Gadgetory

we reviewed the AMD r5 3600 extremely positively just before this review went live this is for the r9 3900 X the 3600 was a strong start to and these third generation of Rison processors or a second generation of Zen architecture and set the stage for continued dominance in the $200 price range of desktop CPUs with that framing in mind it's time to look at the r9 3900 X a $500 12 core 24 thread CPU that competes most directly with Intel's I $9.99 hundred K priced roughly around 45 to $500 for a frequency focused platform this review will look first at production workloads like premiere blender Photoshop v-ray 7-zip and more and then turn to gaming power consumption and additional benchmarks before that this video is brought to you by gigabytes X 570 ARS master motherboard built for the true 14 phase vrm and for a high-end rise in 3000 series builds the gigabyte X 570 master uses a properly finned heat sink for VR I'm cooling accompanied by RGB LEDs in the i/o cover to makes it looks and performance gigabyte also includes BIOS flash features to update the motherboard BIOS without a CPU or RAM the over built CRM is the major feature on this board though and gigabyte has a brand new BIOS for overclocking the new Rison 3003 CPUs learn more at the link in the description below so like the r5 3600 review we're going to get straight into this today because there's too much data it's too dense we can't really delay on stuff we might have some additional information for you in the opening of the 3600 review so even if you don't care about that processor which you should you should watch the first couple of minutes of it after the ad break and see if you can get answer sending additional methodology questions the basics though 3200 Hertz for everything that's done because it works on all the platforms we're testing today including the first generation of Roizen so we got a fair comparison across all the platforms and also it's pretty common to buy and we'll do memory scaling later secondly windows 1903 is used for everything here using the latest BIOS updates for Intel platforms and we're using FC 5 for the gigabytes master motherboard the X 570 master for testing here and then additionally timings have been heavily controlled that's all been stated in the other video so with that said the 3900 X is the target today is 5 CPU or reviewing that particularly as compared to the 9900 K which is another roughly $500 CPU and I think with that we can get into it so overclocking will be included power consumption and just about everything else that you could want for the most part we have thermals coming later as well and these are r9 3900 X is highly competitive in v-ray which is a Kaos group renderer that has grown in popularity among 3d artists the array is measured in time until render completed so lower is better as it means the work will be proofed or finalized faster the 3900 X completes our v-ray benchmark render in 0.75 minutes one stock in 0.72 minutes when an all-court overclocked step four point three gigahertz a time reduction of about 4% like the r5 3600 reviewed the RO 9:30 isn't a huge overclock but we couldn't get it to 4.4 gigahertz with SMT enabled the 9900 K stock CPU following TDP spec completes the render in 1.1 minutes marking the stock 3900 acts as requiring 33% less time to render the 9900 K at 5.2 gigahertz manages in 0.95 minutes behind the overclocked 3900 X by about 14 seconds the 2700 X at 4.2 gigahertz finished this render slower than the stock 9900 K at one point one seven minutes so this is a big leap from core count for AMD and generationally Adobe Photoshop is a very different story for AMD this chart has a clear divide under the 90 to 600 K at 5.1 gigahertz where Intel rules uncontested above that this serves to remind us all that not every application is 100% thread bound and can utilize all of those threats Photoshop likes frequency and that's something we always demonstrate by highlighting the 9700 K and the 9900 K both at 5.1 gigahertz where there's a 2x thread differential but nearly identical scores that's because it's frequency bound the are 3000 series has picked up considerably compared to previous Andy generations with thanks to IPC and frequency uplift but this frequency dependence is still shown on the 3900 X the 3900 X stock CPUs scored 1053 points allowing the stock 9900 K a lead of 2% this is impressive generationally for AMD where the 9900 case stock led the 2700 ex stock CPU by 24% in this case the overclocked 3900 X is within margin of error in this test there are some workloads that only run a few threads which means that a stock 3900 acts with a higher boosting stock turbo table will do better than an all core OC which is lower than some of the individual core clocks when running with limited thread workloads so all core is higher than stock all core but it's not higher than the stock limited thread workload boost overclocking the 9900 K propels it ahead of its stock results by 7% and ahead of the stock 3,900 X by 9% our next benchmark is a cash test our Janu compiler collection benchmark compiles GCC with GCC in a sequin environment but we've also validated on mingw with emphasis - and see similar results when we asked Wendell of level 1 text about this he described it as quote cash hits all the way down note that we are not benchmarking compilers against each other or code quality when compiled but compile time only other compile solutions may favour Intel more heavily but presently this is what we use to test we may add more compilers if we get good suggestions from the programmers in our audience with the r9 3900 X tok cpu the 64 megabyte l3 cache pulls its weight with risings victim cache particularly useful compared to intel on this test as we noted in 3600 review the r5 3600 stock CPU can outperform the 2700 x OC 4.2 gigahertz CPU here because it has more cache at 32 vs 16 megabytes for l3 the our 939 hundred ex pushes to a seven point one minute compile time reduced from the our five thirty six hundred seven point nine minute results by 10% and time required the all core OC doesn't matter here 7-zip testing is next for this one we're using 7-zip compression to benchmark millions of instructions per second per device the ro933 PU reaches 100 3435 mips during compression workloads and that positions it far and away above the next processor which is the 9900 k stocks 66,000 mips a lead of 57 percent for AMD this is a workload that is heavily thread dependent and that shows here overclocking - 4.3 gigahertz all core doesn't do much but it does provide enough uplift to get us 1.2 percent over the stock result for the 3900 the i9 9900 K at 5.2 gigahertz jumps to 70,000 553 MIT's but it's not enough to come close to Andes compression score at a similar price 7-zip decompression is next in this testing Intel doze comparatively significantly better than in compression verses itself but not vs the 3900 X the 3900 X still holds a strong lead it's 140 mm mips vs. 94,000 moves for an overclocked at 9900 K roughly which puts AMD at about a 52% lead in this workload when the 3900 X is stock and the 9900 K is at one of its best points stock to stock the 3900 axial leads the United 900 K is 85,000 MIPS results by 68 percent compared to AMD zone r7 2,700 x of last generation the lead is similar as the 2700 is roughly tied the 99 hundred KS stock CPU previously blender is up next we like this one because it's alter that we actually used for instance we designed our GN a 3d teardown crystal with blender to build a model in 3d space and get it laser etched into glass so this is something that we have a lot of experience with blender is also popular for animation and is the biggest open source solution to 3d modeling and animation with blender 2.79 the GN monkey head stress test creates a mix to work load on CPUs that's heavily thread dependent in this test the ro 939 hundred X doc CPU completes the render in twelve point eight minutes a reduction in time required from the 99 hundred K stock CPUs 20 minute results by 40 percent that's a big climb for the dollar amount and is thanks in large part to the thread increase compared to AMD's our own r7 2,700 X CPU we see a 10-minute render time reduction stock to stock and overclocking isn't too different note that overclocking at 2700 to the same frequency 4.2 gigahertz wouldn't have the same results as the 2700 X at 4.2 gigahertz because they are the same aside from frequency well point out that CUDA rendering is often faster than CPU rendering for blender but there are instances when we use CPU rendering internally instead for the most part these instances are when we exceed cuda memory with a large object or project file as we can run more system memory than we can VRAM or when we need to render a lot of smaller tiles in a flat image with a lot of threads which can go faster than the larger blocks of CUDA our mad mad is what taught us this part despite being a flat image we made it in blender and for this we find CPU rendering to be faster than GPU rendering as we can crush the tile size down to be more efficient to fit the higher thread count CPUs where one tile spawns per thread so that's a potential use case if you didn't know about it anyway overclocking the 9,900 cake gets at the sixteen point seven minutes which is a huge uplift from stock down 18 percent in render time but still not enough with the GN logo render a more intensive raytrace seen the stock r9 3900 ax finishes in 15.6 minutes with an overclocked finishing 5 percent faster at fourteen point eight minutes or five percent less time the intel 9900 k stock CPU takes 26 minutes for this render with the 2700 x stock cpu at 28 minutes and 2700 x4 points you gigahertz all core cpu the same as a 2700 at the same frequency finishing in twenty-six point six minutes this meant that previously Intel was technically in the lead for blender despite an obvious price advantage for near equivalent of the 27 or XOC now though Andy not only leapfrogs Intel at the price point but also completes the render significantly faster now this changes as you go up into h EDT parts but the price gets blown out of our comparison for today for those using the CPU renderers at home the 3900 X is an easy choice in this sub h EDT class 4 cycles rendering Adobe Premiere CC 2019 renders our next for this one we're rendering at one of our 1080p videos from a show floor basically an RNG or an Eng environment using all a roll and b-roll our other test is a 4k 60 encode comprised of a roll and b-roll in the studio with some luma tree and audio correction applied for the 1080p render the AMD r9 3900 X surprised us where the r5 3600 and Andy's previous are 720 700 X couldn't beat Intel the ro 939 hundred X is now finally managing to pull ahead which is a milestone for AMD premiere first favors frequency and IPC but it does make use of additional threads somewhat efficiently just not typically enough where it helps AMD this is why the 31 75 X is better than a 1990 DX e is better than a 9 800 K at rendering premiere despite the obscene thread count on the 31 35 X and lower frequencies it does actually begin to make a difference with that many threads anyway the 3900 X finishes the render in 3.4 minutes when stock come through in three minutes when overclocked with the 9900 k completed in three point eight minutes that's a time reduction of eleven percent which considering the thread count is 50% higher is a nonlinear increase that highlights the importance of frequency in this test overclocking the ninety-nine hundred K to five point one gigahertz gets it to three point six minutes but that's the most we get out of the chip note that as always we're using CUDA acceleration for these so these are realistic times or someone actually doing a render for Youtube upload we are not artificially limiting it to just the CPU because for youtube uploads CUDA acceleration is a good thing the acceleration applies evenly leaving the CPU on its own to make up ground where it can and that's where we start to see the 3900 X pull ahead where we actually didn't expect it to originally the 4k sixty render is more intensive for this one the our 939 hundred X completes the render in nine minutes with the four point three gigahertz all core OC finishing an eight point eight minutes the ninety nine hundred K at stock settings and TDP finishes in eleven point nine minutes were ten point seven one overclock to five point one gigahertz the end result is an eighteen percent time reduction for the overclocked 3900 X versus the overclocked at ninety nine hundred K and for generational perspective although core counts are not equal the 3900 X our performs the 2700 X at 4.2 gigahertz ranked at twelve twenty eight minutes by 30% and the 1700 at three point nine gigahertz at four point eight two minutes by 37 percent time reduction this is mostly from more threads and that's obviously important here but to be fair the price is also a bit higher so the 3700 X will be an interesting comparison for this one which will be in a separate review for us before getting started we need to take a look at how frequency boosting behavior is on the 3900 X we did this for the 3600 as well and it helps to illustrate all core turbos under good thermal conditions in a blender all core workload the frequency average is about forty eighty-seven megahertz across all cores if you're wondering why the frequency isn't hitting the advertised boost of 4.6 gigahertz that's because the listed boost frequencies only apply for limited thread load scenarios which blender is not well next intentionally zoom in on the charge shifting the y-axis around to four thousand and stopping at 4200 megahertz we're also going to just dump all the lines on the chart at once even though it's not particularly legible as you can see here the Coors boost up and down bouncing between the 40 50 megahertz and 40 100 megahertz range and alternating patterns between each core the only time frequency ever exceeds 40 100 megahertz is in the pre-test idle period shadow of the tomb-raiders the last one will show same thing we're just dumping them all on the chart at the same time as shown here forty six hundred megahertz is sparingly reached under single core work loads which are uncommon in these applications except between loading for example for the rest of the gaming scenario where around 40 200 240 225 my guards sometimes up to 40 to 75 mega Hertz or 40 300 megahertz it all just depends on how many cores are actively loaded total war Warhammer to testing is up first we use the battle and the campaign benchmarks for this the campaign map benchmark is our preferred total war Warhammer to CPU test with a greater reliance on CPU power than the battle benchmark and last GPU intensive effects as predicted the distance to the 99 hundred K is wider than with the battle results stock versus stock the Intel 9900 K at 179 FPS average is 28% ahead of the 3900 X is 140 FPS average these two CPUs are mostly tied and 1% and 0.1% low values and there's no perceptible difference to frame time fluidity to the user for frame times but the 9900 K does have a meaningful improvement in average frame rate overall overclocking the 3900 actually yielded some of the best results here that we've seen on a 3000 series part so far placing it 4% ahead so all core overclocking clearly not necessarily the most worthwhile endeavor on the 3900 X the 9900 case 5.1 gigahertz OC is an even greater 30.9% ahead an average FPS versus the 3900 X at four point three gigahertz all corso OC vs. Oh see this is a worst case scenario for the 3900 X competitively it's a game that doesn't benefit from high thread counts that much as illustrated by the stock 9600 K and multiple generations of old CPUs like the 7700 K ranking pretty high on the charts and it strongly corresponds to increases frequency and performs worse with SMT enabled disabling SMT and overclocking the 4.4 gigahertz pops the 3900 X average up to 131 fps from 140 with some respectable improvement in the 1% 0.1% lows as well pushing it past any of the 36 hundreds results with that sums he on the 3600 actually outperformed the 3900 X both stock and OC partly due to thready utilization and scheduling in this title and because of how the all core frequencies behave with boosting differences between the 3600 3900 X the story is mostly the same at 1440p but with results at the upper end limited by the GPU as they were in the battle benchmarks note that the overclock to 9900 K drops from 190 FPS average out 1080p down to 174 FPS average at 1440p while the 3900 axis results are hardly affected this indicates that we are bottlenecking on a GPU over the 900k but that the 3900 X is limiting the GPU in its test the battle benchmark is up next at 1080p if the 9900 K at one so d1 FPS average is about 7.5% I had of the 3900 axes 159 FPS average stock versus stock but as we mentioned before the battle benchmark becomes GPU limited towards the upper end of the charge even at 1080p the lack of major improvement between the stock and 5.1 gigahertz overclock of the 9900 K indicates that FPS averages will max out at about 175 FPS with these settings and with the GPU we're using and we may see a wider gap in other games without this limitation this is further proven by looking at the 9600 K and 9700 k OC positions which illustrate a GPU bottleneck overclocking did a little better for the 3900 X ranking at 165 FPS average but it's still only a 3.3 percent uplift at 4.3 gigahertz all core as I am DS per core boosting and games keeps up with the best all core clocks that we can push the r5 3600 at 159 FPS average is performing function equally to the 3900 acts in this benchmark and as a nearing error for 0on percent lows although not quite as with the 3600 disabling an SMT to limit the 3900 x2 12 cores and 12 threads had a more positive impact on performance than the 24 thread all core overclock this time we also overclocked it to 4.4 gigahertz with SMT off allowing the 3900 ex to stretch to 167 FPS average to be very clear here we wouldn't recommend disabling SMT on any of these processors as it's hit and miss in games and obviously cuts production workloads in half this is more of an exercise to learn about the performance f1 2018 is next this is a DirectX 11 game and gives us a look at the ego engine the stated competitor for the 3900 X is the 9900 K but f1 at 1080p is another task for the 9900 K is far ahead about 13.5 percent stock versus stock with a 313 FBS average - the 3900 access to 76 FPS average the modest 3.6 percent improvement in the 3900 access score from a 4.3 gigahertz overclock it does close the gap somewhat jumping to 286 FPS average as the 5.1 gigahertz OC on the monday had barely any effect and it was bouncing off of GPU limits this is another title where SMT overhead hurts the 3900 axis performance although not so much that it drops to 3600 levels this time disabling SMT and overclocking to 4.4 gigahertz lands it at 302 FPS average closing much of the gap although it again advised against disabling SMT in actual use the narrower range of results at 1440p means that mean the 3900 X is effectively equal and performance to the cheaper r5 3600 whether that's stock with an overclock that's not likely due to a GPU hardware limitation elsewhere in this test either moving on to civilization 6 this game uses a turn time benchmark rather than an FPS benchmark for our charts this is useful for indicating how long it takes each of the 5 AI players to process its turn before passing to the next player if the average turn time for example is 30 seconds it would take 150 seconds after clicking n to turn to return to the players next move civilization 6 suffers from the same problems with SMT overhead that's one of the other titles do based on the 3606 core 6 thread results and the six core 12 thread results for that same processor but it's also the only game benchmark that we score by average turn time instead of FPS I mean you know GPU bottleneck in for even the high-end CPUs lower is better and every fraction of a second is significant because of the low variance between the runs and all core overclocked and 4.3 only reduce to turn time by 3.7 percent which is a fair indication of what's happening in the sieve benchmark the advertised boost clock for the 3900 X is 4.6 gigahertz so in benchmarks that allow the CPU to read single core boost clock the OC results may actually be worse than stock disable an SMT and pushing the OC to 4.4 gigahertz cut the average time further but even with this best case scenario the thirty point four second average turn time for the 3900 X allows the 9900 K to remain 2.3 percent shorter in turn times overclock in the 99th K gets it down to 28 point two seconds which is a 7% reduction versus the best case 3900 X Assassin's Creed origins is our next test as we highlighted in the r5 3600 review this game actually likes both cores and frequency with some level of balance making it one of the best balanced gaming benchmarks we use for CP reviews of all the games we tested assassin's creed gives the 3900 X the best shot at leveraging its 24 threads as with the 3600 turning SMT off worsened performance even with the moderate clock speed increases as it ideally should so we can ignore these numbers for once the 3900 X doc CPU performed at 134 FPS average with low as well paced at 93 FPS 1% and 80 for FPS 0.1% and disabling an SMT while overclocking brought down the FPS to 120 for a reduction of 8% with settings that were best-case scenarios in other games but obviously not all of them the stock 9900 K runs at 140 FPS average here and is beginning to bounce off of 28 eti limits pushing just 4.3% ahead of the stock 3900 acts as a result this is a GPU limitation and other benchmarks with less of a GPU bottleneck we've seen ranges between 13 and 30% advantage for the 9900 K when both products are overclocked for example so it all depends on how the game responds to frequency versus threads although there does seem to be a mean for name forming that about 13% that said the 3900 exit does pull ahead and 1% and 0.1% lows we occasionally have some trouble with stuttering an assassin's creed origins that affects the 0.1% lows but the 1% lowest sample and up frames that it's safe to say the 3900 acts is ahead and frame pacing here the stock 3900 also outperforms the stock 3600 by 17.2% a refreshingly straightforward advantage for the higher-end part overclocking the 3900 acts barely improved the average and it worsened the lows which is to do with the way the frequency bounces around based on boost as opposed to the all core overclock when it's locked this frame time plot will help illustrate what's happening remember that these are the raw numbers from which fps is derived and so Stan is the most accurate representation of real gaming experience and frame-to-frame fluidity is just a little harder to compartmentalize so don't use it for every game the 9900 K starts with overall frame to frame intervals lower than the 3900 X which is a positive thing for the Intel 9900 K lower is better but more consistent is best for reference sixteen point six 67 milliseconds would be 60 fps the 9900 K encounters the same amount of spikes in testing as the 3900 X as this game has a less flat line than others but its spikes are more frequently beyond 12 milliseconds this isn't bad on its own but an excursion from the mean greater than 8 to 12 milliseconds does start to become noticeable as a stutter the AMD r9 3900 X has a higher frame times overall at roughly 6 to 8 milliseconds versus intel's five to seven millisecond range average but it spikes don't hit the same higher peaks that intel does so this is why the 3900 X does better in frame pacing once we use the average bar chart that we saw earlier the 1440p results are like the 1080p results but compressed with a GPU bottleneck now becoming more apparent overclocking had no significant effect on either the 1900 k / 3900 acts at this resolution and the 3900 axes lead over the 3600 shrink-down 24.3% there's a reason we still test CPUs at 1080p even as high resolutions become popular but this chart allows us to see that even 1440p does have points where the CPU start to matter at least for the high-end GPUs as an aside take a look at the r5 3600 SMT off results or the 3900 axes for that matter see what happens to frame time performance as I am these threads are disabled with this game it actually matters in this one GTA 5 is next and is our geriatric came on the benchmark it's still a top played game on Steam and it's the x11 implementation is a traditional one so we like to include it the 9900 k leads the 3900 acts in GTA v stock ranked at 122 FPS average versus the 3900 X is one 10 FPS average that's a lead of 11% for the Intel part although Intel is still ahead it's leaders shrinking with each generation versus the 2700 X the stock CPU the results of 90 3.6 FPS average allowed Intel to manage a lead of 31% so this new climb is significant and most of it is to do with IPC and frequency although in some games the thread count actually matters just not particularly this one the 9900 K improves more with an overclock than the 3900 acted as predictably moving to 131 FPS average versus 113 FPS average of the overclocked 3900 X that's because the 3900 X is already nearing its maximum performance potential without extreme cooling with the CPUs we've tested so far GTA doesn't scale up and performs much with thread counts higher than four so the 3900 X is 5.4 percent gain an average FPS versus the 3600 is mostly down to higher clocks the 12 core 12 thread a 4.4 gigahertz overclock pushed the average FPS up to 118 point five past 113 for the 4.3 gigahertz 12 core 24 thread OC seemingly completely unaffected by the loss of threads as discussed in the 3,600 review the GTA results we've gathered so far are almost identical at 1080p and 1440p thanks to the CPU heavy settings we use in the game but we wanted to show this chart anyway just to give you some extra data shout over the tomb Raider's next using a modified crystal engine and DirectX 12 then I heard K performs at 170 FPS average one stock placing it as 18.6% past the 3900 axis stock 143 FPS average result and that's with Intel CPU becoming GPU limited at the top of the chart despite the GPU limit we have never tested below 1080p in Jen's history actually for desktop CPU benchmarks as we don't believe it to be an accurate representation of real use I think the last time we used it might have been for a fraps versus shadowplay recording comparison but not for gaming benchmarks so 1080p is and has been our floor if it bottlenecks here we just make a note and move on shadow of the Tomb Raider seems to only benefit from SMT up to a point although enabling SMT tank the 3900 access performance as it does in some other benchmarks overclocking the 3900 acts only got its average FPS score up 2% but disabling SMT in combination with a slightly higher overclock raised by 6.6 percent even at stock though the 3900 X outperforms any of the 36 hundreds results hitman 2 is our last game benchmark a pattern is emerging with him and to where the 8 core 16 thread 900k and 12 core 24 thread 3900 acts both benefit from disabling multi-threading while at least 6 core 12 thread 3600 part performs better with SMT on and we can highlight the SMT off result to illustrate that it's not as simple as just saying more than 12 threads is bad though because the 8 core 16 thread 1700 outperforms the 6 core 12 thread 1600 both stock and overclocked what we can say for now is that hitman to strongly benefits from more threads up to a point and begins to scale negatively past that point both for AMD and Intel in any case the stock 3900 act still outperforms any of the 36 hundreds results and the stock 9900 K surpasses the stock 3900 acts by 12.9% running at 136 FPS average vs. 121 FPS average when both the 9900 K and 3900 ax are stock disabling hyper threading and seen such uplift illustrates that we are not GPU bottlenecks in this title so it joins the list of titles usable to illustrate the full range of scaling between the two stock CPUs overclocking offers a slight improvement for each but nothing major for either CPU we showed power testing an hour our 5 3600 review but we'll quickly go over it again here and a blender workload at 100% saturation across all threads the AMD r9 3900 X stock CPU ends up measuring at 148 watched on the EPS 12 volt cables which is far away from the TDP designation of 105 watts it's not even close technically TDP is not one-to-one with power consumption but it's close and being at 148 to 105 is a big difference we need to investigate these BIOS versions and see what's going on with the different platforms for what it's worth the platform does operate within TDP bounds during gaming workloads but not in all core loads like blender we think this may have to do with the stock Auto voltages on some early motherboards but we're not positive just yet there isn't an mze equivalent option that we found and PBO is disabled than all of our review content unless otherwise stated because it's not within spec for hitman two just as a quick measurement we see power consumption more reasonably situated at seventy six watts stock you'll notice that our overclock here consumes less power than the stock CPU by a few watts that's because we were able to tune voltage down below what was Auto selected in this workload more power and thermal discussion after we survive the launch day this was already too much to do at the time we had so we'll table the rest of it for now there's a lot more testing we still want to do for example we still need to dig into the thermals we need to look into the power some more we really want to look into streaming some more for our standalone streaming benchmarks and a couple of other ideas we have for scaley and testing so it's just there were so many products that watch this time that we decided to review about four five of them five of them and in the two launch day period and then we'll just circle back for the other stuff later so hand tight with us for a couple days whole we go through all the follow-up testing but there's a lot more to do yet so power consumptions our only big remaining question mark at this point other than some additional testing you want to do elsewhere and we won't spoil too much of that now but as stated it's it's been like 16 to 20 hours a day for a week and all of this is the result multiple videos multiple 30-minute lon really detailed testing and reviews so we're not going to dig into the power anymore right now we need to because it seems to be a bit high and the Intel platform if you disable MCE you get it down to almost exactly in tell us to find TDP number and again TDP is not one-to-one but it's really close because ultimately it all turns into into heat anyway and into power but yeah so we need to look into that more and the power consumption of the chip could relate to auto voltages they are a bit high and that does lead into another interesting topic which is that on the positive side our r9 3900 ex seems to be we're new to this platform of course everyone is but it seems to be a good overclocker so the voltage was about 1.3 Ford's 1.35 for that one to hold all court at four point three we could not get to four point four on anything until we disabled SMT on the 3900 X and then we can do four point four gigahertz that is so four point three seems to be about the max for all the processors we've tested so far doesn't matter the voltage up until you get to extreme cooling then it'll matter of course and the higher-end CPUs have performed better than the low end CPU we test at the 3600 sample size of one of each so it doesn't really count but gives you an idea in theory the higher NCVS should be been better anyway because there they have more things enabled so that would make sense anyway the 3900 X is a powerhouse and applications that are thread and cash bound and calling it a competitor to the 9900 K and these tasks isn't hyperbole on a and these part for production workloads it's often better as a CPU by a considerable margin sometimes north of 50% once or twice north of 60% much of that is due to the 3900 X's 24 threads though and AMD has been making thread ripper chips with even higher core counts for some time now some of which are now on Steve sales and dropping below the nine adder decay in price the 3900 axon needs to strike a balance between gaming and production performance to be something new and to compete with the 9900 K in a way that thread ripper doesn't already the results are mixed the 9900 K is just better in non thread bound applications and and II can't force game developers to just flip a switch and make everything thread focused overnight which is something we've been talking about for a few years now the 3900 act continues the legacy of the previous Rison processors where we can confidently recommend it as a mixed workload cpu and we can confidently recommend the r5 period so that's a good place with r5 to be but the 3900 acts were more in the range of it's it depends it's not as simple as the r5 is with 3900 X we can recommend it in mixed workloads we can recommend it in sort of budget and budget and quotes tier production workloads and that's if you don't want to just buy a used AMD processor from the last couple of generations because you can get these older ish processors for significantly discounted this one 100 is regularly 200 bucks now now the 3900 X is much more powerful than that so it's not as simple as the 2700 X versus the 3700 X but it's a consideration if you're honest a much stricter budget and thread Ripper as noted it has also routinely been sale first-gen thread if it does have quite a few deficiencies in some areas that the 3900 X solves but if you only need threads it's worth considering for a cheaper solution that's not to take away from the 3900 X though it does well in these workloads gaming so for gaming workloads the recommendation if you're pure gaming high refresh don't care about anything else never work with any other applications maybe work with premier now and then it's close enough to the 3900 X although and he does notably outperform Intel in our test and premier for once and a head-to-head but if you're mostly gaming and maybe premier now and then or Photoshop then Intel would get the recommendation still the 3900 X does isn't it's not a clean sweep it's a good processor but it's not good at everything all the time and that's fine it doesn't need to be so we would recommend the 3600 almost always at its price category specifically at about $200 except for against the older processors that you get for cheap we're only talking modern stuff that's still being made so where we would recommend that basically everywhere the 3900 acts we reserve recommendations for mixed workload user who genuinely actually does some production type of tasks on the side maybe you're a hobbyist video editor maybe you do 3d animation and modeling and you render with the CPU things like that then or transcoding a compression especially decompression stuff like that if you do any of that mixed with gaming it's a good choice if you only game it's still gonna be the 99er K from us so that's it for this one thank you for watching subscribe for more definitely go watch the r5 3600 review we have the 5700 XT review going up as well or up already by the time this goes up probably and a teardown of this coming up shortly 3700 review is next check back go to store like a mis-text net to pick up one of our toolkits to support us directly or you can go to patreon.com/scishow cameras an axis to donate some sleep I think that I think that works I believe there's a sleep bank I can withdraw from thanks for watching I'll see you all next time