Gadgetory

welcome back to her own box today we're going to be talking about a new set of announcements from AMD at their next horizon event in San Francisco as expected the event was well it was mostly focused on enterprise data center and server products which we don't normally cover on Hardware unboxed but AMD did spend some time talking about seven animated trips including epic Rome and seven nanometer Vegas will be going through all that stuff we'll also talk a little bit at the end of the video on how some of these new technologies particularly Xen 2 might translate into consumer desktop products in 2019 but first today's video has been sponsored by asrock and their new phantom gaming range of Z 390 motherboards these are 390 phantom of gaming 6 and 9 include a blazing fast 2.5 gigabytes per second network interface offering gamers and content creators 2.5 times the bandwidth compared to the standard Gigabit Ethernet for more information please check the link in the video description we'll start with second-generation epic CPUs codenamed Rome and this n 2 architecture because the most interesting announcements relate to these CPUs of course epic is AMD so processor line so the specific products won't be of much interest to most of our audience but AMD is doing some very interesting things with Rome that we might see come to third generation rising in 2019 so the big update is a shift in total core count from 32 cores in existing epic processes to 64 cores with Rome and that's been made possible thanks to the use of TSM cs7 animated manufacturing process which is delivering a two times improvement to transistor density compared to the older 14 animator node but the cool thing about Rome isn't the fact that the CPU cores are manufactured on seven animators but a new IO die design with CPU core chip le'ts which separates most Iowa functionality and the CPU cores into their own separate dies as this diagram shows n 2 epic processors will have a 14 nanometer IO die in the center of the chip that houses the memory controllers in this case 8 of them along with infinity fabric interconnects that are attached to the CPU triplets there's perhaps a few other things in there Ayodhya as well but MB has indeed held them yet the chip 'lets themselves are built using seven nanometers and each triplet contains eight Xen two CPU cores along with the set of PCIe lanes so while the memory controllers are in the iodine PCIe remains attached directly to the CPU cores the key advantage to the i/o diet is that each CPU chip late' will have equal access to memory the current configuration for epoch sees each CPU die get its own memory controller so if one core needs memory that's attached to a different controller on a different die there's a latency penalty to access that memory compared to accessing memory attached to its own diet with the i/o die in next generation epoch that issue should be eliminated and every die will have equal access latency to the overall memory pool of course we don't know at this stage whether the i/o die increases latency compared to a core accessing memory from its own on-die controller in the past epoch design but that's something we'll figure out closer to launch and AMD does state the overall latency is improved with this new design with this iodine design AMD can now comfortably scale epic processes beyond the four die configuration of past CPUs so the top end epic roam chips will have eight CPU triplet dies each with eight Xen to cause for a total of 64 cores in the top configuration and presumably there will also be low tier chips we've caused disabled the benefit of using smaller 8 cores n2 dies on 7 nanometre rather than say increasing the existing design on a single die from 8 to 16 cores is a coarser yields and manufacturing costs smaller chips are easier and less costly to manufacture and it allows for easier bidding and this will help with first-generation products on 7 nanometer as you can see from some product shots of an epic roam ship the eight Xen to triplets are definitely very small compared to the massive iodine in the center which almost looks like a GPU in some ways other epic room features include support for 4 terabytes of DRAM per socket and a total of 128 PCIe 4.0 lanes yes PCIe 4.0 is supported with the new epic processes all of this combined allows anybody's new epic processors to provide 2 times the performance per socket and 4 times the floating-point performance per socket though we'll get into that in a moment Kerli AMD isn't willing to any cpu specifics like clock speeds or power consumption at this stage but just going from a 32 course to 64 cores per socket is a massive 2x improvement which is a big deal for data centers that love processing density AMD also provided some details on the Zen to architecture and further information on what seven nanometers allows them to do the big news is that Zen 2 increases the width of the floating point pipeline from 128 bits to 256 bits which should drastically improve one of the key weaknesses to am Lee's architecture and that was AVX workloads this should bring Zen 2 on par with Intel skylake architecture in this regard which as we know is basically the same architecture used in today's 9th gen processors AMD also improved a bunch of front-end stuff including a better branch predictor optimized instruction cashews a better instruction prefetching and more all of this strengthens their architecture to provide better IPC although only didn't specify any exact IP see improvement figures the company also mentioned they're working on Zen 4 which is no great surprise while Zen 3 is on track to use TSM C's 7 nanometre plus manufacturing known on 7 nanometers aim Lee says the new node allows them to provide two times the transistor density provide the same performance at half the power or greater than 1.25 times the performance at the same power AMD mentioned this in both their CPU and GPU presentations speaking of GPUs AMD announced seven nanometer Vega at the event today although seven data meter Vega will be restricted to server products in the form of the Radeon Instinct mi 60 and mi 50 still there's a few interesting things here as well Vega 20 on 7 nanometre uses largely the same GPU design as vega 10 on 14 nanometer in that it tops out at 64 compute units and it uses HBM to memory however vega 20 improves double-precision capabilities moving from 1/16 rate to half rate which is big news for compute applications there's also support for into 8 and into 4 with Vega 20 providing up to a hundred and eighteen terror ops of inter for compute performance Vega 20 includes a second pair of HBM two controllers beefing up memory support to 32 gigabytes with increased clock speeds of 2 gigabits per second the also getting PCIe 4.0 support and off trip infinity fabric links for direct connections to other cards despite all of this the dive of ager 20 is smaller than Vega 10 at 331 millimeters squared compared to 484 millimeter squared all thanks to seven nanometers and of course Vega tween does have more transistors inside in terms of performance well AMD isn't going into specifics on clock speed again but they did list fourteen point seven teraflops of FP thirty-two performance with 7 enemies providing around at one point two five times gain over fourteen animated in performance at the same power consumption the mi-25 had twelve point three teraflops of compute so that's around a twenty percent jump but for compute the real advantage is the massive improvement to FP sixty-four more memory and more memory bandwidth and of course the other new features as well as for product launches the mi sixty will be shipping in q4 this year with the mi 50 coming in q1 of next year epoch room is currently sampling now and will be launched at some point in 2019 although M didn't specify now let's head down well I guess speculations straight to talk about how some of these new technologies might come across to consumer desktop products because there's a lot of things here that will eventually make its way to AMD's next gen CPUs and GPUs however again what's coming up is just pure speculation on our part we really have no idea what AMD is doing on the consumer side for certain as far as third-generation Rison is concerned the big takeaway here is what AMD is officially stating seven nanometers can provide compared to 14 nanometers specifically they claim that they can get greater than 1.25 times the performance at the same power so first generation was built using a 14 nanometer process and third gen will be seven nanometers so presumably AMD will be able to achieve much higher clock speeds on seven enemies than 40 nanometers going on there 1.25 x perform its claimed thousand seven 1,800 X topped out at 4.0 gigahertz single core with the 3.7 gigahertz or core so if we did see clock speeds 1.25 times higher than that for next gen rice and we'd be looking at 4.6 gigahertz across eight cores and up to five gigahertz on a single core at the same power level as the 1800 X again that is just speculation but that follows on from what AMD is saying factored in all the other architecture improvements and we potentially looking at a real beast for third-gen risin the improved floating point pipeline is particularly massive news as this should bring third gen rising in line with intel's current processes for AVX workloads it's also possible that AMD will use the ioad I design which could allow them to put say two eight cores n2 dies on the one am for compatible chip the i/o dive looks like it could easily be cut down into quarters with one corner providing 2d Ram controllers which will be well suited to an am for CPU this would then results in a 16 core CPU on the a m4 platform which would be an absolute beast for productivity workloads without needing to go to thread Ripper and again going on AMD's numbers for seven nanometers particularly stating that they can achieve half the power consumption at the same performance theoretically AMD could use this aspect of seven animators scaling capabilities to double the core count of an a m4 chip at the same power consumption level that would mean something along the lines of a 16 Courtship clocked up to 4.0 gigahertz with reasonable power consumption that's very very exciting and it would certainly be interesting to see how Intel would respond as for GPUs well it's a bit harder to tell how AMD's next generation consumer GPUs will go when vega 20 announced today does not use AMD's upcoming na'vi architecture that will be used for their consumer gaming line so really we have no insights into what na'vi will bring to the table but again we are getting a look at what seven animators might do on the GPU side Vega 20 in the Radeon Instinct mi 60 is about 20% faster than Vega 10 in the instinct mi-25 for FP 32 work an AMD again says they should be achieving upwards of 25 percent more performance at the same power consumption considering Nvidia czar TX 2080 Ti is in the region of 60% more powerful than Vega 64 for gaming aim B will need to do a lot of work on the architecture front to get Navi up to the ballpark of cheering it doesn't look like seven animators alone will get them there that said in terms of R or compute performance the instinct mi 60 with 14.7 teraflops of FP 32 actually beats the RT x 28 ET i with 13.4 teraflops so let's hope na'vi can resolve some of the architecture constraints that prevented Vega from deliver during all of that power for gaming because if they can certainly unleash all of that performance in next-generation architecture and they'll be really great news for gamers anyway that's it for this brief look into what AMD announced today at next horizon we're expecting to hear more about seven animated consumer products at CES 2019 so stay tuned for that in January I'd love to hear your thoughts on what AMD announced in the comments below so chuck them down there consider supporting us on patreon where you can chat with us directly about announcements like this in our discord chat the discord was absolutely flying this morning actually during the announcements so you can be a part of that too if you want and I guess we'll catch you in the next one