it now right it's like we got it you can always complain okay so with that in mind let's investigate the 900 K and see if these solder has been worthwhile we did a few things for this test in first we collected all stock performance numbers and gaming numbers without deleting it so let's just make that very clear right away this was all done after the other testing secondly we originally had the clever idea to disable two cores on the 900 K and fixed the voltage and frequency to equivalent value says he'd find out in 8086 K after deleting the 99 hundred K and after speaking with their Bower we learned too that the 99 hundred K is not close enough to the 8086 K to even fudge a comparison not even if disabling two cores the 99 hundred K actually has a thicker dye package atop the substrate which impacts results in ways which we can't or with our clever trick the dye is also larger which means it dissipates heat over a wider area and will run cooler for this reason our 8086 k results we ran are actually not as comparable as we wanted instead we will compare the 9900 k with solder to the 99 RK with hydro not if you'd like to see the results with liquid metal we'd strongly encourage you go visit their borrowers channel and watch his content he did additional testing with the lidding and collaborated with us on our results will be doing liquid metal in our own testing later and that's their Bower MIT's acht getting into our results let's highlight the 9900 k numbers only so just highly all 9900 k numbers here's the fun part first highlighting the 9900 k with solder clock to 5 gigahertz and with all cores enabled and set to 1.3 for 1 volts after V group is accounted for the results operated in an average core temperature of 64 point 4 degrees Celsius over ambience with ambient logs every second we also logged currents into the EPS 12-volt rails every second which is very important for data accuracy at 64 degrees core and 14 degrees over ambient liquid the most direct comparison is the 9900 KD lidded running 8 cores also at 5 gigahertz this one operates at 69 point five degrees over ambience for a rough 5 degree increase over the solder that's 5 C by switching from solder to hydro not the thing Intel bragged about so much its solder is not much better than a high-end thermal paste it is worse than deleting and liquid metal but again their barrows got those numbers first and we'll follow up later keeping just the 99 100 K excuse highlighted we can next look at these six core tests we ran it's the same thing here with about 63 degrees over ambient for the D lid or about 50 9.6 for the stock soldered version the 8086 k tests are still here but as we learned from Roman and our own investigation it's clear that this data isn't comparable enough to not bother mentioning what we wanted to leave it for reference as for power and voltage during these tests here's what we're looking at power consumption for the 9900 k-6 core 5 gigahertz task was not 194 watts close enough to the 180 9.6 watt power consumption of the 1900 K 6 quart D lid to be comparable the 8 core variation of the nine hundred K measured 260 watts put up against 269 watts and our D loaded test this is within reasonable control although the deleted variants might be marginally cooler if we were able to drop at another nine watts voltage was also logged and measured it's one point three for one volts for all eight core tests in one point three five volts Ross six core tests there was a viju fish you as a result of a C's of BIOS and that's where those numbers come from solder is a reversion back to the Sandy Bridge era this is something Intel needed to do even though it's not perfect even though it's not better than liquid metal because it's a thinner sheet of it it's still better than what Intel was using and this is important and for that we praise Intel for at least doing solder you know it's not the best implementation the kind that we'd like to see but it's better and Intel this isn't a service of course to the community they're not doing it just because they got annoyed with everyone Intel's doing this because Intel needs the extra couple hundred megahertz that the thermal difference affords them Intel has a real competitor now and suddenly having solder is just enough on this 14 nanometer aging the hot process that it can get Intel another one hundred two hundred megahertz and allow some more overclocking Headroom so this was a necessary move and one that we are at least mostly happy with it's not perfect if you deal it and do liquid metal you'll still get probably better results based on what their bearer was saying and we'll have our own numbers on that soon but for most people we probably wouldn't recommend deleting at this point because the solder is good enough it's just that if you're really pushing the extreme then yes it might be worth going that route still it's just a hell of a lot of work to clean it up and we'll talk about that more later in a separate content peace stream benchmarks are next we define this testing and the article linked in description below if you want more information I will play some side-by-side clips without revealing the CPU is while explaining this testing for the basics we're testing with OBS we're capturing gameplay while streaming at various quality settings generally faster is a good enough h.264 quality setting and is typically what we use in our own streams fast and medium improved quality at great performance cost but quickly descend into placebo territory at medium and beyond still it offers a good synthetic workload to find it leaders be the most practical use cases because these CPUs are powerful enough that they have no trouble at faster and even fast most of the time we are testing with fortnight and dota 2 on the 9900 K and the 2700 X when both are stock fortnight is set to high as is dota 2 and both are 1080p and streamed at 60 fps we also measure baseline performance without any active streams to better understand performance loss from streaming streaming is heavily multi-threaded so for people who want quote multi tasking benchmarks this is it starting with streamer side fps and fortnight we observed a baseline framerate of 258 FPS average when not streaming at all note that this is streamer side this is what the player sees not what the viewer sees we'll get to that in a moment and viewer side is arguably more important we were hitting GPO constraints at the top-end in this benchmark the lows remained moderately time that 161 fps 1% and 117 fps 0.1% lows are 720 700 X is next in 2 operates at 200 FPS average for its baseline performance in terms of frame times we're looking at about 3.9 milliseconds with a 9900 K vs. 5 milliseconds or the 2700 X with the 9900 K streaming at 10 megabits per second and using fast h.264 encoding we lose about 27% off of the baseline and land at 188 FPS average loads remain spaced proportionately to the average the 2700 X drops to 124 FPS average losing about 38% off of its baseline that doesn't mean its worst necessarily because we also need to look at viewer side frame throughput coming up next remember there are two pieces to this and at 12 megabits per second for medium encoding this increases the quality at a huge performance cost with the 900k of falling to 170 FPS average with low is now entering worst territory at 43 fps 0.1% low the 2,700 X Falls to 118 FPS average when using the 12 megabit per second medium and code quality here's a frame time plot of the 900 K at 10 megabits per second faster is 12 megabits per second medium and baseline for these plots the most important metric is consistency of the line alongside the lower overall values lower is better but more consistent is better than lower we're looking at the interval frames a frame with 16 milliseconds equating to 60 fps the 12 megabit per second medium result occasionally spy close to 80 millisecond frame times which can create a noticeable stutter or latency difference between surrounding frames and this is why these frame time plots are really important they're not used enough when we're trying to use them more because it shows results where sometimes they would get averaged out even with 1% to 0.1% lo metrics here's the 2,700 X plot performance is overall lower as a result of the lower frequency but the baseline and 10 mega section fast results are still relatively consistent although we don't see a spike up to 80 milliseconds this time must also note that a single frame at 80 milliseconds is not the norm on the 900k either we've seen consistent frame intervals of 40 milliseconds hit throughout the test on both products when streaming with the torture workload of medium quality settings let's take a look at some side-by-side footage of fortnight at 10 megabits per second and fast we won't reveal which CPUs which until the end of this clip for the streamer the takeaway is that the streamer gets a good experience on both platforms Intel is technically superior here but the cost of the 1900 K is a consideration as well we'll talk about that in the conclusion for now either platform looks playable to the streamer what we need to know is if the output is any good to the viewer as that is the most important part of the equation that's what you've been looking at here in the playback as well let's reveal the CPUs and then move on viewer experience has boiled down to percent of frames delivered at 1080p 60fps via YouTube in our testing both the 2700 xn9 900k were able to deliver 100% of frames at 10 megabits per second and fast encoding which is perfectly adequate for any streamer really the 12 magnet per second medium quality setting is entering placebo territory though still has some benefits with fast encoding the 9900 K delivered 87% of its frames within 16 point 6 7 milliseconds or 60fps the 2,700 X delivered 96.5% of its frames within the same window we've seen this behavior before and found that Intel stabilizes its delivery when manually managing process priority this has something to do with task scheduling on each device with medium settings in solace 9900 K is impressive in its ability to still deliver a consistently good viewer experience at 98% of all frames encoded the 2,700 X delivers just 68.4% of its frames in the same test a result of its lower frequency let's get a side-by-side of these two up again with this quality setting it's leaning into placebo tears we want to emphasize that the 2700 acts is still perfectly good for streaming and gaming simultaneously you just want to keep it to 10 megabits per second and fast an adult can maintain higher quality settings but it may be an unnecessary level of quality overall it just depends on how serious you are about streaming ultimately a secondary system would still improve low on frame times and is probably what you want to use if you're a professional highly competitive stream or something like csgo or dota moving on to dota 2 we start again with streamer side fps this is what the host of the stream sees dota 2 tends to favor Intel CPUs hard for frequency dependence following Amdahl's law well and positions the 9900 K at the top of the chart the CPU output to 191 FPS average with low is mixed at 124 fps n 55 FPS 0.1% the 2700 X manages 144 FPS average baseline when we start streaming those numbers dropped to 151 fps for the 900k with fast settings or 134 fps went under medium settings the 2700 X drops to 92 fps and 84 fps respectively and here's some side by side of the 2700 X and 9900 K when using 10 megabits per second fast settings overall both are reasonable performers for the streamer the 12 megabit per second medium frame times get a bit jumpy particularly on the 2,700 X with a 0.1% low of 20 but it's not terrible and ultimately you should be using fast anyway your site performance is where it matters for this one the 9900 K and 2,700 X boats managed to encode 100% of frames at 10 megabits per second fast this means that viewers will see all 60 frames per second when viewing the stream and so will not be able to perceive a quality difference between the 99 30 K and the 27 X they are functionally the same planning favor to the 2700 experts value proposition although it is behind in the streamer side performance if you did want more quality each 12 megabyte per second medium playback retains a lead for the 900 km functionally 100% of frames encoded the 2700 X falling to 92% of frames encoded you're looking at side-by-side footage now of the 2790 900k at 12 megabits per second medium will reveal them at the end of the clip and again it's not terrible and can be compensated for with a permanent overclock or with a slight reduction in settings for the 2700 axis performance the 2700 X does better here than what we saw in a fortnight part of this down to resource allocation and how the games work of course frame capping the for either game on the streamer side would also help and would pick up performance on the stream side the output as well although that may be inadvisable for some ultra competitive players if you need every single frame on your end but maybe a secondary system is better for you anyway let's review all those CPUs before moving on power consumption while streaming is an interesting topic this chart is for fortnight power consumption with the ASUS Maximus 11 hero that we used for our 9900 Kay there's a stricter adherence Intel stock policies than with some of the other boards this means we see a sharp drop-off in power consumption when testing under full stock conditions the CPU Falls to 100 watts a load and stays there measured at the EPS 12 volt cables leaving more performance available if we were to remove power targets and limits some of the other motherboards shipping today will exit these Intel power specs and draw more power than what you're seeing here it just depends on what board you're using and how well they follow Intel's policies the 2700 X pushes closer to about 120 to 125 watts power draw in town manages to achieve better overall combined throughput for both the player and viewer side experience while maintaining a lower power consumption for which the 9900 Kay deserves acclaim Intel has done well here to optimize their output maintain high frequencies and not suck down a ton of power until you overclock it which is something we'll do later this is of course at a significantly higher cost than the 2700 X competitor and that's a massive factor that will play into our conclusion getting into the game benchmarks we start with Far Cry 5 this one runs on the dunia engine by Ubisoft and is also used in our GPU task bench at 1080 P V 9900 Cape which is 157 FPS average 119 FPS 1% lows and well times 1 to 2 FPS 0.1% blows overclocked the 1900 K ends up at 163 FPS average for a reasonable gain of 4.5% over these stock to 9900 K the 8700 K ends up around 155 FPS when overclocked to 5 gigahertz meaning the stock 9,900 cat in your equivalents these stock 8700 case it's closer to 141 FPS average when looking at the i5 CPU we noticed a clear drop in frame time consistency as the frame to frame interval became more sporadic represented by the i5 eight 600k 0.1% low metric the r7 2,700 at 4.2 gigahertz meanwhile ended up at 111 FPS average within margin of error of the overclocked r5 2600 at the same frequency target and again the 2700 at 4 points you gigahertz and the 2600 are within variance of each other at this point and can be considered roughly equal the 900 K has a clear lead in far cry 5 and performance follows a similar trend as an assassin's creed which you'll see later both our Ubisoft games but they are different engines although those engines may share some code I have 1440p performance becomes capped at around 147 FPS average as the GPU is that leaned on more heavily United I heart ek maintains 146 FPS average the 8700 K at 5 gigahertz again meets its performance and the 8600 K does the same except its thread deficit does pose problems for frames of frame interval consistency the 2700 at 4 points you gigahertz post roughly the same performance as before as we'd expect because it's almost fully CPU constrained the overclocked 2,700 2,700 X are not fast enough to keep up with the 1440p throughput under normal settings with a 20 80 TI if we were to boost options the high settings and fully leverage the GPU then the natural expectation is that all results will be dragged down to meet the GPU cap f 1 2018 runs on the ego engine by Codemasters and is our first representation of a game of that nears 300 FPS it's at this point that most people would probably be happy with just about any CPU on this chart even the r7 1700 and it's 183 FPS result the 99 100k posts an impressive 284 FPS average one stock thanks to coupling it with the 2080 Ti and the overclocked pushes it to 291 FPS average for the 8700 K at 5 gigahertz we see a 270 FPS average result that permits the 99 hundred KS stock CPU elite of 5.2% if we compare the stock 8700 KS to 48 fps average to the 99 hundred K the difference is closer to about 14% that's with both stock the 2700 at four points you gigahertz ends up at 212 FPS average with Louis still reasonably times this posts a 27% improvement and the 900k although it's sort of misleading as a stat at this point to state the percentage improvement fps scales nonlinearly a 27% gain at nearing 300 fps means a whole lot less than a 27% gain at 60fps the difference from 48 to 60 fps using this example is often noticeable in gameplay whereas the difference between two hundred and seventy FPS and two hundred and twelve FPS is really not observable to most people although certainly some pros could probably tell the difference moving on some 1440p the 9900 k is now squished down to 240 FPS average by the RT x20 atti bottleneck this limits overclocking improvements heavily and also caps the nine hundred K stock see beauty 238 FPS average the 8700 K is just behind at 235 fps and the 2700 at 420 gigahertz manages to a 1 FPS average this posts a slight dip in spite of the GPU limitation versus the previous round we've seen this in the past with AMD CPUs like in battlefield 1 from last year's testing Assassin's Creed origins uses the anvilnext 2.0 game engine by Ubisoft and runs on DirectX 11 at 1080p and medium settings we are minimally GPU constrained in an otherwise GPU intensive game the 9900 K leads the pack at 135 FPS average 1 stock and also illustrates that we are hitting GPU bottlenecks at this point the overclocked only gains us 2 FPS or 1.5% and so we can't see the unconstrained performance in this game still we balanced between realism and proper methodology here and going too far one way or the other doesn't really result in a great benchmark going lower than medium is unrealistic and so will point toward our other unconstrained tests for examples of the top and performance like f1 2018 still the 900k leads the 8700 K is stock CPU is 112 FPS average by about 21% on both our stock overclocking the 8700 K pushes it to 123 FPS average closing the gap but only because we are limited by the GPU on the 900 K this is a realistic example of how even a 2080 TI can become bound by 9900 K and so it's still important to show when the gains of a higher on CPU become CAF by even and unrealistically or unreasonably expensive video card the 2700 X was only tested in Assassin's Creed origin and ashes to illustrate that it is functionally the same as an overclocked 2700 or just below it anyway so that's here just as a reminder of that performance in this game we see the 2700 X line between an overclocked 2700 at 4.2 gigahertz and a stock 2700 as you'd expect there are 2700 post 104 FPS average results this puts it ahead of the stock 8600 k and led by the 9900 k a stock CPU by about 31% this is also why we recommend buying the non X CPUs in most instances because it's so easy to overclock and you save some money the 900k holds a clear lead here that's not unexpected at this point at 1440p assassin's creed origins illustrates a clearer GPU limit at 122 FPS average where the 9900 case it's under both configurations the stock and overclocked results are tied with the difference of being within margin of error 8700 K isn't far behind at 117 FPS average and the 2,700 at 4.2 gigahertz posts a 102 FPS average result differences absolutely still exist at 1440p but they do diminish for civilization 6 we use the AI benchmark to analyze turn time processing this is represented in a unit of time not famous per second but rather seconds and so we're looking at the number of seconds to process a single turn this is taken from an average of five turns processed multiplied across four runs if it takes a CPU ten seconds to process one turn and you have six AI players in the game that would be about one minute per full rotation before it is the players turn again that starts to get noticeable a faster CPU reduces this wait time resolution is also irrelevant here the results would be the same at 1440p the 9900 K when overclocked posts a turn completion time of eleven point four seconds remember lower is better and so it's leading the pack the stock 9900 K operates at eleven point seven seconds per turn with these 79 60 X overclocked to four point six gigahertz tied at eleven point seven seconds as well so a mistake seems like frequency to some degree so we are seeing a boost in the 79 60 X from twelve point six seconds when pushed to four point six gigahertz this statement frequency dependence following on dolls law is further reinforced when looking at the 8700 K and 8600 k5 gigahertz results both of which end up at around eleven point eight and eleven point nine seconds so our seven twenty seven hundred four points u gigahertz ends up at twelve point seven seconds per turn the total time requirement to complete turns is reduced by eight percent on the nine nine hundred K versus the overclocked 2700 which is the same as a stock 2700 X it's an eight percent time reduction in the most suitable comparison GTA v runs on the rage engine from Rockstar and is our oldest tested title a popular announced remain on the bench for now at 1080p and with high and very high settings the 99 hundred K operated around 172 FPS average boosted by the five point two gigahertz overclocked to 179 FPS average note that we are running into engine limitations here best illustrated by the dismal i5 results for the eighty six hundred K this is a story we busted open about a year ago but it's worth reiterating when i5 CPU is bounced off of the 187 FPS marker in GTA 5 they stutter insanely and instantly hard to resolve this you'd actually want to run a lower average framerate in GTA so that it avoids the frame limiter at the high end this issue seems to primarily affect I 5 CPUs as we discovered previously it's clear that Rockstar also hasn't fixed it since the discovery either way the primary focus is elsewhere the 87 arcade stock CPU posted a 159 FPS average with the overclocked results at 168 FPS average that's a market gain which puts it near the 9900 Kay's stock CPU like most games GTA does it like frequency first and foremost our 720 700 are 4.2 gigahertz again basically 2500 ax ends up at 143 FPS average allowing these stock 900k about a 20% lead this is about what we'd expect at 1440p scaling remains the same or about the same it's just the ceiling has dropped down the 900 K now caps at 160 FPS average when overclocked or stock as do the 8700 k + 79 60 X CPUs this is a result of hitting the GPU limit the r7 2700 at 4:20 gigahertz still remains below that limit and so posts a similar result to its result at 1080p we're at around 143 FPS average on the 2700 X stand-in here blender performance is up next for our in-house GN monkey head Ren that's built specifically for CPU stressing the 9900 EK finishes the image in 20 minutes marking it as 9% slower than the stock i 970 900 x x 299 cpu overclocking the 9900 k2 5.2 gigahertz pushes us to a 17 minute completion time roughly tying with a moderately overclocked 7 900 x as for the 2700 x that's represented by our overclocked 2700 at four point two gigahertz remember and he uses the same cpus all the way down the line so overclocking a 2700 to this degree is basically the same as 2700 x over fox 2700 ends up at 22 and a half minutes for a speed improvement of 11% on the stock 9900 k or 24% on the overclocked at 9900 k our next render test is a single frame from the jeon intro animation which will just go ahead and replay it back now for this one the 9900 k finishes the render in twenty five and a half minutes compared to the stock 7 900 X and overclocked 9900 K at twenty point eight minutes each overclocked 7900 X still manages a time reduction of 13% from its thread advantage the overclocked 2700 at 4.2 gigahertz completes the render in twenty six point six minutes allowing the 9900 K overclocked CPU a time reduction of 22% to twenty point eight minutes well that's worth the extra money is dependent on how time intensive your workloads are and now we bring it all the way around back to the conclusion Intel gets major credit for doing solder it's not the thing that everyone dreamed of when we look at the results but it's still way better than the therole paste until you used to use the dow corning paste now technically the 9900 k is also still the chart-topping gaming cpu so Intel gets credit for remaining yes as they say the best in quotes and it's in quotes because we're at best means a whole lot of things different people it's not the best value it's not the best price but it is the best just in terms of raw throughput whether or not that matters to you depends on how neurotic you are about your frame rate if you're the type of person who absolutely needs to hit 240 FPS on that new monitor and you've got a video card that can sustain it then sure but a lot of the time the 2700 X or e the 8700 K and Intel can't be mad for commending their previous CPU are both fine parts for achieving high frame rate gain and that's what you have to look at and figure out to you how much does having the fastest really matter versus having an extra couple hundred bucks because Andy's a real competitor now and Intel's not alone anymore in the arena so we have to consider AMD as an alternative these days when it used to not be the case and as always thank you for watching subscribe for more we go to store documents axis dinette to place your order for this limited edition Ron for the graph logo shirt with the multicolored foil that's it for this one thank you for watching I'll see you all next time