Gadgetory

the rise in the hype train reached velocities capable of entering low-earth orbit as we ramped into today's launch we had four days with all hands on deck to benchmark the and the risin are seven 1800 X that were reviewing today socketed in aces high-end X 370 crosshair motherboard our risin are 7 1700 review opposed tomorrow with the 1700 x review posting the day after that today we're looking at the 1800 X and blender premiere thermals and gaming workloads and before getting to risin this coverage is brought to you by a thermal takes new contacts island 12 CPU cooler the contact silent 12 fully supports the am for socket type and aims to be available around the 30 dollar mark thermal takes contact 12 uses a hydraulic bearing for longer and quieter operation and would do well on an AR 7 1700 s learn more at the link description below speaking of links in the description this review is massive it's a new architecture so check the link in the description below for the full review its nearing 10,000 words has a bit more depth than we have here because of time constraints to check the article for the full rise in architecture and benchmark but we can go into most of it here the core specs are on the screen now the r5 1600 X and 1500 act interestingly it will offer sub $300 parts with 6 cores and the 12 threads in quarter 2 with our three parts do in the second half today we're looking only at the r7 1800 X for some brief details and major changes to the rise in the architecture center around the ccac's and at replacement of bulldozer modules rise in CC X moves to implementing a single FPU per core rather than one per two integer units on bulldozer and it also uses a single sim D and private l2 cache per core Andy has moved to wider instruction level parallelism but still uses narrower floating-point units and Andy is running 512 K of l2 cache for core at 1.5 millimeters squared per core this 512 K cache is eight way associative so Andy doesn't need probes to interfere with low latency caches during snoops the SRAM is less dense than Intel's but is still 60 SRAM for tagging data with eight T at four state l3 cache totaled 8 megabytes at 16 millimeters squared of the total die area which by the way is 44 millimeter square l-3 is also 16 way associative and shares pol with all cores the l3 cache can be shut down if only l1 and l2 are in use which helps with automated power savings each CCX is a fork or a thread module means that we won't be seen dual-core CPUs out of the currently known Zen architecture and these l1 and l2 bandwidths are up about two times over previous architectures with l3 up about fivefold the victim cache is shared across the ccac's and operates at the frequency of the fastest core which means that it also reduces frequency alongside the fastest chord to further provide power savings for each integer units there are 4 Al use in to AG use or address generation units with a 2 K op cache present and Angie's got a new micro app cache that is also something we'll be talking on the article and these Q have gotten significantly larger versus bulldozer though the company does still have major architectural differences from Intel one example would be am these AV X units the company is running to 128 bit wide units instead of the 256-bit wide units in skylake and is using two 256 bit load port this design takes less space over all contributing to andis smaller area per core than comparable Intel chips but does it mean that we should expect different vector performance andy is also running one fewer aju than its direct competition which should help the company clock higher and will also lower power consumption this is potentially at the cost of performance and workloads that can saturate 3ag use with two loads and to store for clock but Andy thinks that it's worth the trade-off as for an answer to Intel's fiber and the has one they're sticking with linear voltage regulators to reduce complexity heat and retain conversion efficiency and given how Intel's fiber was received that might be a good choice if you're curious about the m4 chipset differences we already have that content on the channel search for and port chipset differences or hit the link in the description below to the full review let's talk about memory for now in the leaks leading up to aizen's launch there were plenty of toasts by users and other sites that talked about potential memory limitations on Rison this was true to some extent though maybe not as exaggerated as some folks may have made it the official memory support for Eisen right now is 2667 megahertz that's what was largely suggested for benchmarking we were able to clock up to 29 33 megahertz on the kit provided and used that for our tests and we're not able to sustain higher clock rates than 29 33 for a few reasons one of them we have a high next I Corsair kit it officially is rated for 3,000 megahertz but to get any kind of higher frequency you would need to do a few things one custom reference clock and V dim tuning and to ideally step through memory training and then three note that there is a distinct lack of memory sub timings in Rison and that's more of an ante Amanda motherboard things that would be less on ASIS and more on Andy this makes it a little difficult to get the higher clock rates that some of the kids might sell at and at the time we were working on this review which is of course prior to launch the only officially validated kit of memory is the Corsair 3,000 megahertz of engines LTX kits that we had access to which again we clock to 29 33 megahertz and after the theory there was time to get to the benchmark as always the full testing methodology and all the platforms used it's defined in the article in the description below we have to obviously change motherboard memory depending on architecture stuff like that it's all defined down there and we're going to start with the risin thermals versus clock over time and then we'll be moving on to Adobe Premiere Blender and gaming workloads this first chart shows two things primarily we're looking at CPU Junction temperature in blue and average frequency of the cores in red ideally the redline is perfectly flat when under heavy load indicating a more stable core clock the blue line should never exceed 75 C as that is technically AMD's maximum junction temperature for the 1800 X or t.j.maxx we were able to intentionally trip a thermal shutdown by heating the CPU beyond 75 C and validated that yes t.j.maxx is in fact between 75 and 80 Celsius gauging thermal performance on AMD isn't exactly easy and the diode reporting has some inconsistencies at times that manifest form of massive reported temperature drop in reality we're able to use thermal probes mounted to the package to gain some additional insight as to the accuracy of the measurements and then we can make appropriate adjustments to our testing methodology if we zoom into the chart you'll see that the clock rate becomes less stable and starts to fluctuate in increments of 25 to 75 megahertz at once temperature approaches 73 to 75 Celsius this is and these internal sensors working with the clocks to keep temperatures under control down clocking and rapid intervals and up clocking again to keep thermals below 75 C and by the way this is with an NZXT crack and X 62 at max fan speeds so it's running pretty hot this is an 1800 X anyone putting this chip under 100% workloads for long up times like blender rendering and things like that should seriously consider better cooling than the nock to a unit that Andy provided most reviewers and that's why we switched thanks to the m3 support on the crosshair to an X 62 now we also log voltages versus applications power draw things like that for comparative power draw charts link in the description below but we can go over the basics now with 185 at watts total system power consumption again that's total system power consumption we see the rise in r7 1800 X operating at a steady boost frequency of 3.7 gigahertz when all cores are engaged this number only ever increases to excess power range of four point one if a single thread workload is underway which is pretty uncommon for most of our audience but we do have some tests to show it the voltage it to sustain blender at this clock rates and 100% load ranges from 1.2 to 1 V 2 4 3 Cinebench 1t pushes us up to 4 point 1 gigahertz at times but fluctuates between three point seven and four point one this workload generates a voltage of about one point three nine five to one point four six 1v core with total system power consumption at 92 watts finally pov-ray 1t runs a 94 watt system power draw with 4.1 gigahertz fixed for the frequency fully an acts of our range for the entire bench and a core voltage of about one point four one seven volts let's open up with rising strong points our video production workload is a real-world test pulling from files and exports from our EVGA icx review so this is a real render for our youtube channel color correction transforms and other calculations like warp stabilization may be included in the render check the article below for specific render settings configured to its stock clocks the r7 1800 X completes the path in sixty 7.4 minutes with the sixty nine hundred K completing the video render in sixty eight point five minutes if you prefer percentages that's a change of one to two percent and rendering terms you'd be saving about one minutes for every hour of encoding ignoring all context that's not impressive but context is what matters the context here is that the 1800 X is a five hundred dollar CPU with high end motherboard support - like me crosshair available for two hundred fifty five dollars Intel i7 sixty nine hundred K is a thousand dollars and high-end motherboards cost three hundred plus quad channel memory of course is lost with Verizon but it's also not the most beneficial item in this type of test that said anyone running production renders knows it's the GPU that counts for most rendering tasks despite some specific use cases where a high-end CPU can be beneficial for premiere we'd never use the CPU as a primary render workhorse CUDA and OpenCL acceleration we're supported are significantly faster and this is where Andy's argument starts to fall apart this chart puts our render numbers into perspective relying instead on the GTX 980 FTW and CUDA for the rendering process the result is expected our render times dropped from 60 minutes to 20 minutes or thereabouts rendering on the CPU takes three times as long and using a rising CPU and GPU is not any faster really much slower than using a sixty nine hundred K and the same GPU to render and the deserves praise for accomplishing comparable render speeds to intel's double costed cpu the sixty nine hundred k but we also have to frame that praise within reality and the reality is that for this type of workload you're probably using cuda anyway so the argument is a bit weaker when considering that option moving on to blender GN uses a custom-built blender benchmark for its cpu rendering performance analysis the benchmark mixes for rendering motion blur multi bounce rate racing and various transparencies and mattes the Andy it Rison r7 1800 X completes the frame render ingest under twenty eight point seven three minutes with the i7 69 100k completing the render in twenty nine point five minutes this puts and these five hundred dollar eighteen hundred X about on par with you one thousand dollars fifty nine hundred K in the blender rendered tasks at least when both are in stock configuration the r7 1800 X is therefore about one to one point one percent faster normally this would be unsurprising or uninteresting but the price makes and these are seven platform a worthy consideration over clocks we see the sixty nine hundred K at four point four gigahertz reduces render time into twenty five point three five minutes while the rise in chip overclocked is at twenty seven point eight minutes so Intel now pulls ahead and the falls behind on overclocking but still half the price Intel also overclocked so definitely higher in memory if you were to actually try which we didn't do too much with that and that's potentially useful in some specific tasks as for other devices the three hundred forty dollar seventy seven hundred K a stock hyper threading on completes the same render in forty two to four minutes with the overclocked variants finishing and thirty eight minutes we're curious to see how the rise in our seven seventeen hundred holds up to the seventy seven hundred K in this task for reference the 83-70 completes in ninety point two two minutes so either way Andy's doing a lot better than previously rise and holds a lead here even if it does come down to a difference of about one point one percent but now it's time to move on to gaming workloads so if you want to see the rest of the synthetics and other test that includes fire strike Cinebench and time spy and pov-ray benchmarks you can find all that in the link in the description below and the full review article moving onto the battlefield one the sixty nine hundred k and seventy seven hundred k overclock so some loss and lows are bumping against the wall around one forty-one to one forty two FPS average we're still seeing the high IPC and clock rates of a 70 to 100 k planting the chip about equal to an i7 sixty nine hundred k this remains true for the 6700 k5 five seventy six hundred K and the i7 4790k by five sixty six hundred K begins to drag behind a visit 137 FPS average and that CPU is eventually trailed by the r7 to 1800 X overclocked to three point nine gigahertz which lands somewhere around 135 FPS average one percent lows hover around 109 with 0.1% at nine before for the overclock subversion of the CPU and considering the low percentile performance of neighboring chips that's good performance overall however considering the price and age of those nearby chips the 1800 X doesn't look so great in this particular benchmark ultimately the $500 r7 1800 X is effectively equivalent to an i5 4690k in this battlefield one benchmark the i5 4690k came out a few years ago and once retailed for $240 trade off the bound as usual with the r7 being veteran production the one thing is clear the 1800 X is not impressive in the battlefield 1 department and is matched evenly by 2014 Intel architectures you may have noticed a black bar across the data let's remove that now disabling SMT or simultaneous multi-threading on the r7 1800 X results in performance that boosts us to 135 FPS average from the 132 FPS average of the stock clock r7 1800 X because the gamers Nexus also measures frame times we benefit from the knowledge that disabling SMT further increases 1% low and 0.1% low metrics by upwards of 30 fps in some cases around 30% better 1% low values with SMT off you'd be better off disabling SMT than overclocking in some cases so if you do both you'll get even more of a performance jump and by the way Andy is not alone on that previously in much older architectures when we tested years ago Intel hyper-threading did add a bit of overhead to the point where in some games and applications disabling a hyper threading might have gotten you 1 to 2% FPS increase so not as big with differences and Eocene here with this SMT toggling but also Andy is not alone the overhead from multi-threading simultaneous or hyper threading however you want to call it is real and it's something we've seen in the past it just so happens that Intel doesn't really see that effect anymore total war was only just added to our charts in time for the 2600 K reviews so all the i7 CTS haven't been fully subsumed by the warhammer bench just yet we're hammer also gets finicky about the stability of some overclocks so as you'll see in the frame time metrics for the 6900 K they can jump up and down a bit to the point where it's not always the most usable metric with this particular game where hammer appreciates clock speed - great extent we see the trade-off between cores and frequency when ricochet in between the 7700 km 5900 K processors by 770 700k pulls ahead in every regard when pitted against these 6900 k thanks to its 4.5 and 5.1 gigahertz clocks and lands around 186 27 FPS average and let's stock as for the rise in r7 1800 X the new at Zen arc CPU land between the 70 350 K stock and the 5.0 gigahertz overclocked variant total war shows a clear frequency focus further emboldened by the 1800 X's at 5 fps or about 4% increase from a 200 megahertz overclock partly for this reason among other architectural advantages Intel again holds the lead even with processors of significantly lower cost we still wouldn't recommend that you go out and buy an IV 73 to PK not for $180 anyway but the 1800 X performs just slightly better than the over price but still cheaper I 3k skew CPU let's move that blackout bar total Warhammer now shows the biggest change in performance when disabling SMP the AMD r7 1800 X moves from 127 to 132 FPS average depending on the test and speed to 153 FPS average with SMP off the table Anandi's most advertised property is increasing performance in the best case or in the worst case resulting in basically no performance difference SMT seems to hamstring frame latency to some extent and creates enough overhead to burden us in total Warhammer let's look at watchdogs to next at 1080p high we're seeing the sixty nine hundred K top charts around 119 FPS where we're beginning to encounter a GPU bottleneck on the 1080 FTW the low percentile performance it's at 94 FPS 1% 77 fps 0.1% comparatively the half-price r7 1800 X is operating around 284 FPS average with 1% lows hanging around 62 fps and 50 FPS 0.1% in terms of raw frame rate that is a performance delta of approximately 35 fps and averages between the two comparatively the 7700 K stock operates at around 113 FPS average with the 6700 K skylake CP around 110 average even the devil's canyon 4790k circus 2nd quarter 2014 is outperforming the r7 18 hundred x please not 100 FPS average disabling hyper-threading on the 7700 K plants was about equal to the r7 1800 XS empty enabled so eight core 16 threads further demonstrating and architectural advantage for Intel right now in gaming workloads outside of the raw core and thread count of the AMD CPU overclock in the 1800 X pushes to 87 FPS average with marginal improvements in the lows the maximum sustainable overclock with our chip and motherboard seems to be about 3.9 gigahertz resulting in an increase of 3.2 percent from stock so we're at 3.7 gigahertz on all cores stock or 3.9 with EOC and ultimately for watchdogs to the 1800 X performs mostly equally to an Intel i5 7600 K CPU for $240 disabling SMT reveals that performance is largely the same in watchdogs - not as profound as in total warhammer netting an output of 85 FPS average versus the original 84.3 FPS average so that's um he doesn't really help Andy here but it's not hurting either and by the way watchdogs too will utilize all the threads to some extent not the only thing that matters in this case clock rate still matters but this illustrates quite clearly that just having more cores doesn't when all arguments by default and that's something we've experienced time and again let's CPU architectures that venture into a high core count territory for time reasons you can find the ashes of a singularity Metro last light in GTA in the article in the description below if you want those benchmarks we move on to a little bit of 1440p now when we presented some of our early findings to AMD they suggested that we benchmark the CPU by creating the GPU bottleneck with 1440p resolution still it's a little misleading to the overall capabilities of the CPUs that such a benchmark means that future GPU upgrades could be bottlenecks by the CPU which wouldn't be known if we were incorrectly passing the CPU by imposing GPU bottlenecks regardless we ran a few anyways despite only having four days to test the CPU here's a look at watchdogs 2 at 1440p high the AMD r7 1800 actually when take in Andy's suggestion of burdening the GPU is still performing at a frame rate deficit to the cheaper 7700 K now of course the next argument might be that the visible difference between 86 and 77 FPS average is not profound and that be a fine argument it unfortunately doesn't make sense to spend more money and get less that's just bad value besides when considering the CPU and platform tend to last the longest in the system upgrade pathways are going to be better on the platform which has the highest baseline performance frame times that suffer a bit with the 1800 X file in 248 Giovanna jean-louis 60 fps 1% low versus the 7700 K at 68 0.1% in 74 1 % this isn't something where we're seeing stuttering he'd have to fall a lot lower for that so mark that very clearly not stuttering but it does indicate that Andy has other potential weaknesses that could emerge under the right conditions or later on the next argument might be that mixed workloads users like content creators would benefit from production advantages afforded by Rison that's a much stronger argument but still has caveats encounters will point you back to the premier benchmarks for that discussion moving on to battlefield one at 1440p we're definitely developing the AGV bottleneck and beginning to invalidate the wants of 120 + 144 Hertz users the 1800 X using the GPU as a crutch is now more or less on par with these seventy seven hundred and sixty nine hundred ksed use and we're still seeing a decreased frame latency versus Intel but just like with loss talk - that's not producing any noticeable stuttering to the end user it's just measurable in this particular title benchmarking incorrectly does make the numbers look better for the 1,800 X but the 7700 I perform zit is cheaper and really we have to defer to the r7 1700 at $330 to see how andis future looks with gaming titles we'll be testing that hopefully tomorrow at this point you might be feeling a little disillusioned that when considering and these previous tech demos keep in mind that most of the charts leaked and the and the official benchmarks shown were based on Cinebench or were gaming workloads that were inflated in one way or another by doing a few different things in benchmarking the Sniper Elite demo and the frequently looked at the skybox when reloading and often kept a larger percentage of the skybox macrosse1 than the other side by side Intel processor a skybox has no geometry which is what loads the CT with draw calls and so it will inflate the frame rate by nature of tasking with a chaotically conducted methodology as for the battlefield 1 benchmarks that were demonstrated at the Sonoma event and the conducted be is using other chaotic methods wherein and II would zoom more frequently or at different intervals than the side-by-side Intel CPU making it mostly impossible to actually compare the two heads ahead so you're left with a number that basically says these are more or less equal and most importantly of all these demos were run at a 4k resolution that creates a GPU bottleneck meaning we're no longer observing true cv performance there's an argument to be made that low-end performance doesn't matter if you're stuck on the GPU but that argument doesn't always hold you don't buy a worse performing product for more money especially when GPU upgrades in the future will eventually even out those limitations as bottlenecks external to the CPU vanish as for blender benchmarking although Andy was looking at deltas and so it doesn't matter too much Andy did demonstrated splendored benchmarks using different settings to what we would recommend and it's blender testing and the execute surrenders using just 150 samples per pixel or what we consider to be preview quality and runs unoptimized well somewhat at 32 by 32 tile sizes not awful but we'd recommend 16 by 16 they also render at 800 by 800 resolution in our benchmark render using 400 samples for pixel for release candidate quality 16 by 16 tiles and a 4k resolution this means that our benchmarks are not comparable to an B's but they are comparable against the other CPUs we've tested as for Cinebench Andy ran those tests with the 6900 Kay platform using memory and dual channel rather than its full quad channel capabilities that's not to say that these results would drastically change but it's also not representative of how anyone would use an X 99 platform Cinebench isn't everything that neither is core count as software developers move to support more threads if they do perhaps Andy would look better for now it is clearly not a good buy the 1800 X that is in the gaming market there is an argument to be made for the 1800 X and its value in production workloads again premier and blender it looks pretty good compared to the $1000 Intel chip half the price it performs at least equally if not slightly better other than when you overclocked Intel so there it looks good but again you have to consider are these workloads that are production focused something you can push to a much faster GPU even if it's a 1070 or something like that because at that point a lot of reasons by either the $1,000 or the and the $500 chip there are uses for the software acceleration but if you're one of those people who needs that you can hopefully figure out how the performance looks on your own the r7 1700 may prove a better value for gaming we'll have those benchmarks shortly for now though the 1800 axis of disappointment and is not a processor we would recommend for gamers at its price point even mixed workload users should consider when and where software acceleration is better than GPU acceleration if ever wellyes overall the 1800 X is an impressive competitor to the 6900 k and it's significantly cheaper so we'd recommend the 1800 X over the 69 hundred K if you are in one of those use case environments where it makes sense to have such a thing for now we're going to suggest waiting for the r7 17100 of you if you're a gamer and hopefully buying in the lower end market because if that might have a better chance at competing with Intel considering the giant price to performance disparity between the lower end chips and the higher end chips from both AMD and Intel that's all for this one thank you for watching as always at patreon.com slash gamers Maxis if you would like to help us fund this type of in-depth analysis link in the description below for more information definitely subscribe for seventeen hundred and seventeen hundred extra views and in attendance which tear down that would be posting the shortly I'll see you all next time you