Gadgetory

and these technical press event board information for both AMD Rison and Navi including overclocking information for risin Navi bass boost and average clocks architectural information and block diagrams product level specifications and extreme overclocking information for isin with liquid nitrogen we understand both product lines better now than ever before and can brief you on why am these working on we'll start with the navi specs die size and top-level architectural information then move on to rising and the also talk about rate racing during its tech day throwing some casual shade at nvidia and so doing and we'll cover that here as well before that this video is brought to you by MSI's x57 a motherboard lineup MSI's x5 70 motherboards will be available on July 7th with the launch of Rison 3000 series CPUs and now include flagship brands like godlike ace and creation the MSI edge will be among the more affordable mid-range motherboards appealing for most rise in 3000 builds while the godlike has more overclocking support an abundance of LEDs and OLED and a server grade PCB for improved PCIe for signal integrity keep an eye out for MSI's x5 70 motherboards for the rise in 3000 launch or learn more at the link in the description below so first off note that AMD did not give pricing information to us today and that instead will be unveiled at the e3 event which will be happening or will have happened right around on this video goes live so we'll do our best to post a comment and pin it to the top with the prices once AMD does unveil them but if we don't do it please post the prices below so that everyone can see them and get that information we do have a whole lot of other stuff though including a ton of architectural information and this event overall yielded a lot in terms of details on deeper down specs for the products some of it we're saving for later because there's too much of it to go through all here but we do have most of it for this I want to get one gripe out of the way first there's these tech days they they're primarily actual good tech information but there's a little bit of marketing screwy in this one and that was with trying to show a comparison of CPU thermals between two different CPUs by pointing a thermal camera at a heatsink and there's a lot of problems with that one of them the scale on that image was 23 to 27 degrees if I remember correctly which is insane for a scale because then you have white hot and red as a point two degree difference and separately pointing a thermal camera at a shiny metal heatsink doesn't do anything it doesn't tell us what the junction temperature is in fact you want the heatsink to be hot because that means it's working and also you're probably showing the reflected temperature of the room or the environment not the temperature of the device itself or maybe you you could deal with emissivity issues as well you really need like a blackbody object for that for measurement and then ideally a scale that's not four degrees wide just for reference what we would like to see here is an actual Junction measurement where you're looking at the internal processor sensor for thermals or you're at least putting like a thermocouple in there between the heatsink of IHS something like that anything but pointing a thermal camera at a heatsink that's that's actually useless so that was the only real gripe and for that to be the only major gripe coming from us is pretty damn good so just kind of set the stage with that I mean that's that's the main thing that we have to complain about and you know our content if that's the main thing to complain about them that's not so bad let's get into the GPU specs frequency and die size first and then we'll walk through the risin parts after that Andy has two GPUs it plans to release in July of 2019 with the exact date TVD at time of filming rise in 3000 to mostly lands on July 7th so the GPUs will push out around the same time the ApS will also launch on July 7th with the CPUs the two cards will be the RX 5700 XT and rx 5709 XT both moving away from GCN and towards the new rDNA architecture we've known GCM to be extremely limited in cache bandwidth and in scalability past 56c use so our DNA will be a refreshing change that should be interesting to study GCN will stay alive for now for the high performance deep learning and non gaming market segments but our DNA will be the game in architecture going forward the RX 5700 xt will host 40 compute units with 64 stream processors per Cu or 2560 stream processors for the 40c urx 5700 xt this is the same s P Percy you count as Vega 10 also add 64 stream processors for Cu but note that you can't linearly compare these streaming processors one to one because the architecture has changed a lot in an AMD performance demo the company showed the press that it's Navi 10 GPU that's the one being used in the 5700 series post about 14% higher performance versus Vega 64 with about 22% lower power consumption this wasn't am the internal demo so obviously we need to validate that at home base but that uplift is promising for our DNA is significantly smaller first GPU speeds were listed to media as 1905 megahertz boost 1750 5 megahertz for gaming workloads and 1600 5 megahertz base wear boost here really just means the peak clock or the rare opportunistic clock the delta between boost and gaming tells us a few things but one of them is that a.m. these chosen blower cooler is potentially a poor choice once again to no one's surprise there should be more thermal and power Headroom there to clock higher and we think that'll be proven once their custom designs out or when we put a water block on the card well talk about the negatives more later but let's get back to the specs and the architecture notes first as for GPU size AMD also provided that information to the press the vega 64 die for reference measures at 495 millimeters squared with Navi 10 measuring that 251 millimeters squared this is extremely important and it's for a lot of reasons by significantly cutting the die size and also by eliminating the highly expensive HP m2 andy is able to cut its manufacturing cost significantly and finally start competing with nvidia directly in price and he claims that the RX 5700 xt will be the direct RT x 27 t competitor we don't have the price while filming this but we'll make sure it's posted below as noted earlier and fill it in for us if you can't either way we know her fact that na'vi 10 will be a lot cheaper for AMD to make then vega because the die is smaller and there's no HBM to instead moving to GD d r 6 so these are good things this means the price will come down for the consumer as well in theory anyway and if an these claims increase of performance per watt and reduction and power consumption are even remotely true which they really should be there's no reason they wouldn't be then this could be and these property entry into the gaming market for the past couple of years we've basically gotten refreshes of refreshes which is extremely boring for everyone you have an RX 580 refresh into 480 and RX 590 refresh in 580 and and obviously didn't do a whole lot to gain and the any ground other than compete kind of with a 1060 class hardware so Andy is is still not competing at flagship level we would really like to see that if only for the halo marketing effect that's an important class even if a lot of people aren't buying 28 ET is for example it's a marketing move where the halo effect of having the best product on the market and that's really what it is in the gaming world even if it's not the best value because it isn't the value of having the best product on the charts is that people will look at that and then they'll think I want that but I can't afford it they'll buy the next tier down so Andy really does need to get a flagship out there at some point but they're starting with a for TCU part that will compete with an RT X 2070 and that is a really popular market segment so it's not a bad move it's just one that we've seen repeated time and time again by am the this time however it does look like on paper that the the specs are more competitive than the previous attempts at competing in this price category so that's the the hope that we see some renewed competition with the new architecture that is promising but we do really want to see something the flagship territory at some point it's just it's not gonna be this time at least not this launch but maybe this architecture later down the line in short the 5700 XE has four TC years eight gigabytes of gddr5 14 gigabits per second for four to forty eight gigabytes per second of bandwidth 17:55 megahertz gaming frequencies 1905 megahertz boost likely limited by the blower cooler and we'll have to see what better solutions can do later the RX 5709 xt will cut down to 36 e use 2304 stream processors with an 8 gigabyte GD r 6 frame buffer and GPU speeds of 17 25 megahertz boost 16 25 megahertz gaming and 1465 megahertz base and ii was forced to acknowledge the real-time ray-tracing marketing battle this time but doesn't have an immediate plan to support it from our end we think it's fine that there's immediate support for real-time ray-tracing it's just that the market demand for real-time ray-tracing is it's sort of hard to gauge so we know that there's not a whole lot of support for RTX when and video launched with it and the no games supported it for one and so it's interesting to see the script flip a bit between everyone saying RTX has no value to now people are saying well but it doesn't have RDX and that's something you'll see more in our discussion with Gordon from PC world in a separate video but to give credit to and video they have forced AMD to acknowledge ray-tracing and start talking about it and AMD is a way of doing this was mostly to say well we'll push the features and the hardware when the two are ready to be shipped together so throwing a little bit of shade at Nvidia for that and they talked about a quote driving the ecosystem so that gamers can take full advantage of the features they pay for which is a fair comment to make and a bit of a jab but as a quick aside and a separate one note that AMD is planning on making a seven nanometer plus a navi GPU for late twenty twenty or twenty twenty-one and that's aiming to support some form of maybe hybrid ray tracing in the future we don't have exact details but and they did note an interest in moving the sport parts of ray tracing with full ray tracing support more targeted for cloud computing later indicating that this is not something I am the expects to be doing locally anytime soon and that's probably realistic the industry isn't isn't quite ready for it yet as we've seen with our TX total board power should be 225 watts with a fifty seven hundred XT 180 watch the 709 XT we asked if V BIOS would be locked and due to security concerns AMD is locking down custom V BIOS modifications we then asked if power plate table hacks would work this generation and we were told with a laugh that we'd have to learn new tricks to bypass the power limits it's a little bit of a letdown from the friends use case of power modding but we're hopeful that we or more likely build Zoid can find new power bypasses the reference card will be using a blower design there are two different shroud designs that we've seen but the PCB is overall mostly the same with one change the V R M so schoolers a blower cooler unfortunately it has a vapor chamber this is a design we've seen struggle a hundred percent of the time in the past Andy is adamant that this is the way they want to go with it primarily for reasons of using it in poor ventilation cases which we've done that test and a lot of the time the dual axle cards do still work better because you're sucking all the heat from the whole system through the GPU otherwise but either way and he has made some moves in the right direction by restricting the noise this time so the rpm ranges from a couple hundred rpm at the low end when it's idle to about 1800 rpm or so max when it's under load and that will vary depending on how cool the ambient case temperature is but Andy says it's limiting the noise level to 43 DBA now a quick note that doesn't actually mean anything right now because we don't know the distance that number is measured at so 43 DBA without the comparative of of their measurement before the original coolers really doesn't mean anything to us presently but it's not because they were trying to hide it just because they genuinely they didn't know and they have to go find the answer for that and we won't have it before we film so what we do know is that Andy said their measurements more or less matched ours and for what that's worth we found that the previous blower designs ranked about 59 DBA under 100 and so if Andy's capping at 43 this time that is a significant reduction in noise remember that this is not a linear scale it's a logarithmic once that difference means a lot more than the linear Delta would be so that's potentially a good thing but we need to test the card to see more the reference 5700 XT PCB will have a seven phase vrm we don't know the power stages unfortunately did ask but not something they had for us today it's probably a six plus one configuration but Andy can confirm that for us today again they're not hiding it it's just a matter of getting information quickly enough and then the 5709 xt is going to have one fewer phase that'll probably be a 5 plus 1 power stages we need to follow up on and get full full information on that for the analysis of how much power the the PCB can handle we also for what it's worth didn't see a V bio switch on the card so it looks like a single bios this time it should be a pretty final retail version the card that we saw but there might be one more revision of changes so maybe we won't commit to a firm statement but it does look like probably single BIOS on that card before diving any deeper on the GPU stuff let's talk CPUs for a bit this was a very complex and long tech day there's a ton of information we won't be publishing here today on the CPU side of things AMD did have an EXO c2 demo at the event and we learned that Andy was hitting about 5.35 gigahertz all core on the 16 core CPU which until now is not officially shown and this was using an MSI godlike motherboard something for which we have some some initial coverage from Computex and builded will have a full PCB breakdown in his channel but godlike board is about topped here for MSI 4x OC and this matches the numbers we mostly heard previously we've also heard 12 core this from a board partner should do 5.5 Plus approaching 6 gigahertz on Allen too so 16 corr 5.35 gigahertz fell into does make sense we don't have h2o numbers just yet unfortunately but we are trying to get them and as for the rest there are a couple of critical notes for overclockers that you may be happy to hear if you do benching then PCIe gen 1 no longer has to be used for Rison for the actual 3d benchmark or for benchmarking in general so now you can do 3d benching with Rison and enable PCIe gen3 or for previously you had to disable those and go down to Jenn Wan in order to run the benchmarks properly cold bugs will depend her CPU but the one that was being run for Andy's demo was run at about minus 150 degrees Celsius for the Alan 2x OC demo for memory overclocking reisen 3000 is also looking more promising than previously it's supposed to fairly easily support 30 600 megahertz XMP without issue and then we've also heard on MSI's boards of going upwards of 50 100 megahertz for the memory at timings of 1821 21:56 and 1t for the command raid that 51 megahertz 50 100 megahertz overclock was done without Elan's who on the cpu so that's just a memory overclock that's a result of better trace layout and a results of a new controller design some of the more noteworthy architectural changes to rise in 3000 CPUs focus on higher IPC and higher game performance interestingly and these engineering discussion the senior architects and and the fellows thought that Zen to woodsy frequencies regressing which is something that you do typically see what they process reduction for example down to seven animators for this one engineer's originally thought the jump to seven nanometers would result in lower clocks relegating Zen to to a server part but it's actually coming out to desktop client so that's quite the change what it was going to be otherwise the team was able to figure it out and some of the key changes included the following AMD doubled the size of floating-point Engine two to three 6 bits meaning that AVX of 256 can now run a single cycle on Zen 2 so that's a big change and he has doubled its load store bandwidths for data storage and movement integer execution was also improved primarily by reducing resource contention during integer operations and he also doubled the l3 cache size more data and local cache should mean fewer hits to memory but also reduced effective latency to system memory and the further highlighted that this improvement in the cache access and in prefetching algorithms should reduce the effective memory latency and as an aside and he has made the decision unfortunately to rebrand l3 as game cache so if you see that name in the future that's what that means that is what l3 is now Andy showed some more internal benchmark charts and while we know there is an improvement to gaming of course we'd advise looking out for our own benchmark test whom the processors launch some of the internal charts looked a bit GPU constrained down which one you look at but we do anticipate performance uplift if only from the IPC increase and of course from the frequency increase so these are places where you will see performance increase especially in games versus the previous generations memory is also more easily tuned to 3600 megahertz as noted earlier and beyond to 5000 with the msi setup that was used for memory extreme overclocking probably by top PC and he also walked through architectural changes that will speak to in more depth approaching the launch for some quick stats we learned that Andy has moved its integer execution to a 92 entry integer scheduler from 84 previously it has increased physical register file entries from 168 to 180 that has grown the reorder buffer from 192 to 224 which is meant to help with making up for the latency hit by increasing l3 size Andy has increased the historic from 44 to 48 with increased load/store bandwidth by 2x and that should recap us for now so again there's a lot more we have several documents on all the breakout sessions there's a lot to learn here and we'll try to cover that over the next week or weeks as we approach the launch but for the pricing we don't have it that time of recording leave it below or we'll do it if we catch it in time thank you for watching subscribe for more go to patreon.com/scishow and Razak cents or stored at cameras XS net sports directly we'll see you all next time