Gadgetory

and these financial and analyst day was hosted hours before this recording where the company revealed new information on Vega thread Ripper the CPU mobile and data center products and professional versions of Vega which will be due out in June potentially indicating further delays for the gaming market the target is firmly on the back of end videos P 100 chip for now and thread Ripper meanwhile is working out public demonstrations at Computex in a few weeks and we'll be going over all of that in a bit more in today's news video before getting to that this is brought to you by our patreon backers at patreon.com slash gamers Nexus a big thanks to those of you on patreon who are in discord and were able to help us with covering of this news story as rely on a lot of you to grab screen jobs during the livestream event let's start with Vega as its of the greatest interest lately and these Vega video card bears the FE badge so this is the first vega card that was shown it was at their analyst and financial day that was streamed earlier today as of recording anyway and the fe card not found as edition it is the frontier edition so Vega frontier Edition is the first one that we know about the card is aimed at data scientists it is not gaming targeted and as such they were talking big numbers primarily in things like deep learning which is a big topic lately especially with the v1 hundreds announcement just recently so what we know thus far of the Vega fe card is that it's got roughly a two-to-one FP 16 ratio to FP 16 to 1 FP 32 we're at 25 teraflops FP 16 compute power that puts us somewhere around 13 teraflops based on what AMD was saying of FP 32 or single-precision computes FP 64 or double-precision was not revealed at a time of the analyst meeting so we don't know where that is today but they did show us some slides of performance using deep bench which is a Baidu benchmark for testing neural network capabilities of video cards and graphics and computing hardware including Intel's Knights landing and videos P 100 is on the chart that P 100 accelerator and then Andy put their own Vega F II card on the charts as well andy claims that Vega will outpace n videos 2016 P 100 accelerated GPU and using the deep benchmark for machine learning positioning at Vega at 88 milliseconds completion time versus 133 milliseconds completion time for the P 100 this is the only test of its kinds at Andy showcase thus far so we can't draw definitive conclusions from the data but it is the first real look at performance we've been given for Vega outside of the initial CES demos that were pretty limited and just restricted to gaming at 60fps the V 100 of course isn't present as it won't ship until quarter three and also note that of this deep bench test result rajma Kaduri said quote I'm not declaring victory here over Nvidia or anyone else this is basically stating that Andy is finally on the chart a year ago we would have taken as being on the previous charge we are on the chart with a fairly impressive number so that was their goal is getting something out there that can do deep learning or some kind of machine at processing with regard to gaming there were two tests that Andy demonstrated at the event with the Vega card they had they were doing a VRAM thrashing scenario where it was simulating exceeding vram capacity by limiting the capacity on the vega card so they took what would normally have a 16 gigabyte HBC of HB m2 memory on this particular model of Vega GPU they were showing and they restricted that to two gigabytes then ran an HPC a/b test were they disabled HPC or high bandwidth cache in one test and they enable in another if you don't know what HPC is we talked about it in our CES news video when we first heard about HPC and got some more details on it but the idea was to toggle it on versus off to see what happens in a scenario where you've exceeded the vram capacity of the video card the other test was with Sniper Elite for that song we're going to start with Sniper Elite seems to be a bit of a favorite benchmark for AMD lately it shows an v cards performing pretty well and at all of our tests like the recent crossfire test we did where there's almost 100% scaling in that particular game and even the DX 12 test where we look at g4 560 bottleneck you know GPUs so they use that for this benchmark as well benchmark was really just a demonstration and it was the game running 4k with undisclosed graphic settings we don't know if it was ultra or high or what so we can't really run those tests on our own hardware and try and guess at where Vega will be and other than that we saw a range of 60 to 63 FPS on the screen with no further information was about a 20-second long demo if that where they walked on the beach climbed a ladder and stabbed the guy in the heart so that was the Vega demo and in the latter climbing part of the demo the FPS spiked over 70 briefly 62 63 FPS 4k undisclosed graphic settings tested in a vacuum so we don't really know where it compares to Nvidia hardware or even other ante hardware right now but based on what was previously demonstrated at CES where Andy showed Vega competing with the GTX 1080 in a 60fps locked scenario it would be reasonable to assume that Vega is probably still targeting something like a GTX 1080 rather than a 1080 ti4 performance at least with regard to where it's competing on the market but that is entering the realm of speculation so we'll go on to the next test which was the demonstration using HPC toggling and this test was the simulation of VRAM thrashing by limiting Vega at least the one on display at ajust 2 gigabytes of its high bandwidth cache which was done because games don't use anything close to the total allowance on Vega currently Andy toggle the HPC on and off of the 2 gigabyte limits showing the performance bumps from roughly 56 FPS average to roughly 72 FPS average and rise of the Tomb Raider specifically in this mirror I'm thrashing scenario this was a simulation it's not real world and it's a chosen slide for an analyst meeting so we don't have a whole lot of depth yet but this gives an idea as to what the company is hoping to do with its high bandwidth cache and the theory is that HPC helps in vram limited scenarios should one present itself in a way that would speed up performance because you get the extra speed and the way that the storage actually works at a little lower level which is something we've discussed in the past and we'll discuss again soon once we have more white papers on Vega 4k 60fps is still the target that's the target that was set two years ago that's what they're hoping to do with Vega so it looks like that's probably going to be met and is something that was line it up in competition with the GTX tena idiot as Andy previously positioned the card at the CES meeting most of the presentation was focused on professional applications that are both Verizon and for Vega on the Vega side this means a lot of talk about the new SSG which will bear the GPU on it and they also talked about different deep learning capabilities data center type stuff so what does this mean for gaming well from what it looks like right now the professional data center targeted Vega card should begin shipping sometime in June that's what AMD was saying on stage so with June as a target for the professional audience that would be again data center enterprise data scientist and anyone doing heavy compute stuff with that targeted in June it's probably looking like gaming will hit in July or August gaming tends to follow professional deployments of new architectures so that would be reasonable to assume but again that's speculation so just a note as far as the SSG line we talked with Rajat Kaduri previously about the Radeon SSG we sat down with them at the A&E offices probably late last year and talked about that the new one is carrying Vega for the GPU it is CAD targeted and will work in other applications like CAD and the idea is that it's basically got a two terabyte SSD strapped to it and that will store all the data for large project files like massive CAD files with a lot of individual moving parts so you store that on the GPU it is non-volatile memory it's just like an SSD but it's very close to the GPU core so it can access quickly and hopefully load up those files faster so that's something that has existed but now it's got Vega on it as well and it's something that they were showing at the event with this demonstration Andy was also showing an 8k video premiere test or demonstration where they basically had a premiere window open Adobe Premiere that is and a live timeline scrubbing test with 8k video using the Radeon 2 terabyte SSD with Vega and it was fairly smooth but we don't have any actual numbers for you other than to say things like fairly smooth and maybe show some of the footage that we captured of the event but that's all we've got for now on Vega other than Vega mobile so Andy also discussed mobile deployments of their rise in CPS at the event which will be accompanied by Vega integrated graphics chips in the near future and he says that these integrated Vega graphics cores will run a generational improvement of about 40% of the preview I GP solution by Andy and claims a 50% power reduction over the previous generational version risin mobile will include two and ones for folks interested in that market and we'll get a MD back into laptops in a bigger way in theory and there was some quick discussion on thread Ripper there's a lot on epoch EP why see the other one we'll talk about that moment thread Ripper was briefly mentioned it's a 16 core 32 thread setup it is an h EDT market solution so high-end desktop that would be probably competing with Intel's rumored I 9 lineup that we don't obviously have confirmation yet especially from Intel side of things but that would probably be where it's targeting because the current horizon CPS already battle with Intel pretty well in h EDT but thread Ripper will go a bit beyond that it will be an expensive CPU comparatively versus what Andy is already shipped 16 cores 32 threads first live demos will be at combi txt this year we will be there recovering Computex in about two weeks so if you haven't subscribed already do it so you can see our Computex coverage and will be covered in thread Ripper at that point vague information on thread Ripper that they gave at this event where that thread Ripper will expand memory bandwidth expand i/o bandwidth and will focus on production level and professional applications but that's about the most we've got right now other than all the other rumors that have been out there for a while and the last piece of news was on that epic CPU that I mentioned so epic ep YC is the new name for Naples Naples was the code name for the server architecture that we first learned about at the Rison events d4 aizen's launch naples is a 32 core 64 thread CPU and epic is the official name of it so the code names were as on the roadmap Naples we learned today that the next code name is Rome and then after that is Milan so they're going with an Italian influence there so Naples is coming out soon in the form of epic they are targeting Intel and their 95 percent market share of x86 servers that's where AMD hopes to compete they do have an x86 license they're one of the few people who can actually compete in the server division and takes on that market share from Intel so that's where epic is targeting they had some initial demonstrations including Linux in code and compile passes and the compile paths that we saw in the live demo with the analyst meeting position dandy matched against intel's a five to six nine nine a version for which is important signifier that cpu the intel one was at twenty two point five seconds and the and the epic cpu for the linux compile Sam test was fifteen point seven seconds to complete so that's looking good at least for that one test the tests were done in dual socket configurations for those particular numbers so two and two and they also had a later test with epic in a single socket configuration versus a different Intel CPU it was the e5 two sides two six five zero version four thank you in self making that easy so it was burst his that in a dual socket two of those versus one epic CPU showing a complete compile time of thirty three point seven seconds on the and the epic CPU and thirty seven point two seconds on the Intel i5 CPU final notes here Andy is finally providing some roadmap so again it is Rome and Milan next those will be on Zen two and Zen three respectively they will be on seven nanometer and seven nanometer plus respectively which is a nomenclature we've seen lately from Intel as well as Nvidia and now AMD and that's really what it looks like for the whole roundup that's Andy's whole the whole press conference kind of or analysts and financial conference recap quickly epic will ship before end of June big is looking like June for the professional market and later probably for gaming maybe July or August and other than that discussion for the name epic and other name is discussed here can commence below in the comments as always you can subscribe for more go to patreon.com/scishow busy thank you to those of you in discord for joining in to watch the same life as a lot of fun if you want to join in next time there's an event like this you can go to that patreon.com slash gamers nexus site to learn more thanks for watching i'll see you all next time you you