The best AMD Ryzen 9 3950X 16-Core,Processor review
The best AMD Ryzen 9 3950X 16-Core,Processor review
pros
Performance powerhouse
Cheaper than HEDT
PCIe 4.0
Fits in AM4 socket
cons
Needs extra cooling
Limited gaming advantage
TEST SYSTEM SPECS:
CPU: 3.5GHz AMD Ryzen 9 3950X (16-core, 73MB cache, up to 4.7GHz)
CPU Cooler: NZXT Kraken X62
Motherboard: Aorus X570 Master
RAM: 32GB HyperX Predator RGB @ 3,000MHz
SSD: Samsung 860 QVO 1TB
GPU: Nvidia GeForce GTX 1080 Ti
PSU: Phanteks Revolt X 1200
Case: Praxis Wetbench
it's a really big day for all PC gamers and content creators and everybody who loves high performance processors it was only two years ago that AMD doubled the core counts of intel's competing mainstream products like for real this time and then just when it seemed like until it finally clawed their way back into the fight AMD sent us this for review the Verizon 939 50 X is a desktop processor that not only has 16 true multi-threaded processing course it even has a higher boost clock than any of its predecessors guys I haven't been this excited for a consumer CPU since I got my first quad core in 2006 I am almost as excited about it as I am about this segue to our sponsor team group SSDs are cool but our GB SSDs are even cooler the t force delta max SSD features a large mirror like luminous surface area for that sweet RGB goodness grab it at the link below and check out their early thanksmas sale ,when AMD announced this thing back at e3 2019 the biggest question everyone had was how I mean we knew that they were doing it with two separate two CCX dyes just like the risin nine 3,900 X
except that that processor not only had two cores per die disabled it also ran pretty toasty surely then with 16 instead of 12 cores and the highest boost clocks out of the entire risin family it couldn't possibly perform the way that AMD was claiming that it could except it does almost to the point where Andy is killing other more lucrative parts of their own product lineup nevermind the competition cuz here's the thing at it's not cheap but it's also taking the position that was formerly held by AMD's thread ripper series of CPUs and with only a few
trade-offs so first up you lose 40 CPU PCI Express Lanes bringing the total to 24 with four of those being used to communicate with the chipset so if you need a ton of PCI Express expansion thread Ripper might still hold some appeal for you but remember two guys PCI Express Gen 4 means that the link between the CPU and the chipset is effectively twice as wide as first and second gen Rison that means less congestion on that link even if you are using mostly PCI Express gen3 devices downstream so second you lose quad-channel memory and the extra
memory capacity that thread Ripper offers although the 3950 X can still handle up to a hundred and twenty eight gigs of RAM so most people will probably be fine especially given that the supported memory speed is far higher and as we all know that's super important for AMD Rison the icing on the cake here then is that you no longer have to contend with switching Numa versus yuma nodes for a given workload oh and the motherboards are much less expensive to get into too so all-in-all the 39 Dax is a faster CPU compared to first and second-
generation thread Ripper and as a gamer or even a prosumer you're giving up very little of value unless you're into multi way GPU setups or huge nvme storage arrays so you don't miss our review of the next-gen thread rippers now though it's the moment you've all been waiting for glorious it's been a long week we used a number of different platforms this time around so here's what we came up with
there's more AMD than Intel here because frankly speaking Intel doesn't really have a lot to compare against it and we just want to do a shout out here for puget for providing us access to a kora 999 60 X which Intel wouldn't send us for comparison presumably to avoid direct 16 core to 16 core you know graphs and wow I can see why they did that because this thing is a Productivity monster remember guys we're just looking at bars on a graph here so it's easy to
lose track but these are thread rippers and core i9s that we are comparing it to and the 3950 x is on top in virtually every test we threw in it thanks to its high core count and pike or clocks compared especially to intel's h EDT platform now pay particular attention to the benchmarks that measure time here we are shaving literally minutes off of the 3900 X which already trounced Intel's 9900 K and then comparing directly to the 9900 K the 3950 X finished our handbrake transcode in nearly half the time what's more impressive here is that the single
threaded and lightly threaded results are also quite competitive with everything else in our test suite last gen thread refer case is looking noticeably last gen right about now especially because thread rippers main disadvantage came into play when its professional users wanted to take a break at the end of the day and play some game here the 3950 ex shines though as advertised it is consistently as fast or faster than any of AMD's lower core count chips which means that we are looking at the processor that somehow combines the best of high core counts and high
frequencies guys this is the closest that AMD has come yet to decisively dethroning Intel in gaming and it's so close in some cases that if I was sitting over at Intel watching this I would probably need a new pair of Underpants at this point by the way LTT store comm but then all of this must come at a price mustn't it and not just one that hits your wallet right wrong so we were expecting the 3950 acts to be of a blazing inferno but under synthetic load we are nearly
10 degrees cooler than the 3900 X and that things down for course now as we can see it does dip slightly below base clocks under these conditions but so too does the 3900 X and if we were hoping for the power consumption to reveal some horrible secret tough luck fare to the 39 50x also manages to draw 4 watts less than the 3900 I like what blood sacrifice did team red make to do this when we look at it closely we see that no blood sacrifice was required in fact the 39 50x achieves this through some very very careful power management it actually runs at a V
core of less than 1 volt under load in prime95 and only just more than that incentive an R 20 it's actually only in lightly threaded workloads that the voltage really ramps up and there well it doesn't affect total chip power consumption as much because there's so many fewer course that are actually drawing power so thing is while AMD recommends that you run out and buy a beefy cooler for the 3950 X unless you're overclocking I'm not so sure because this thing is easier to cool than a risin 7 when we ran it with a memory overclocked and the BIOS
performance enhancements enabled to relax precision booths restrictions yeah it did draw more power and it got hotter but again not by a huge amount the core only went up by about 10% and we're still within 10 degrees of stock under load not to mention that the performance improvement this got us in multi-threaded workloads put us well above our stock results and even got us on par or better than our stock core I nine ninety nine hundred K in terms of gaming so what then is this just really good binning we're paying for Hana like Intel's ninety
nine hundred KS that they recently released probably I mean there's a reason that AMD you've decided to hold off until now to release this chip after all but now that it is here it seems like if you've always liked high-end desktop CPUs for their huge core counts but hated their outlandish prices and expensive motherboards you finally have the processor that you always dreamed of honestly the guys this thing is incredible it's crazy fast it works on reasonably priced boards even has ECC memory support if you want to use it in a budget workstation this thing is so good but if Andy wants to have any hope of us recommending a third gen thread
Ripper they are gonna have to pull a serious rabbit out of their hat .
The best AMD Ryzen 9 3950X 16-Core,Processor review
the AMD r9 3950 x16 core CPU following up the 3900 acts except at this establishes the 3950 access functionally an h EDT or high-end desktop CPU except on a mainstream socket the 3950 X it doesn't have a modern direct competitor and is flanked by the 9900 K and 3900 X in the range or 79 80 X II and future 10 9 8 exe at the range we've tested the 39 50 X stock and overclocked for power consumption thermals video editing performance production workloads like blender and Photoshop gaming and more and today we're reviewing
the AMD risin 939 50x CPU before that this video is brought to you by EVGA r-tx 20 atti XC ultra the 20 80 TI XE ultra is what we use in our cpu reviews to avoid GPU bottlenecks the XE ultra uses hydraulic dynamic bearing fans for reduce the noise features our r-tx support for DX our titles and uses a massive 2.75 slot cooler the cooler design allows the fans that's been slower and quieter while syncing heat further leveraging a mix of l-shaped and traditional fins to maximize airflow or contact learn more at the link in the description below so quick important note on this one first we are going to try overclocking this with liquid nitrogen the day this
video goes a lot so that's going to be probably later in the afternoon we'll shoot for somewhere around 4 p.m. to latest 7 p.m. Eastern Time start time likely pushing for 4 though for that it'll just depend on how a launch day goes for the CPU so check back what we're gonna do in that stream if you don't care about the overclocking the first part of it we're going to explore the thermal scaling for frequency versus liquid nitrogen temperature so what we'll do is we'll use
liquid nitrogen as a tool to keep it first above ambient for example 90 degrees Celsius all the way down to minus 80 - 90 maybe even minus 150 depending on how far we can drop the temperature and that will allow us to see how does the frequency scale without human input with just colder temperatures because this is like a GPU or a boost that way now so that'll be today keep an eye out for that for the stream for the rest of this for the review today a couple of things this is a 16 core 32 thread processor is advertised with a boost at 4.6 gigahertz as a
reminder that does not and has not for an extremely long time mean anything other than 4.6 gigahertz single core AMD's rise in 3000 series it means something a bit special beyond that though which is that it's the single peak opportunistic boost under some special circumstances we'll talk about that today as well the advertised TDP is still 105 but AMD has changed its cool recommendation to a to 80 millimeter CLC which changes the thermal resistance for the formula to get TDP out so we're not sure how the change in theta CA is not reflected in the 105
watt number unless they also change TK s and T ambient to adjust it anyway that number is not really accurate or useful for anything for a user and we have a whole video on that but that's that's the basics here that's enough of that there are a lot of charts today we're going to get straight into it before getting into any performance results we need to better understand how the CPU is performing when stock this first chart is for all core average frequency during blender and uses an intentionally zoomed in y-axis to better demonstrate the frequency
behavior we'd prefer a scale starting at zero but in this instance it just looked like a flatline and wouldn't be useful this is a fully stock test other than our custom memory configuration for reviews looking at the beginning of the benchmark we get a reminder that rise in sea views behave like modern GPUs they boost dependent first and foremost on things like thermals and power budget as we illustrated in our initial 3000 series reviews with - 82 positive 80 degree charts to get full scale in one stock in this chart using the same 280 millimeter X 62 CLC at full
speed that we always do the rise in 39 50 X EPA started with an average frequency of 39 63 megahertz but fell down to 39 24 megahertz within 100 seconds the CPU averaged about 39 thirty megahertz all core frequency overall fluctuating about plus or minus 10 megahertz once steady-state was achieved and thermals were flat this is in blender which is an AV X workload that loads all cords evenly and constantly this next chart will help us better understand these single thread scenarios running Cinebench single threaded we measured the maximum core
frequency has typically around 40 through 40 megahertz with a couple of spikes to 45 20 megahertz upon starting the task before any sustained we did have a 45 seventy-five megahertz clock but it couldn't sustain we're pretty far away from the 4.6 gigahertz number on average which will influence obviously results compared to any SM use that may hold higher to each CPU is different these days so it's not like the old days of Intel where stock CPUs were the same the short of it is that it's out of our control we could switch CPUs and hope for a better silicon
we could switch motherboards but neither of those is a realistic solution for the consumer you get full boost when there's no load on the CPU and when it isn't doing anything as the CPU has power and thermal budget to do so we don't think it's fair to say that the 1350 X does 4.6 gigahertz if it's only under extremely limited conditions so until Andy fixes this via GC pushes or marketing if one of those as possible we question the legitimacy of such claims being able to achieve 4.6 at minus 100 degrees doesn't count but we'll see but AMD won't get a pass from the
community forever for stuff like this Intel for example would be crucified if it did the same and Andy is not the underdog anymore it's the it's it's the equal dog if that's a thing so is that a thing the point is Andy's outstripping Intel sales and DIY enthusiast buy for two watt in our audience right now so Andy can't get the underdog past forever on its marketing we were curious about if the slightly older BIOS would make a difference so we updated this chart with different BIOS we're at f10 a instead of f10 see then reap lauded in Cinebench r15 one threat
the result was a clock that appeared more stable at forty three forty to forty to fifty megahertz rather than the forty three fifty two forty three hundred bouncing around that we got on f10 C but we also didn't get any of the Upward Bound spikes toward forty five twenty or forty six hundred at the end of the day technically speaking the f10 CBO scored a little bit higher but we're not sure how much of this is variants we ended up going with f10 C for full testing as it's the most recent consumer ready BIOS and although more variable it had some spikes toward
the higher frequencies so with the X 62 CLC closes liquid cooler we were able to sort of get 4.5 all core it held for a little bit in blender which is our current stability benchmark it's one of the more intensive workloads so a 4.5 all core on this thing sort of held but we did eventually lose it and we'll have to push higher with a better cooling solution Allen T will stand in for now but we think with this chip if you ran it under a big custom loop or with perhaps maybe just a 360 with faster fans for example you should on our chip be able to do four or five not necessarily at 24/7
voltages but it should be doable we ended up settling on four four and we did four four at a set voltage of one point three six two five which after accounting for everything like LLC which we only had a turbo for most of this the GATS of voltage the voltage it actually runs at is something closer to one point three to five volts for V core SPI 12 and it changes but if you look at the motherboard readout instead but if we go by that number to one point three to five which is really damn good when compared to the 3900 X's requirements to get similar clocks we really
struggled with that at four point four we needed to turn out some t off basically so four point three was the common overclock for the previous set of three thousand three CPUs blender was the only application where our one point three six two five or whatever was set voltage was not enough to sustain so we changed that to one point three eight or so for set and the gate voltage is something like one point three five for a four point four gigahertz all core for our CB reviews we only test with the multiplier over clocks and I kind of leave it there this one can't do nineteen hundred like our other CPUs in this family can but it does get up to eighteen sixty
seven megahertz and that's close enough so this is within typical range of achievable F clock horizon three thousand as a side note with extreme cooling F clock you actually have to drop to fourteen sixty seven or so because otherwise you can't get about five gigahertz because you can't get cold enough and it doesn't like cold Adobe Premiere will be our first production benchmark our first render benchmark is at 1080p for a trade show are on GE ng style shoot this video is rendered in approximately 2.8 minutes with the 1350 X whereas the 3900 X
required 3.3 minutes for the same render despite boosting Delta's the 39 50x still manages to reduce the render time requirement by a massive 15 percent it might not sound like a lot in the scale of 30 seconds here but most reviews will take a lot longer than this to render and that'll calculate out against longer projects to the 79ad XE which is now 18 core part that will soon be replaced by the 1090 DX e managed to complete this render in 2.9 minutes overclocking the 79 ad XE or the 1980 XE it's the same thing just one is soldered and higher
base clock ends up posting a render time a 2.4 minutes a 14% reduction from the 39 50x stock CPU although we think we can do higher with a bigger cooling solution the 4.4 gigahertz RC is the max we can pull out of this configuration with this solution it's fair after all because we use the same cooling limitation for all the other CPUs on the chart so we're not going to break the rules and get a special bigger cooler for AMD for this specific thing that's why we're doing Alan to you later for this setup the 1350 acts manage 2.7 minutes not a worthwhile change from a
stock as for the cheaper 9900 K the 39 50x runs and 22 6% last time the 3900 X already beats that in this test though so there's nothing new here Vista 900 K as a quick aside the 29 t WX doesn't play well in our premiere renders relative to the lower core count CPUs so it's just not a good value for this workload nobody has limits to what it will efficiently work with with a 4k b-roll sequence for the next chart Adobe Media encoder completes the render an eight point
two minutes stock for the 3950 X nine minutes for the 3900 X stock CPU and 7.7 minutes for the stock 79 ATX e CPU and eleven point nine minutes for the stock 99 hundred K until doesn't have an exact one-to-one in price with the 39 50x right now but it does have two CPUs on either flank the 9900 K is beaten by the 39 50 X and the 3900 X pushing us toward on average and AMD recommendation for this specific task the 79 80 XE beats the third and 50 X the runnin time reduction and that will continue to the 10 9 8 exe but both are expensive in terms of value
and ease 3950 X undeniably offers a good value and performance sat between two Intel CPUs for our Adobe premier benchmarking Adobe Photoshop is up next we use a set of filters warps transforms translates and resizes to benchmark Photoshop against image manipulation capabilities for the CPUs the route metric for this is time required for each task and then we convert that using the puget suite into a score higher is better this one is often the counter to the more even thread distribution and applications like premiere premier wasn't always as well
balanced as it has become although it's still not great it's much better and Photoshop may one day go that direction too for now though Photoshop really likes free see our example we always give is to highlight the 5.1 gigahertz 9900 k + 9700 k @ 5.1 as well squaring roughly equivalently which underscores just how little the extra threads matter in this application it wants frequency first the 9900 KOC then remains the chart leader here and that be true of the KS with the same OC it's the same thing just more efficient the 3950 x ends up remarkably
close to the 9900 KS and actually does finally achieve a technical victory over the 9900 K of course the 9900 K is cheaper and users who primarily live in Photoshop could save by opting for the 9900 K instead of the 3950 X but achieve better performance with an OC and rough equivalence without the upside to the 3950 x is that it will prove better in blender up next and in premiere previously photoshop holy remains an Intel favorite scenario the 3950 x critically demonstrates that just like adobe premiere this is a stronghold that Intel will fast lose if it
doesn't push something to market that's better than a refresh Intel is losing ground genuinely this time and needs to find a way to try and get back in there the 3950 x runs approximately 3.8 percent faster than the 3900 X and 13% faster than the 16 core 79 60 x1 stock the 39 50 X is about 2 percent faster than a 4.6 gigahertz 79 60 X indicating an architectural advantages and overclocking the e39 50 X doesn't appear worth it here with our thermal solution blender is next this is another real production workload it's not synthetic we replaced Cinebench with this years ago way before Intel was replaced in Cinebench so we were the hipsters of using blender
and we have finer tuning / benchmarks with blender to make it last longer to allow us to better show scaling between processors in the upper tiers we also use blender internally for our own animations and product designs like those you can find on store documents accessed netware we're planting 10 trees with Ian's reforestation projects for every single item that you order through November anyway for the results our GM logo blender render required 12 minutes to
complete on the third + 50 X 1 stock remember that this is a tile-based renderer with one tile spawned for thread present that makes blender scale perfectly with more threads present not one-to-one but very well the third a 50x stock result puts it between the 18 core stock 79 80 X E and the 16 core overclocked 7 960 X which is extremely competitive placement for AMD the 9900 k is obviously no competition in a battle of threads and the 798 exe now a much better deal than it was still isn't cheap enough to be worthwhile in this workload against the 30 or 50 ex
that changes a bit with an overclock or the 99 8 exe eat the same thing at 4.5 completes the rendering 16% less time and a stock 30 or 50 X although the power requirement is obviously substantial the 2990 WX still has value here completing the same rendering 25 percent less time than the 39 50 x stock although it's double the cores and threads it's not half the time requirement so blender does not scale linearly as stated it's still like some frequency as for the 3900 X comparison if you're wondering whether you should spend extra at the 50x
reduces the render time requirements Y was 22% whether that matters will largely hinge upon if you're doing this type of work for money or for a hobby if your money-making ability is bottlenecks by your rendering pipeline and requirement well it's probably worth the upgrade the GN monkeyhead render is a different type of workload for this one the 3950 x stock cpu finishes in 10.2 minutes which puts it up next to the 790 DX e stock cpu actually requiring 12 percent less time despite the core deficit and near the overclocked 796 the x 16 core cpu the
overclocked ATX e obviously chart tops here but the 4.4 gigahertz 3950 x encroaches on intel's h EDT territory in a serious way the 2990 WX is not on this chart because we need to rerun this particular task for that one but we'll do that before our thread ripper 3 review in a couple of weeks let's do one more production test before games will speed this up a bit for 7-zip looking at decompression performance the 30 m50x ends up second only to the 2990 WX the stock 39 50 X runs 186 MIPS achieving 86% of the performance of the 2990 WX that's not bad for half the
course these 79 60 x @ 4.6 gigahertz allows the stock 39 50x lead of 15% and that's with an OC for the Intel 16 core the 39 50 X leads the 3900 X by 31 percent a notable improvement for those who care about decompression workloads and for time budget will skip v-ray this time the stack is mostly the same as blender anyway with a 39 50 X favorite gaming benchmarks are up next we'll start with our most recent edition despite it having more limited stack of CPUs in the chart Red Dead Redemption 2 also has some serious for engine issues for high frequency
medium thread-count processors like the 9700 k which appears to be a game side issue that red Dead's team is working on the lows on the night 700 k that you see here can be easily worked around by running higher graphics settings and imposing a bottleneck on the GPU cutting it down to 130 fps cap for example in this respect that dismal 0.1% low performance is only applicable to read that to at 1080p with something like a 9700 k and it can be kind of fixed check our previous research video for that the 3950 x runs at 140 FPS average with low is
pasted similarly to the rest of the high-end cpus at 82 FPS 1% and 73 fps 0.1% 3950 x can't quite reach the maximum frame rate achievable under a 20 atti with these settings which is about 151 FPS as illustrated by the 9900 KOC and stock results both bouncing off of the GPU limit this also means that we can't fully see where the 9800 K would perform when completely unlimited by the GPU but this is still a realistic scenario no one's going to lower resolution to 720p and we've actually never ever tested at such a low resolution below we've never tested low 1080p for CP reviews except for really really old ap use that were low-end so we cut it off at
1080p for anything that's DIY enthusiast and that's the way it's been basically forever so that's our floor the night I heard K then has more theoretical Headroom if we were to drop the GPU load with low settings or lower resolution but it's not useful here the GPU is the limiter period with this game the 99 hundreds about 8% ahead of the 39 50 X when both are stock that puts it roughly with an error of the 8700 K stock CPU and about 3% ahead of the 3900 X stock see view at 1440p we imposed GP bottleneck at the top and even more and restrict the chart down
there's no point retesting the CPUs that weren't already at the top of the chart since they're just gonna score the same value 700 K now has good performance because the GPU has disallowed the 97 RK from entering the engine framerate threshold for bad performance on 8 threads the 3950 acts at 14:40 P ends up at 118 point 6 FPS average so we've got some AMD specific GPU driver overhead that is coming into play here the 900k stock establishes a three point four
percent lead which isn't enough to worry about that's a measurable difference and a repeatable one and outside of error but it's not a meaningful one we are about now so it makes sense that they're nearly equal borrowing whatever overhead there is civics is next we use turn time benchmarks - this one turns our process for every AI player in the game so longer single turn times are multiplicative against the AI player count the 3950 x completes the average sieve turn
in about 30 1.4 seconds when stock and 30.7 seconds when overclocked in our benchmark the 3900 X for reference finishes the average turn in thirty two point eight seconds the stock the stock 3950 X takes about four point three percent last time to complete than the 3900 X the 9900 case stock CPU completes the turn in twenty nine point seven seconds or in five point four percent less time than the 39 50x stock CPU and as a reminder it's also cheaper so if you are only gaming from what we've seen so far the 1900 K is going to be a more cost-effective choice
than the thirty-nine to the X but once you start adding other work on that may change GTA 5 is next giving us an older rage game to look at the 9900 KS at five point two gigahertz and nine hundred KS stock still have a five FPS gap between them demonstrating that the GPU bottleneck is distant from the 3950 X the 3950 X ends up at 114 FPS average one stock just outside of error from the stock 3900 X and not meaningfully different if you're between the 3900 X and 3950 X and mostly gaming and that's those are the only two you're considering then
the 3950 X so far is not a meaningful change that's a different story again and thread heavy work if you do actual work outside of gaming the 900k stock CPU leads at 131 FPS average an improvement of about fourteen point three percent over the 3950 ax stock CPU the lows are not meaningfully different between Intel AMD Intel has a technical lead but the difference is nearing or within errors as lows are inherently an average of less data 1440p illustrates a GPU bottleneck imposed at about 132 FPS average shown by the United hundred KS at 5.2 gigahertz the stock 3950 x averaged 113 FPS average with lows at 91 and 84 for 1% in 0.1% overclocking
provided no meaningful uplift to the 3900 X ran at 108 FPS average letting the 3950 X lead by about 4.8% the night I heard a stock CPU hits at 122 FPS average leading the 3950 X stock CPU by about 7% when constrained by the GPU there would be a bigger Delta here if unconstrained illustrated by the 1080 P chart previously but the GPU does become a limiter at some point and is a realistic consideration insult tends to maintain a slight lead even in these scenarios due to some type of overhead but it's reduced hit man to at 1080p is our next one here
the 3950 accents about 127 FPS average with the 3900 acts at 121 FPS average the 9900 k'naan s maintains a lead at 136 FPS or about 7% we also know that it isn't a GPU bind here as we have at least until 144 FPS average before that kicks in for f1 2018 at 1080 P the 3950 X runs stock at 277 FPS average with low spaced as expected for this title the 3900 X stock CPU managed 276 FPS average demonstrating functionally 0 difference even the lows are nearly exact which is saying a lot for numbers this high because it is a wider error the 9900 K runs at 313 FPS average a lead of about 13% over the 30 and 50 X 1440 P has the 39 50x stock CPU at
221 FPS average allowing the 9900 case Toxie be a lead of 10% the 900 K is at the GPU limit here but there's still room for scaling at 1440p we can see it's limited by the GPU since the 900k stock and 9900 KOC results are equivalent and stuck at the same frame rate shadow of the Tomb Raider at 1080p has the 3950 X running at 154 FPS average although we noticed a 0.1% low and frame time consistency dip with our settings in this title this correlates with the 2990 WX in game mode which would be an equivalent in core and thread count as game mode
disables half the cores it seems that this game is one which struggles with the higher core counts on AMD cpus in the very least and is likely indicative of a latency issue performance overall is ok but the experience would be better on a 3900 X or on a 9900 K assassin's creed origins is up now for this one the 1080p configuration has the third 950 X doc CPU at 136 FPS average measuredly but it remarkably ahead of the stock 3900 accent 134 FPS average the night I heard K maintains the top spot technically at 1:40 with a lead of about 3% lb atlast frame time
consistency 1440p doesn't change too much the 3950 x runs 116 it's average here which is about tied with the 3900 x9 her case stays a bit ahead and averages though mostly in a meaningless fashion lows are fine on both but AMD has a tactical victory in the lows and a technical loss in the averages if we're gonna play that game let's close this out with just total war before we get to the power benchmarks total war Warhammer two's campaign benchmark has the 3950 axe benching at 138 FPS average which is about in line with you twenty nine ninety W axe in game mode this is one of those games where the excess cores are actually hurting performance it's
still okay and you can certainly play at a high frame rate relatively but there may be games that we haven't tested where disabling cores like this one would be advisable if running into issues with games on these 16 core tries in parts disabled in a couple of course should be an annoying but functional solutions just annoying because of the restarting involved to really do it properly the 9900 K isn't even in reach running a massive 30% to lead this is a some kind of scheduling or resource conflict that we don't see for many games it's in a few but overall the 3600 being
ahead here should be an indicator that something's wrong on a software level and that's not to say that the game can improve it because this isn't as simple as being quote unquote poorly optimized if it just simply didn't build it for an architecture like this let's move on to power some quick power benchmarks are next and then thermals after that we measure power consumption at the EPS 12-volt rails which is more accurate than a measurement at the wall both have their place but EPS 12 volt rail measurements are much closer to the actual power
numbers for the CPU only and only failed to account for things like vrm efficiency losses which aren't much on our test platforms with a CPU only all core workload our GM logo render with blender has the 3950 acts more power efficient than the 3900 X the 3950 X is a better bin than the 3900 X so that's why at least for our two samples anyway and it can be run at a lower auto or manual voltage than the 3900 X for this reason the 1350 x ends up requiring lower voltage and therefore consumes less power the 3950 X was 137 watts for the stock test the 3900 X
required 147 point 6 watts for the same under Auto stock conditions manually tuning both we see that the 4.4 get your Hertz 1.3 5 volt 39 50 X and a 251 watt power draw rapidly outpaced the lower core account 3900 X at 170 watts for a 1.3 4 volt OC this is why the vrm quality mattered a lot if you were buying for future proofing so to speak although the 30 or 50 X is more efficient when stock it's every bit as capable of being extremely power hungry it's all relative though and better bins will be lower voltage at a given clock but more cores will draw
more power once you start pushing things the 9900 K for reference runs at 91 watts when running within the Intel spec with MC e disabled the 9900 K is more power efficient for the framerate and gaming workloads but is less capable and less power efficient and applications like blender Sinnott ban charge xx is one of our less populated charts for power as the benchmark is newer the 30 or 50 X runs at 136 watts for multi-threaded rendering and blender while the 9900 K runs at 146 watts if let's surrender for longer like in blender the 9900 K would drop in power measurements that's because the 9900 K hits PL 1 at 28 seconds and since
Cinebench completes within that time period the power consumption is allowed to run at on average one point two five times TDP although that can be changed by the motherboard maker speaking of TDP and these number is obviously still a made-up fairytale and useless and we have a separate review on that topic if you're curious as to why Intel isn't totally free of criticism either of course to be fair but anyway the 3900 X was at 148 watch for this test establishing better per clock efficiency for the 3950 x1 stock overclocking the 39 50 X got it between 246 and
262 watts depending on the OC and the third Y hundred X was at 173 Watts when overclocked thermals finally the last chart if you'd like to see why we were so bound by thermals for overclock this chart will help illustrate it first off remember that thermals aren't a fire strikes core it's not a number that you can compare between reviewers or your own hardware you don't get 90 degrees Celsius you don't get 60 degrees Celsius another reviewer doesn't get 70 degrees Celsius it's all depending on the test environment the hardware the fan settings the cooler all of that stuff so this is purely useful for real comparison and give you a baseline of
a2/ad CLC for the full stock settings on the 39 50 X running an X 62 Kraken CLC at max pan and max pump speeds and an ambience of about 22 degrees Celsius on an open bench we hit steady-state nearly instantly in blender and maintained about 60 to 62 degrees Celsius for the hottest of the two CDs see CD one was cooler and how that around 57 degrees Celsius our 4.4 gigahertz overclock required a lot more voltage to hold in blender than it did for some other tests so we ended up finalizing settings at about one point three five volts to one point three
seven five volts depending on what the test was in this instance the core set of one point three eight or so with a turbo LLC was one point three five volt SPI 12 or one point three nine if going with the motherboards readings either way the OSCE ran at around ninety five degrees Celsius for a sustained blender run at our settings definitely not recommended without a bigger cooling solution but totally achievable if you used a custom loop or a 360 with super loud fans
or something wrapping up to 39 50 X we're actually fairly positive on this one it's a CPU don't buy this for your gaming only machine that's not what it's meant for it's a waste of your money and that's all there is to it but if you're doing things like not gaming work for example you do a lot of compression a lot of decompression you work with Adobe Premiere don't we Photoshop even though 3950 X is breaking a lot of the rules where previously AMD was just not doing well in Adobe applications and that's changed it changed with the 3900 ax and in
Photoshop it's specifically changing sort of with the 3950 X Intel is still a very strong competitor and Photoshop specifically and probably the better buy from a strict value standpoint 900 K gets roughly equivalent performance to a 3950 ax and our Photoshop testing at a cheaper price point but the important aspect here to look at with Photoshop specifically is that Intel is not going to have its stronghold in these types of applications forever even if Adobe doesn't continue to advance the software and it has done with premiere Intel is losing ground as I am
the advances so this is actually becoming a pretty serious situation and if doesn't actually do something aside from a refresh sometime soon it will become a dire situation but anyway the 3950 axe is not the only processor you should be looking at of course for less production focused pcs something where you're not doing tile based rendering not doing premiere that's we in Photoshop 900 K is still the best gaming option for the most part and if you are doing some of
those production workloads but don't have that much money there's no reason to discard the 3900 ax unless you make money off your computer where you find yourself bound by rendering times in a way that impacts your ability to make money there are a couple of quirks with gaming just like with the 1950 X the later the 29 WX these higher core count CPUs have some games that just really hate them and don't know what to do with them so in shadow of the Tomb Raider and in total war for example performance was in total war anyway in the
campaign benchmark was below an hour 5 3600 it was tied with a twenty nine ninety W X with game mode enabled which has the same amount of cores and threads so that's an example of a game where it doesn't work out well it's still playable it's still fine but if you wanted peak performance you either buy something else or in those specific games you turn off some of your cores which is a huge pain but at least it's doable and you can do that through rise in master if you're afraid of BIOS be stealth restart in terms of where the CPU is best suited we really like how it's doing with the premiere for the price it's still cheaper than Intel's 18 core it's a losing
the the 18 core is actually losing against and these 16 core 39 50 X and some of these benchmarks and yeah you can overclock it but it doesn't really gain enough back to be worth it and of course if you care about this type of thing the power requirement for doing so on the 18 core Intel CPU is is substantial so the third and 50x we think is a great fit for someone who is probably making money or trying to make money with their computer doing things like premiere and again even Photoshop if you're only Photoshop probably don't buy this but if you use Photoshop in addition to stuff like premiere you're a heavy user of both the 35 50 X is a
great solution if you are a user of blender 3d animation and modeling software that does any kind of tile-based thread dependent rendering the 3950 ax is at its price point a great cpu choice for your solution make sure you get a motherboard that has a an appropriate vrm though because 16 cores is a lot and not all these boards it's a mainstream a m4 socket you can buy a board that's meant for an r5 3600 and use it with the CPU if you want but it won't be a good experience so do some research on that we have a video on the best motherboards for Rison
with a m4 and you can figure out what you need for the third and fifth DX from that review in terms of what's next well Intel's got the 10 9 8 exe coming out we'll need to look at that obviously it is a bit of a refresh so we're not sure how much the stack will change with this CPU but it's going to be it's kind of in the ballpark but if you're going you're probably not stretching to a thousand so Andy's kind of out alone at the soda price point that's going to wrap us for now this was actually a pretty straightforward test for the
CPU no major issues or headaches like we had with Rison 3000 and that's because most it's been worked out now and overall we are positive for the CPU just remember there are certain places where it really doesn't make sense and then a lot of places where it's incredibly good so links in the description below as always.
Learn more about Ryzen 9 3950X
ModelBrandAMDProcessors TypeDesktopSeriesRyzen 9 3rd GenNameRyzen 9 3950XModel100-100000051WOF
DetailsCPU Socket TypeSocket AM4# of Cores16-Core# of Threads32Operating Frequency3.5 GHzMax Turbo Frequency4.7 GHzL2 Cache8MBL3 Cache64MBManufacturing Tech7nm64-Bit SupportYesMemory TypesDDR4 3200Memory Channel2PCI Express Revision4.0Max Number of PCI Express Lanes16Thermal Design Power105WCooling DeviceCooling device not included - Processor Only.
The best AMD Ryzen 9 3950X 16-Core,Processor review
To find out more from Amazon link below
Post a Comment