Gadgetory

I've been received the ire of the community for its product stack until today sets forth in an attempt to validate the challenged existence of its X 299 platform new skylake X and new KB wake X CPUs we're starting with the 10 core 20 thread core I 970 900 X which is Intel's new $1000 CPU that takes place of the one $1,700 I 769 50 x10 core CPU this is a significant price drop over time but Intel is no longer competing with only itself skylake X's precipitous launch lands between Rison and thread Ripper due out in early August and today we're looking at game streaming VR benchmarks premier blender and other benchmarks of the new AI 970 900 X before that this is brought to you by Corsairs vengeance RGB memory which uses binned ICS for higher overclocks and uses the sm bus to write its colors rather than relying on cables learn more at the link in the description below let's start by revisiting some of our earlier 7900 X discussions from when we deleted the CPI Computex first of all that RFID chip in the corner we got some clarification on that that shift is actually not capable of storing data because it's sm bus is not physically connected so it could do something if it were connected but on the HED TCP use it is not and the reason that RFID chip is there at all on the package is because Intel is using the same package for its server versions of these CPUs as it is for the h EDT versions of these CPUs and that just means that it's kind of a leftover because again same package for both types of products but it's not physically connected so that's good in some ways though it could potentially provide functionality could also provide things like backdoors and we don't have to worry about that at all because you just can't read or write to or from the RFID chip on the package but that also bleeds into the next point which is the really interesting substrate design of the 7900 X when we deleted the CPU we showed that the Seaview almost resembles an LGA 1150 x substrate atop the 2066 substrate intel couldn't answer why the design is like this citing quote manufacturing reasons which is the same reason they cite really for anything like use of thermal paste rather than solder and our present hypothesis having not seen the server CPU is that the server CPUs might use it more of the lower substrate and ditch the upper substrate part the protrusion but Intel is keeping the same package for everything really not sure if there is a legitimate technical reason for this approach at this time or if it's more of a cost savings in as Intel just wouldn't tell us as for the CPUs themselves our current understanding is that the 14 core 16 core and 18 core CPUs will be based on the same silicon from a 20 core sieve you with some of them disabled or removed we're not sure 100% about the lower-end skews like the 7900 X which is a 10 core part but we do know some about the die size because of that deleting process earlier so the 79 100x measures in at roughly 334 millimeters squared some rounding involved and for reference the previous generation 10 core 224 core parts range from 246 millimeters squared up to 456 millimeters squared in the enterprise division respectively but anyways the substrate design is genuinely interesting hopefully we'll learn more one day at some point we do know it uses thermal pay it's probably Dow Corning and no solder for this one even though it is a larger die we'll look into the thermals in a completely separate video because there's frankly enough for today we're not going to really spend time going over a brochure with specs on it that's not really the point of these if you want that you can check the table on the website the 7900 x is under test today and is used with the acs prime deluxe x 299 motherboard for full testing methodology again check the article link in the description below we're focusing on game streaming benchmarks including single and dual stream output simultaneously be our benchmarks to validate intel's own marketing language as we figure a few others will bother with VR benchmarks and then dive into the production workloads we're saving some stuff for later like thermals and power because we're up against deadlines overclocking will also be discussed in more depth in the article since overclocking results we're still rolling in as we filmed this video starting with streaming benchmarks this is the first time we've ever done this type of benchmarking for actual game streaming via OBS and so it's exciting but there's also still a lot to develop methodologically and a lot to develop in presentation method there's a ton of data is not necessarily easy to figure out how to display it so there's there's more to come later but for now we have a baseline we're using two main tests one is exporting to a single stream just twitch via OBS at 6 megabits per second 1080p 60fps with h.264 encoding and then second we're streaming to both twitch and YouTube simultaneously so you've got dual stream output from the same box that's playing the game and that's done with the same settings for twitch but with 10 megabits per second for YouTube rather than 6 because it supports a higher bitrate and finally we do have an extra FPS baseline benchmark without streaming and another benchmark with envy encoder or the GPU accelerated encoding rather than CPU so we can get a baseline for everything else and all the other main options in the system in addition to the FPS numbers we're using login to measure what we're calling delay frames and drop frames another term that we use for VR benchmarking for these we've got usual frame time and frame rates metrics for the host systems game output to the player and then the drop frames and delay frames are used for the viewers so that you get an understanding of what the actual quality is going out the lanes happen when the CPU can't complete the encode in time to hit it's roughly sixteen point six seven millisecond window before the next refresh and so this tends to be indicative of more of a GPU limitation we can run into these with high-end CPUs if we are run into a GPU limit because the CPU can keep up fully a drop frame however is indicative of in this scenario a CPU encoding issue where the CPU is actually just skipping the frame or dropping it entirely so you end up with sort of almost micro stutter in the absolute worst case scenario if you're dropping 50 percent or more of your frames and that is just because the CP is getting overwhelmed which will certainly happen at some point as you pile on more simultaneous encoding streams because it see if you can't keep up with the workload and we'll talk about more that's going through it but we're starting with the r7 1800 X on the I 970 900 X of course both in stock configuration both with 32 hundred megahertz memory and full testing methods again on the new benchmark check the article below it'll answer all the stuff in more detail let's start with a single stream first as it's the most common use case streaming this which only we saw no drop frames on the 7900 X and no drop frames on the 1,800 X both are adequate performers with delay frames not noteworthy or perceptibly different between the two here's a look at the differences in terms of frame rate the I 970 900 ex post an in-game frame rate of 104 FPS average 79 fps 1% lows and 31 fps 0.1% lows with the r7 1800 X sustaining 96 FPS average 68 fps 1% low and 27 FPS 0.1% mode let's put a chart on the screen below this one this charge now is that you can see both of them shows the performance when there's no streaming at all with just the game being tested standalone in its same windowed configuration because again it's a stream benchmark and we're doing a/b testing we're going from 112 FPS average to 104 on the eye 9 7 900 X when compared to no streaming and with the 1800 X we're going from 109 FPS average to 96 FPS with the twitch stream going the frame time performance is significantly improved when we stop streaming as you can see the low numbers and justifies why you would still want a separate capture machine for really competitive gaming or games where frame time variability can affect your ability to play competitively moving on now to our next chart we're looking at streaming performance when going to both twitch and YouTube simultaneously there's a lot of data to process here so we're just going to start with the chart of twitch performance from streaming to both services and then look at YouTube as well these tests are meant to illustrate performance when outputting to destinations like when the user has an audience on both platforms and it's becoming increasingly popular and it's traditionally donald's every capture machine to alleviate load on the gaming machine so we're going to see if I still needed here here's the chart performance the twitch during this dual stream benchmark post 0.13% drop frames on the i9 so 900 X averaged across multiple test passes of the same duration which is largely imperceptible to both the viewer and the player delay frames are 5.0 3% which we think but we're still learning as this is new testing is because the CPU is performing well enough to bump into GPU limits the 1800 X has a much harder time with this workload and these 1800 X is driving 54% of it's frames twitch which results in something of slide show for playback for comparison and the encoder is still choppy for about 20% of its playback on the 10 to 80 with the 700 X even with two env encoder engines running so generally speaking the CPU encoding is still going to be preferable when you can get away with it that's clearly not always the case adding the YouTube chart to the page now the I 9 7 900 X sticks to well below one percent of total frames dropped with GPU frame lags still around the same level and the EC is about 78 percent of its frames dropped here and struggles with the h.264 profile and dual stream outputs to make things easier to understand we're putting the actual stream footage that went to youtube on the screen now showing the 7900 X next to the 1,800 X again remember that while this was going on we're also exploring to twitch so there are two encoders going on simultaneously one for each stream the extra four threads really help Intel in this specific use case but just for sake of understanding performance here's the screen shot of utilization on the rising system and the Intel system 1 under this intense workload and again this is h.264 with the encoder set to faster rather than very fast so you could lower your quality and possibly still sustain it but it would be a bit rough on the rise in CPU as for fps this is what it looks like the Stax chart shows an average FPS of all three tests we're down to 75 FPS average on the 1800 x still fine if it weren't for the stream struggling anyway and 89 FPS average on these 7900 X now we're faced with the challenge of analyzing what this means the first big disparity is in frame time so the average FPS on both AMD and Intel on this very specific test looks just fine when you're streaming really to any of those two scenarios what doesn't show up in averages is what we see in frame x point one percent lows in this case or we do have a frame time chart as well and that is the case that the cpus get over 70 FPS and 0.1 percent lows and our charts when there is no stream going on and then they fall below 30 or around 30 when streaming is happening which is not really the best case scenario especially if you're trying to stream something like csgo that will get you killed but for a lot of other games that might be just fine so it's something where it's going to vary on a game by game basis whether that's accessible to you depends on what kind of content you're working with anyway this frame time chart shows the dual streams versus no stream on the i9 7900 X clearly illustrating the improved frame time consistency on the benchmark without the streams running that's where those 0.1 percent load numbers are derived high variability and frame the frame latency is what we're seeing here with all that detailed a few notes the Intel I 970 900 X is an impressive CPU for this dual streaming use case scenario certainly if you must stream to two sources simultaneously and if you cannot build a secondary capture machine then this isn't a bad solution in fact a $1000 CPU the 7900 X may actually be cheaper than building a dedicated capture machine in addition to the gaming machine but you still lose performance in frame times so for people who are ultra competitive we'd still suggest building a secondary capture machine if you have the space and the budget to do it if that's absolutely out of the question then the CPU gets you closer but still outputting to just one stream is the best middle ground between them Rison does well with single stream performance here and does dominate in the price to performance argument versus the 7900 X by quite a lot particularly considering you could buy an r7 1700 instead of the 1800 X which we would certainly recommend because the 1800 X is superfluous you get overclocked to 1700 and get the same performance or better than the 1800 X out of box and that way you can run a cheaper solution than the 7900 X stream your one source twitch or YouTube and really do quite well in the price to performance market but to be fair to Intel the 7900 X does crush everything in the dual streaming test so far so that's not to say our settings couldn't do it you just really have to lose quality to do so both CPUs have their place just depends on what you're doing let's move on to another set of harder benchmarks this is VR testing we debuted our VR testing methodology in our r7 seventeen hundred versus i7 seventy seven hundred KB our benchmarks where we noted that the CPUs were imperceptibly different from one another you would absolutely not be able to confidently tell the difference in a blind test and if you didn't see our previous VR test we'd strongly recommend watching those for a full description of how this procedure works because it's quite complicated and the most calm kaida benchmarking we've ever done not going to re-explain it all here but the real reason to reintroduce these benchmarks now is to validate some of intel's marketing slides that we publicly disagreed with in the initial news announcement about these products in their slides intel would lead you to believe that their existing non-case cue still high-end CPUs are insufficient for VR gaming that's plainly false in fact oculus rift officially recommends recommend not requires recommends an i5 4590 cpu or better there's no K on there the HTC vive recommends the same hardware and in our own testing you would do fine with an i7 non-case q with really pretty much any game you could throw at VR so these slides served more of a point to find a way to say VR it does VR it does it really well rather than actually say that this is the capabilities of each of these products Intel's cannibalizing their own existing product line to make this one look better which we don't necessarily agree with so we benchmarked it to see what the performance would actually look like in an objective VR test since those really are still quite rare because it's hard to do and it's kind of interesting results but maybe not surprising now the main point here is going to be to compare the 7700 K to the 7900 X seen as we're really just validating Intel's own slides but we do have 1700 benchmark numbers as well if you want the hardware capture from the capture machine which is extra complexity check the article we're going to do just software here let's look at elite dangerous versus the 7700 K shown in blue the i7 7700 K technically outperforms a 79 hundred K here and that should be somewhat expected it's got a clock advantage which is often more beneficial than a thread advantage and generally speaking VR games don't really have the best multi-threaded programming right now that said the differences again imperceptible to CPS are effectively identical in actual perceived performance but if you are buying for VR gaming and nothing else this shows that clearly the 7900 X is not only unnecessary VR but technically closer to the 11 to 13 mil second frame time cut-off point before entering into drop frame or warp miss territory these 77 100k meanwhile stays closer to nine most I can average frame times again with a ninety Hertz interval we need to hit that 11 to 13 millisecond frame time targets prior to the run time kicking in and applying warps and finalizing animation and now just for a quick showcase here's the 7900 X and the r7 1700 in stock configuration the 7900 X it does indeed outperform the r7 1700 which was also outperformed by the 7700 K but between all three of these none of them are really running past the run time in any meaningfully different way in fact we've got a better way to display this data so here's a bar charge for elite dangerous with average FPS in terms of delivered frames of hmd we're seeing a perfect 90 Hertz on all headsets a simple extrapolation of unconstrained frames peg vi9 7900 acted below the 7700 K stock CPU but the differences again not noticeable our dropped frames are still just nine out of 5,400 intervals that's really not bad you are never going to see 9 out of 50 400 whatever percentage that may be in this type of scenario and these 1700 also does well here really all of these CPUs are fine and so there's no need for a $1000 7900 ax to do this VR workload though Intel tells us that it wasn't just marketing for VR gaming that was also for VR content creation except that's not what the slide said so that's not over benchmarking right now here's a look at leat dangerous average frame times in milliseconds I 970 900 X runs an average frame time of 9.3 5 milliseconds plus or minus 0.25 as there's some variance in these tests that's below the 7700 K for which our stock and OC numbers are effectively identical and are within our error bars and just ahead of these 1700 OC with dirt rally now we see a similar experience the I 970 900 X is technically slower and fame times in this interval plot than the 7700 K both stock though the two are realistically again imperceptibly different by 970 900 X would be a monumental waste of money for vr gaming considering that Intel to own i7 7700 K the best performer we've seen or tested for virtual reality by this interval plot and fame time charts every 900 is again technically better than the r7 1700 but again it's not noticeable and the 7700 K is still technically better than both at this particular workload here are some more bars since people seem to find that more palatable the 7900 exit lands at middle of the pack none of these cps are bad they all provide 90 FPS to the HMD again it's not worth spending $1000 when two 300 ish dollar CPUs from either cap including Intel zone can provide the same or better experience and that's an objective measurement as for frame times briefly the 7 900 exit lands at 8.15 milliseconds averaged between the 8.7 millisecond average of the overclocked 1700 end of 7.4 one millisecond average of these 7700 k performance is good but it's not better than intel's other products or and these other products that we've tested and that's fine the cpu does not have to do everything the best and that's what we want to convey to the marketers of the world we have this same talk with Andy's r7 1800 X when they boasted all kinds of gaming things like 4k gaming this and that we talked about all that then and the points not to pick on anyone the point wasn't to pick on Andy then the point is not to pick on Intel now even though it seems to be cool frankly the point is that Intel you don't have to BS people if you have a good product it does fine as the things that does find that as we'll see in a moment and as we saw in streaming a second ago but advertising this as a vr gaming best of platformer is not only wrong when your own products do better although they are all basically the same it's unnecessary and it's trying to bandwagon on to a phrase to letters that Intel clearly thinks will sell more products but all this really does is produce more marketing BS that harms the companies because it's it's just unnecessary and there's I mean that's really all I have to say that I guess despite the marketing the thing is not any better at vr gaming than other cps we've tested and just Intel it's ok you don't have to be the best at every single thing let's move on to more suitable tasks for this type of processor using jion's in-house blender scene including optimizations for rendering on CPS or GPS as appropriate we found the 7900 ex effortlessly chart tops all other tested products when it stock configuration the 700x completed the scene rounder in 21.5 minutes using its 20 threads to render 20 tiles simultaneously for the 4k scene this is the first CPU we've tested that managed to outperform a GTX 1080 with CUDA rendering and that's counting 256 by 256 GPU optimized file sizes the 7900 X improves over the stock 69 100k i7 CPU by nearly eight minutes or a 26% reduction in time required that's massive generationally and the more competitive pricing of Intel's modern 10 core part tells us that it was folly to expect Intel didn't have anything to compete that said the r7 1700 overclocked 3.9 gigahertz comes closer to competing with its 28 minute render time with the 7900 X about 23 percent faster than the $310 part regardless there's no denying Intel's lead here for production studios that have the budget and would kill for the extra render time as the CPU is a winner at its price Rison though should give everyone pause on pricing if absolute top-end production isn't the objective and the buyer is more of an enthusiast or artist without a studio budget the r7 CPUs can be overclocked to compete quite reasonably and would still be a good buy for much cheaper they are ultimately a completely different class of hardware for this type of thing now particularly with X 299 s memory advantages over Rison so this is one of those scenarios where it just depends on who has the money and who's paying for the product moving to Adobe Premiere we're still slower than mercury acceleration with CUDA but we're seeing steady improvements over all using our EVGA icx review from February to render which includes dozens of clips in various states of post-production we find the i9 CPU completing the software render in 54 minutes or about 7 20 percent time reduction from the overclocked 6900 K and 21 percent reduced time required from the stock 6900 K that's a decent jump and the r7 CPU is completely ran at about 62 minutes on overclocked even the 7900 X with a 13% time reduction from the overclocked r7 though with a higher price of course again trade-off of if you're a studio or an individual that's not great value for the average end-user given the $700 price hike but in that again a production house would consider then again for this type of premiere workload you're really significantly better off with a GTX 1080 and CUDA anyway we're at twenty minutes for those renders even with an i5 CPU we're just doesn't care about the CPU if it's given ample GPU to work with Cinebench and synthetic tests we'll be in the article below moving on to some games total war Warhammer saw an update posterizing launched that improves performance significantly on both Andy and Intel processors Intel saw a major uptick in frame time consistency reflected in our 0.1% lows and Andy saw major uptick in SMT enabled performance CPUs with an asterisk have been retested with this update while CB is without an asterisk have not been retested CPUs without the star would see the improvement of a couple percent in average FPS if retested stock vi9 700x operates at 168 FPS average placing it behind the i-5 7600 K and ahead of the 69 hundred K stock CPU and ahead of the overclocked 1700 X CPU with our clock DRAM the stock i7 7700 k CP u ranks at around 190 FPS average where we begin bumping into GP limitations which illustrates why h EDT CPUs aren't really meant for gaming despite some of their advertising some games will use the threads but many are still frequency intensive first battlefield 1 also got an update after aizen's launch moving to version 1.0 8 and improving CP about mine performance again the asterisks on the items that mean that they have been retested I nine seven nine hundred X runs 144 FPS average placing it roughly tied with a $330 700k CPU and better yet the $240 7600 K CPU the r7 1700 X was overclocked core memory speeds runs 136 FPS average marking at the 7900 X about 5.7 percent faster not a huge gain we also have gaming benchmarks for Ashes and watchdogs 2 linked in the article below with the synthetics each of the conclusions for the different things was kind of self-contained in its section but go over it briefly here the CPU is impressive for multi stream output it does really well and outputting to streams to YouTube and to twitch Rison does pretty well with one stream to the point where you really wouldn't need a some 900 X is not a huge difference at that point but for two this is the best thing we've tested it's just a matter of is that a use case you will encounter and also it better for you to just build a separate stream box the reason you would do it again is if you want the lower frame time variance for competitive games if you don't care about that then this is a cheaper alternative to building two complete systems for the most part for VR gaming this is a monumental waste of money and you should buy something helps for production work like premier or blender is even a better example in blenders they were big enough gains that the 7900 ex could be worthwhile to people who do blender rendering and similar tasks professionally and who might want to do it on the CPU rather than the GPU although again if you start pumping 10 to 80 eyes into it the argument gets a nebulous but for those people this is a good product at a good price for people who are more enthusiast artists or non production studios Andy is a great competitor at the price that it's at looking at something like a 1,700 overclock you get really good in close performance premier that performance is closer still in our testing you're better off with the GPU and our testing that's not the most conclusive because we can't test for every single use case but from what we've seen you're better off with a GPU anyway so it's it's better than the other CPUs I guess but that's it for this one so plenty of other stuff in the article below as always you can support our testing like this and our new 4k camera which is eating one gigabyte per minute right now if you go to patreon.com/scishow so you can buy a shirt like this on a restocking tri-blend this week actually stored our cameras next has done that and adding that part cost me about another gigabyte so thank you for watching subscribe for more I will see you all next time you