Gadgetory

Red River brings MD closer to where its risin architecting started with server class epic chips that later were worked down into consumer lines and so thread rivers should more fully expose and these multi die architecture and its benefits and performance but CPU to ship at $800 to the 1920 X and $1,000 the 1950 X both of which are on the bench review today and they also both pose a serious threat to intel's newly launched skylake x high-end lineup we're focusing on game streaming power thermals blender and premiere performance and other items along the way with a lower weight splash of gaming just for perspective before getting to that this video is brought to you by synergy the software that lets you share a keyboard and mouse between multiple systems if you have limited desk space and multiple computers to command synergy removes the need for separate peripherals or a KVM and works as over the network software use our link below to get 50% off the basic or pro version we've a lot to get through today so there will be no fluff at all you can click the link in the description below for the article if you want background information on thread Ripper but for now we're just focusing on getting to the benchmarks from the video before going through those a few items of note on thread referee overall overclocking is not going to be the most impressive thing you've ever seen with thread rippers there are reasons for that we'll explain those but in some of the charts you will see lower performance with the all core overclock than with the out of box performance the main reason for this is XFR or extended frequency range and with thread ripper Andy has a two hundred megahertz XFR over the advertised boost so if advertised boost is 4.0 it goes up to 4.2 and it can do that on four cores this means that applications which use fewer cores like games and a lot of instances will do better stock because they can use that XFR to boost to 4.2 gigahertz or something like that plus 200 rather than an all core boost a 4.0 where half or more of the cores are never even touched to begin with so if you see it results that looked higher with the out of box config that's the reason we're still showing 4.0 gigahertz overclock and some of the charts though just because I'm sure people are curious but we were not stable all core over 4.0 without just absurd voltages so that's the first note the next note is there are two memory modes with Red River distributed and local these are also known as uniform memory access and non-uniform memory access or uma and Numa distributed is the default mode and tends to work best for content creation tasks as uma isn't restricted to one die and allows the scheduler to do whatever want Numa tends to be better in some games what restricts workloads to one die and the memory attached to that die there are creative applications that minimize thread sharing and synchronization between cores so they're not that latency intensive and that's why you would see better performance with the other mode as opposed to games which do actually care a lot about latency and synchronization in most cases because you might be synchronizing things like the physics and game logic threads or AI threads or things like that so Numa tends to be better in these use cases we do have tests of these in a V but they'll be coming out separately however we have tested for today the AMD bundled game mode which disables half of the cores and switches the memory access mode to local let's start with streaming benchmarks while gaming these are two tasks that are intensive and perform simultaneously so stream game video playback from the stream and then we have other benchmarks that are dual stream or streaming while recording and the reason that those came out is because the single stream benchmarks while gaming were simply not intensive enough to show any difference whatsoever between the thread Ripper and the 3900 X there it was just too easy for them and so we started introducing other tests and one of those was the dual streaming with dota 2 streaming to twitch and YouTube simultaneously like we do the 7900 X review the other one was a streaming csgo while recording that locally and they were also both really not that intensive when you've got 32 threads to work with but we'll go through the results anyway we're focusing on the more intensive tests here since the single stream output tests were just trivial for both CPUs to handle to be fair they're also trivial for an r7 1700 to handle as previously so it doesn't really teach us anything new that we can focus on here first up is our simultaneous streaming of dota 2 to youtube and twitch which we ran with the h.264 faster preset the output is 1080p 60 with YouTube's bitrate at 10 megabits per second and Twitch's bitrate at 4 megabits per second these outputs cover all the bases with the most common stream setups when we pulled our audience as seen here you end with this tremendous load on the CPU is we drop effectively zero percent of the frames on all tested devices we'll look at the 1920 X in the very near future that would be fair to assume that that would perform about the same given where the ten core Intel and 16 core AMD parts land that is to say probably dropping 0% to frame as far as the viewers experience is concerned both of these CPUs output effectively a perfect 60 FPS to 2 streaming services simultaneously you would not be able to tell the difference as a viewer and that's a good thing they're both good at this task here's a look at the FPS output to the player now that we know what the viewer experience is like when dual streaming the 1950 X manages 79 FPS average to the player though dips down somewhat in frame times toward the 1% and 0.1% slow its frame times the 7900 acts performs well ahead here but we also know that dota 2 does tend to favor intel parts we were previously benchmarking dirt rally for this test as well but unfortunately it doesn't work here because thread Ripper just doesn't work with dirt rally without disabling out the course that's more on the game development side because it just doesn't know what the hell it's looking at when it's East 32 threads so it freaks out and crashes the 7900 axe is doing well though but regardless of that thread Ripper is still hanging on and could be tuned for better performance is perfectly acceptable here considering that the stream output is delivering 100% of the frames in a game that tends to be against and these favor to begin with so things are looking good to start with for thread Ripper ultimately if you wanted extreme frame rates and consistent frame times anyway it'd probably just be best to build a separate capture box off of the work both still do well here though these 7900 is outperforming the 1950 accent player side same rates we have to give Intel major credit here there is 79 stocke CPU when streaming is performing more or less better than Andy is 1950 X when it's not streaming particularly if we're assigning more weight to the lower end frame time performance so the 7900 X is doing very well in this particular benchmark and with both at $1,000 it's probably the better buy if you're playing csgo specifically but this is not the complete picture because again we're testing dota2 and csgo right now both of which come from the same company and both of which probably show the same favor in terms of optimization so for a more complete picture what we'll have to do is revisit this once we get to the 1920 X streaming bench works and try to add a couple more games so keep an eye out for that strictly for this game intel's doing well just keep in mind that there is a lot more to the complete picture as we learn more about thread Ripper as advantages but we'll look at power metrics in a moment to see how that battle shapes up and see if the 1950 X can claw back any advantages outside of FPS during the csgo recording and streaming tests with medium encoding speed which we had to step down from faster because it faster just simply wasn't intensive enough and neither was medium for the record the I 970 900 X was consuming 193 watts of the EPS 12-volt rails when under average peak conditions the thread Ripper 1950 acts consumed about 134 to 140 watts average peak power this is a significant reduction from what the 7900 X uses that's a good thing the 7900 X does outperform the 1950 action player side frame rate with this game but the 1950 X outperforms it in power consumption competing with a 27% reduction over the 7900 X with equivalent viewer side output just the player side that changes let's plot temperatures on the right axis now in the same charge during this synchronized csgo workload we see the 700 X ramping up to an average peak core temperature of 60 to see the 1950 X is around 57 c on average ambient is accounted for in these measurements as well not a huge difference particularly when considering that t.j.maxx is different for each CPU but that's what we're looking at for temperatures during the streams we could do a lot more with this but most of the actually all of the other that we ran we're just so much lower intensity that it's they're both good they both deliver honor to some of the frames there's no point in looking at the data it would be like it's just it's a pointless benchmark we ran the test anyway because normally those are actually pretty hard for CPUs to handle but when you're looking at a 10 core twenty thread part and a sixteen core 32 thread BART they're irrelevant so that's what we have for now the numbers here indicate that if you wanted to with thread or especially you could step down from the medium preset that we use to something even slower which means a higher quality output technically although you start entering placebo territory it could handle it it could keep up if you wanted to do that but for now we're using medium it all looks good and the CPU is perform effectively identically on the stream viewer side output one other item of note here you play around a lot with affinities and priorities with streaming we know this we've done it a bit in the past didn't do it here because it just wasn't necessary so there's something to be said there if you would rather not fuss with any of it then certainly one of these two CPUs would be a good solution to that assuming you don't want a separate box because now you never touch the affinities or priorities unless you run into a really odd use case like their rally or codemasters games which just don't work but that's not a big deal you turn on game mode and rise a master and it works fine at that point though you lose half your cores but that's a that's pretty limited sample size from what we've seen so far and it should go away in the future but that's what we're looking at if you are ok with playing around with things then you can buy the cheaper CPUs assuming this is the only thing you want to do with them anyway and just while we're here one quick note on VR it'll be fine vr works well on the 1700 it works well in the 7700 K they are imperceptibly different and you won't see any benefit from going to these higher core account part anyway moving on to power next this is a good time to segue into those we're measuring at the EPS 12 volt rail for power consumption this is not aggregate wall draw like we used to do from the power supply directly to the CPU we also have high performance mode set for these measurements to keep that in mind idle via thread Ripper CPU the 1920 X and a 1950 X we're both consuming about 8.6 watts which is within measurement error of the r3 and our seven CPUs accounted for especially by the motherboard change overclocking gets us up to 22 watts to 30 watts draw Idol when using the high-performance power plan comparatively we measured the 7900 X at 49 watts idle in the same power plan though both CPUs can draw less when using a more conservative performance mode and blender the thread Ripper CPUs sit within a variance range of one another again both that around 145 Watts on the EPS 12 volt cables this puts our thread refer CPUs consuming about 10 percent more power at stock than the overclocked 17 100 the 7900 x stock CB runs the same blender test at 171 watts but it's overclocked experience at 2:24 third row remains remarkably efficient and it's out of box States but starts guzzling power when overclocked on higher voltages to sustain an all core Oh see we're at 274 watts when overclocked on the 1950 X and to 12 on the 1920 X and as you'll see in our benchmarks for blender it's not worth it total war Warhammer Paul's 75 watts when the 1950 axis is at its stock configuration using creative mode and 53 when the 1920 X's in its stock configuration note that these are with Auto voltages on the zenith the stocks are 900 x consumes 93 watts for the same task with overclocking pushing both flagships to 100 watt territory for prime95 Cinebench fire strike and others check the article linked below here's a quick look at thermals over time in a prime95 28.5 L FFT torture workload note that this chart shows spikes as the test iterates between larger FFT sizes so you'll see a thermal torture scenario enumerated as power and fft cycles iterate the assembly line hard ax runs warmest here using our X 62 and set a delta T over a meter around 50 C peak average as FFT size progresses we see that the 1950 ice begins to heat up more and reach what is more or less a steady-state at 44 C delta T over ambient keep in mind that distance from TJ Max is different on AMD and Intel processors so the significance of this temperature will vary between them until maintains a distance from TJ Maxx of about 25 C in this test we think that AMD is t.j.maxx on threader is about 85 to 90 see that we haven't been able to confirm this directly that assumption is based on observations of when thread Ripper throttles in one of the tests if that's the case the summe 990 C t.j.maxx with a high end and D maintains a distance of about 22 to 25 C also not so different in this case anyway more thermal discussion in the article getting into production workloads next we start with our in-house blender animation for CV workload benchmarking the threader 50 views don't even need a highlight they're all on top that will highlight them anyway the 1950 X completes our render in 15 minutes a remarkable speed considering we were only recently impressed with the our 779 it's 27.6 minute completion time this performance places the 1950 X ahead of the $1000 700 X CPU on overclock to 4.5 gigahertz even the $800 1920 X manages to keep pace with this 7900 X a line item coming in $200 cheaper and about same rendered performance it's clear that the 1920 X carries the trend of lower skew AMD CPUs offering the best value that said the 1950 X does reduce render time by a still noteworthy 19% over the 1920 X it's not a bad price for such a reduction assuming you're doing something that will actually leverage it otherwise the 1920 X is looking impressive from this testing and serves a good value proposition off to a good start on this one Adobe Premiere is next this test uses one of GM's own project files as a benchmark for a real workload and positions the 1950 X again at the top of the chart well aside from the CUDA accelerated workers anyway premier still benefits from boosted CPU performance for certain types of renders though it is not the most optimized application you'll ever use regardless and I think that the X completes the render in 41 minutes with the 1920 X finishing the render 14% slower at 46.6 minutes just behind this is the 7900 xcp at 54 minutes it was overclocked variant does clawback a good amount of ground for perspective the r7 1700 X overclocked to 3.9 gigahertz finishes the rendered 62 minutes showing that the similarly clocked 1920 X provides a 25% time reduction from its extra cores pov-ray multi-threaded rendering is next and post the 1950 X completing the workload in 46 seconds followed by the 7900 X 4.5 gigahertz OCC VF 51 seconds surrender time increases 11% here and is followed next by the stock 1920 X CPU for perspective our highest-scoring r7 CPU completes the words in 76 seconds multi-threaded for render time increase of 64% from the 19th of the Xbox CPU here's the single threaded version of this test here it's clear as day that Intel still holds a significant lead in single threaded performance the 1700 X is impressive in this regard and completes the 1 threaded worth load in 565 seconds for a 22% time reduction from the 1950s stock CPU that's a big jump in Intel's favor and also coincides with Intel's still stronger performance and other single threaded intensive applications but that's also not news don't buy the rise in architecture CPUs if you want the strongest possible single threaded performance or highest IPC for most folks in the enthusiast content creation audience though thread count holds a lot of relevance and so rising architecture is completely valid and should be considered in those use cases moving on to games briefly here because it's really not the focus we're more interested in the game streaming aspect of it if you skip to the part where we talked about how overclock and core is impact games now is a good time to go back and watch that before confusing comments are posted that's discussed right after the ad spot walk through a handful of game benchmarks next again as this is an h EDT platform we're more interested in the X through 99 configurations from the standpoint of gaming while streaming but we'll just look at gaming anyway it's just not weighted as much for a TD T and the thread rate for CPUs benefit in total war Warhammer from game mode which requires a reboot and disables half the cores then switches the memory mode to local from distributed stock with the distributed memory mode and 16 cores the 1950 act operates an average of around 127 FPS with lows at 82 and 42.8 rebooting with local memory access and half the cores enabled we operate instead at a significantly boosted 146 FPS average with lows also significantly boosted to 109 1% and 68 fps your 1% lows that's about a 20 FPS gain for each category the average FPS bump is about 15% with game mode in this particular title and is well worth it here's where it gets interesting notice that the all core 4 gigahertz overclock which overrides xfr and other features is for me noticeably worse than the stock configuration this is partly because of acts of art which occasionally boosts threads to frequencies between 4.1 and 4.2 and he has done well with xfr here that's the story in the takeaway the range has expanded significantly you now with plus 200 megahertz are abused and it functions well enough that it actually be worse the overclock in some games like this one as for the 1920 X for seeing competitive performance the 1950 X largely because just based on thoughts right now a game doesn't understand what a 16 core CPU is or what to do with it compared to levy 7900 x stock CPU operates a 168 FPS average with 70 FPS 0.1% low values and so a significantly boosted over the 19 to the X in this particular workload 1440p won't change the scaling much unless running something very graphics intensive or with a lower end GPU like lower end relatively like a 1080 Nandi eye but 4k will 4k will be the great equalizer here if gaming on higher resolutions the performance gap somewhat minimizes that it becomes a GPU bottleneck at that point o games that don't understand the threads will still need gaming mode to resolve frame latency issues even at higher resolutions as usual though none of these HEV TCPS are good value for gaming you're way better off buying something like an R 5 or an i7 7700 K for something like this but those won't handle workstation tasks nearly as well GTA 5 is our next title and posts an average FPS of 116 for the 1950s stock CPU with 1% letters at 62 and 0.1% calculating out to 32 FPS these low in frame time performance metrics are pretty damn bad but that's because the game doesn't know again what to do with the CPU the frame times are inconsistent enough that we're experiencing visual stuttering to a point where it's really not worth having the extra cores enabled fortunately enabling the gaming mode toggle and rise and master helps resolve this issue game mode brings up the worst 0.1% frame times to about 50 FPS effectively eliminating the stuttering or hitching average is roughly the same the 1920 act stock review isn't affected in the same way given us a lower core count and so enjoys the benefit of a 50 fps 0.1% low out of the box with averages proportionately higher overall we think part of this is because CPUs are the same core account as the 90 20 X have already existed so it's not unreasonable to expect some level of optimization of the code for them for reference the 7900 x operates at 145 FPS average and this boosts FPS about 26% of the 19 to the X stock CPU ashes escalation post the 1950 exit 47 FPS will close at 33 31 there's one of the few tests where we consistently saw higher thread counts utilized and is also one of the few tests where we see a performance loss by enabling gaming mode and they acknowledge this and noted the loss as expected is simply because ashes usually uses those threads and distributed memory mode this positions the 1950 X about 4.5 percent behind the 7-yard x not bad given previous results 1920 x isn't far behind and actually does lose to the 1950 X in this particular title unlike some of the others which is again because the extra cores are actually utilized rather than just confusing the game natural last flights another game where we see the 1920 X behind the 1950 X and see those cores utilize somewhat but you can check the article below for that game and others we have a lot more to do with Red River this is one of the more interesting products we've worked on the last few months it has a whole lot of options for testing and for use cases as a user so we'll be iterating on this especially in the thermal department but for now one thing to get out of the way first like these 7900 X just like we said for the 7900 X actually just like we said for the 1800 X don't buy it for gaming only if you're only gaming just to stop and go buy something that's $250 something like that because these aren't for you so that stated what these do well on the gaming side are things that would be more enthusiasts type tasks like recording your games while playing them and encoding that on the CPU rather than rely on on env encoder or something like that if you prefer not to use shadowplay with enemy encoder other than that we're looking at benefits primarily in production workloads blender is really impressive for thread Ripper it's it's well ahead the 1950 X actually posts a good gain over the 1920 X so it's not just another pointless step to a processor that is really hard to justify existence of in this case the 1950 X Julie has purpose it's doing well it's 20 or so percent ahead the 1920 X but the 1920 X is still really damn good if $800 which looking at the this market and the segment is also pretty damn good and very competitive and that makes these two CPUs some of the more exciting ones that we've looked at recently in the game streaming side while they do not offer a noteworthy advantage over the 1700 X at least with the type of streaming we tested they do offer a good value proposition if people look at the 1920 X and we need to do more with that one on streaming but it's fair to say where it will fall that's said to subscribe and check that when we post that information strictly from the standpoint of where these CPUs perform best blender performance its chart-topping premiere performance aside from CUDA acceleration its chart-topping and you can still use that we finally found some applications that will use the extra cores without just CUDA accelerating beyond the the usefulness of those extra cores and then for other things like streaming it's good it's just not it's not special but it's good and not special strictly because the 70 I know tax also is fine for those applications because turns out streaming while gaming although intensive is still not enough to these CPUs to really start crumbling you need to do other stuff in the background maybe if you did streaming while gaming and while rendering something in blender then you start to see differences we also start departing from real-world use cases at that point but if that's you sort of looks good for it we didn't talk about gaming performance too much in this video it wasn't really the focus we're more interested in streaming wild gaming or productivity but here's the thing if you are doing a lot of gaming it is your primary task be careful because it may be the case that something like an r7 1700 or an overclocked variant is a better buy for you because this CPU the 90s at the ice tiny 20x they do have problems with games it may be the case that the games don't launch a lot of the ones we've tested we just see lower performance scaling overall than with Rison 7 for example because the games don't understand what they're looking at and something like Rison 7 is more acceptable to them things are built for them a little bit better so be aware of that that's not to say that the cpu is bad in any way it's just that if you are doing something like for example streaming to a single service so it's a low workload stream relatively you're streaming to one service and you're playing games Rison 7 is going to be a better buy for you if you're streaming to multiple services capturing live or using the more intensive encoding options then thread Ripper should be a bigger consideration but until that point rising 7 makes more sense unless you're doing other things like blender or premiere work or any kind of CPU crunch where the CPU will be pegged to a hundred percent those are the cases where threaded four makes sense but if you are primarily a gamer even one who streams just go to the article linked below look at the gaming results and consider that maybe horizon 7 see if you would be a better buy for you because the performance can actually be superior due to the optimizations and other issues with thread rippers unique features that make it so good at production so all that said the 700x is still a good CPU it's gotten a bad rap because x-29 was frankly rushed until pulled and launched quite a lot and that hurt their launch it's still a good CPU and it's just being challenged now so thread Ripper is a good launch we can give to AMD that they really ironed out the difficulties with Rison wouldn't rise and launched it was a complete mess behind the scenes leading up to the review the communication internally at Andy was bad the communication with vendors from we heard it was bad and the motherboard support at launch was mixed now with that several months behind us they've worked to really fix all that so thread Ripper has rolled in all of these fixes and all these really painful learning points for AMD into a platform that finally is actually pretty stable so we obviously haven't tested everything possible for thread Ripper but from what we've tested with this platform and the CPUs we have we really didn't run on to any stability issues that would cause me to say I'm afraid of you this in a real environment whereas previously there's like okay these these are kind of pretty good in some use cases but the memory concerns are a little worrying the crashing or whatever is a little worrying things like that they've all been pretty much ironed out now as you've seen in our coverage of the last few months so a thread Ripper we can handle II give the winning award I suppose over the 700 X and a lot of those applications and it's a good buy that said not everyone should buy it just remember like what you're doing with the PC and advise something appropriate for that the r5 1600 X is good and the cheaper and the i7 7700 K is good in the higher end of gaming otherwise look into this one thanks for watching as always you can subscribe for more patreon.com/scishow stuff without directly click the link below for the article or go to gamers nice and squarespace.com to pick up a shirt like this one as always I'll see you all next time you