Gadgetory

reisen that is the name that AMD has gone with with their new md's mcp uses a first one out of the gate that's really been detailed under the Sun architecture risin is a combination we're told of horizon and Zen which at least there's a reason for the naming but it's not really what's important here what's important here so we have some preliminary information on the caching the speed and the core count things like that some of its already really been discussed for a long time now so it's even been confirmed but we'll go over what was shown at the horizon event before getting to that this content is brought to you by Corsair and they're 270 our case which is about a sixty-dollar mid tower ATX form factor enclosure that has really easy to use these of installation features and makes first time building it pretty straight forward link in the description below for more information on that so back to rise n here's the thing we're looking at today first of all this is all coming out of the horizon event that was live streamed via YouTube by Andy officially and recapping the most important stuff we already know that Zen is using a victim cache it was sort of discussed again briefly and it's fully associative so this means that you're really only losing one cycle of performance if you have a cache miss and that's important but this has been at least loosely known on the perimeter for some time now some Zen architectural details have already been published officially by the way not just leaks so that's kind of known but it is victim cache that was confirmed here 20 megabytes l2 plus l3 this improves latency and that's kind of a big thing that Andy's pushing forward with this architecture the latency reductions made are across three different things so cash is one of them and then just in the switching frequency switching boosting and down clocking and stuff as necessary to better work with whatever the current workload on the CPU is so those are main things improving latency is also being done through better prefetching and through better predictions branch prediction things like this that are really buzzwords for most major CPU architectures in the modern era but are coming up in full force with Zen because again if we look kind of back in time with AMD cpus they haven't had any really major architectures and quite some time now the APU chips were really the main area we've seen innovation FX although it has received refreshes and updates the architecture has really not been that different since bulldozer piledriver steam roller at all had some changes but Zen is the first really truly major step for AMD since many years to cut at this point so to that end prefetching is a big thing that they were talking about and branch prediction this reduces hopefully reduces the requirement of shuffling data around is shuffling your whatever you're working with at the time between the cash on the CPU package and the system memory so with better prediction that's hopefully lessened movement of the data and therefore improve the latency because if you're transacting everything locally on the chip and you don't have to reach outward to get it or you're not ditching things in cash that could actually be useful in a moment then obviously you improve your processing time and this for those of you who have followed our GPU coverage when we talk about the architectures of both Polaris and Pascal it's a similar trend there now two very different things in terms of what you're actually designing it to do but the steps and improvement are pretty much in the same area so we've got reduced latency and you've got better boost functionality that's the next major point talk about with Zen and specifically with the rise in CPU that we saw briefly at the event so the improved frequency boosting or management you can call it similar to what GPUs do reisen will basically look at the current workload and then add a 1000 to Hertz interval it so a thousand times per second it can modulate the frequency to match the workload now this is important because really other than sort of benchmarking and overclocking scenarios things like that you don't want to blast your CPU at one hundred percent of its complete maximum clock capabilities all the time because it's just not power efficient there's no real point to do that if you're doing basic tasks and this isn't new to the industry because CPUs have been boosting for a long time now but the improve movements are new so risin you're getting the 1000 Hertz change rate basically maximum potential change rate of the frequency from the marketing materials we have now anyway and it's got the the base clocks kind of obfuscated it either it's got a lower bcl cave and normally or it can change it basically said 25 megahertz increments so I don't know if that's a lower be clk all together and a higher multiplier or if it's something else just in terms of the granularity with which the clock can fluctuate based on load that I'm not clear on right now and we won't be for a little while yet but we are supposed to be briefed on Zen pretty soon theoretically CES is generally a time frame where more of this information would come out so we might have more of that in very early January but that's what I got for now so I'm not sure if it's a lower bc okay and a higher multiplier or what's going on there but 25 megahertz intervals is the the key phrase that they used when discussing boosting on risin now boosting the reason i said it's sort of similar to GPUs is really just because the parameters under which zen is going to boost are comparable and that's because then the risin chip I should say is equipped with according to md hundreds of sensors and those sensors do things that you would expect them to do their sensors for temperature sensor voltage down to the millivolt incentive sensors for wattage down to the mill of wat and these things look at current workload current parameters working conditions of the CPU is it too hot is it too low that is it low enough voltage that you could boost the voltage and increase the clock rate or what's going on so that's in there that helps with abuse functionality and with regard to thermal sensors specifically I suppose the theory would be that just like the way both AMD and NVIDIA GPUs will boost rise n will sort of in some distant similarity be able to boost if it's got enough thermal headroom if it says ok I clearly have plenty of temperature range here to grow so it can boost its performance and then does so until hit some other limitation that's detected by those on chip sensors is whether that's voltage or power or maybe thermal at a later time when it's boosted higher so that's what you're looking at there this sort of ties into dvfs so the dynamic voltage and frequency scaling the curve dvfs curve normally is again just because it's we it's been non-stop GPUs for six months so you'll excuse me hopefully for for drawing on that recent memory and knowledge but again with dbfs the frequents the curve basically your voltage and frequency is somewhat tied together and with risin we're looking at a lower dvfs where in theory you're going to get higher clock rates with a lower voltage or in the very least maybe not that's a direct higher clock rates too low voltage but in the very least you get higher performance per watt and the the boosting and down clocking functionality will just hopefully very aggressively switch depending on the context so you're not burning power where it doesn't need to be burned and that tells us that Andy is taking this time power consumption a lot more seriously than the older generation CPUs that have largely been dying out on the market so that's a that's a bigger deal this time that's a good thing because power consumption is a major deal I mean even just in mobile that's the only thing really that matters for a lot of mobile devices so that is good to see well we'll have to test it in house and see how it actually performs functionality we've kind of gone over everything that I actually know so far everything else is either going to be it's just going to have to wait till unbreached and we can do a more in-depth article on the architecture how it works without speculation release dates and prices obviously that's the big thing so reisen quarter won 20-17 still target launch and then Naples which is the server variant is still targeted for quarter two prices i do not have I they were not announced in the demonstration md compared their reisen chip which is eight cores 16 threads at 3.4 gigahertz against a broad well Egypt the i760 900k somewhat comparable in clock rate they compared them head-to-head that's an expensive Intel chip it's also eight core 16 threads and the performance with reisen was effectively identical in some tasks were shown to the i760 900k and those tasks include the handbrake CLI transcoding functions where they were more or less tied if not slightly better on verizon and then a render task with basically shading and rendering sort of a 3d modeling program where reisen and Intel through the broad well each ship were tied in terms of the time required to complete the task so it's good news we have to see again how it performs in house gaming we were not given any FPS metrics it was just basically like look it's a indeterminate how different that you can't see actually difference between the two that's not really useful because if you've locked the frequency to the display doesn't really tell us anything other than they can both do 60 Hertz which big deal no 11 surprised about that so we'll have to wait for gaming but we do have a baseline for the rest now I was always wait for our review and further discussion before making any purchasing decisions would say that every time but just worth putting out there patreon link official video gonna help us out directly link in the description below for more information subscribe for more comment with your thoughts I'll see you all next time you