Gadgetory

welcome back to harbor unboxed last week we've got our first look at am reason Raven Ridge desktop AP uses a m-- D release the risin 5 2400 G and rise in 322 energy and both were very impressive despite a grueling 4 day grind doing nothing but working on our coverage there's still loads of things that I'd like to cover in much more detail such as overclocking and of course just testing more games since publishing our review there's been a heap of questions asked about these new Raven Ridge ap used but the one I've been spotting probably the most frequently has to do with memory allocation for the Vega GPU now unlike a typical discrete graphics card like the Vegas 64 model that I have here most integrated solutions don't have their own dedicated memory so if we take this Vegas 64 graphics card as an example it has 8 gigabytes of dedicated memory so an inking about memory buffer just for the GPU where it can store data and then access it quickly when it needs to because this is a high-end graphics card it not only has a rather large a gigabyte buffer but the bus it uses to access the memory is very fast HBM to provides a 2048 bit wide memory bus and with the memory clocked at 1.2 5 gigahertz this allows for a bandwidth of 483 gigabytes per second further down the food chain like at the bottom you'll find graphics cards such as the rx 550 because the compute performance is around 9 times lower it doesn't require an 8 gigabyte memory buffer either in fact the rx 550 would be flat-out taking advantage of half that capacity and in today's games really works just as well with a 2 gigabyte buffer regardless of capacity though it uses memory clocked at 1.75 here you hertz and with a 128-bit wide bus can be fed down at 112 gigabytes per second so a typical rx 550 has a 2 gigabyte memory buffer and that allows it to shift data in and out of the memory at a theoretical peak bandwidth of 112 gigabytes per second now if the game requires 3 gigabytes of vram but you only have two year bytes some game assets then spill over into the system memory this is often referred to as Ram I covered this in a bit more detail in a video earlier this year that investigated how much RAM gamers need basically shifting data in and out of memory is significantly slowed up the Raven Ridge GPUs for example a limited to a memory bound with them around 35 gigabytes per second that's for a system equipped with dual channel ddr3 200 memory so in the case of the rx 550 when accessing data locally using the onboard via Ram it has a bandwidth of 112 gigabytes per second or a theoretical peak bandwidth but when accessing from the system memory it's limited to around 35 gigabytes per second so it takes at least three times longer to process the same amount of data we're moving through the system memory however if your computer then runs out of system memory game assets a move to the local storage device this means your hard drive or hopefully an SSD and depending on how fast that devices and how heavily its it with data it's at this point that you'll very likely see a noticeable dip in framerate as the bandwidth is now reduced to around 500 megabytes per second with an SSD if you're lucky keeping all that information in the back of the old noggin the Vega GPU integrated into the Raven Ridge apu has no local memory to speak of now some integrated GPU is like the Vega M graphics that'll be found in the new Intel KB Lake G processors they do have their own dedicated memory and this greatly enhances performance but it also greatly increases the cost as well since AMD's Raven Ridge ap user budget solutions it wasn't going to be possible to include HBM to memory and it simply doesn't make sense at this point in time so with no dedicated vram they use system memory exclusively and this means they are restricted to our bandwidth of around 35 gigabytes per second with the system using dual channel ddr3 200 memory of course bandwidth is just part of the issue here memory capacity also plays a key role a system with a base model rx 550 graphics card for example in a gigabytes of ddr4 memory effectively has 10 gigabytes of total memory to play with but when using vega 8 or 11 graphics on a raven or GPU you now just have 8 gigabytes of memory to share between the CPU and GPU generally speaking we notice does a very good job of managing memory and prioritizing applications for best results that said at least some portion of your system memory will be partitioned and allocated to the integrated graphics the Raven Ridge ap used for example use a method called unified memory architecture or luma for short right now most am for motherboards let you set them resize somewhere between 64 megabytes and 2 gigabytes depending on the size you select will determine the amount of system memory that is allocated exclusively to the vago graphics that is to say once allocated it can only be used as a graphics memory and therefore can no longer be accessible by the operating system or the applications that run on it now this is where I'm seeing a bit of confusion and misinformation some people are claiming that for best results reviewers need to test with the frame buffer set to the maximum possible size and right now that's 2 gigabytes however this isn't necessarily true and in fact most of you watching this video who have built or plan on building their own Raven rich system will want to do the complete opposite you're actually far better off selecting the absolute minimum amount of memory that you can allocate to the GPU because as I said once you do allocate that portion of your system memory to the graphics processor that's all that can be used for for example when in Windows doing things that don't really require much video memory at all you just have a significant chunk of your memory they're partitioned off that you can't use so if you know select a 2 gigabyte buffer for example on a system that has 8 gigabytes of ddr4 memory that would mean that you now only have 6 gigabytes of system memory available but you're a gamer and you want the maximum gaming performance possible so just set it to the highest value for those sweet sweet frame rates well know again you don't want to do that as I discussed earlier when using a discreet graphics card once the graphics memory or via Ram fills up the game assets are just loaded into the system memory in the case of the Raven Ridge GPUs we're using system memory exclusively so regardless of whether you allocate 64 megabytes or 2 gigabytes it doesn't really matter for example if a game requires 2 gigabytes of video memory but you've only allocated 64 megabytes usage to spills over into the shared memory and because it's all the same memory the bandwidth remains the same and therefore within reason so too does the performance as I said windows manages this very well so by allocating 2 gigabytes you're just restricting the operating system's ability to better manage the system memory when trying to make sense of why AMD was offering a 1 gigabyte or 2 gigabyte frame buffer I had initially thought that maybe reserving a certain amount of memory so 2u bytes would ensure maximum gaming performance as the operating system did have to shuffle around things especially when using just a gigabytes of RAM a poster you know 16 gigabytes or more however after testing various configurations I found this had little to no impact when gaming are certainly nothing you'd actually notice using both 8 gigabytes and 16 gigabytes of dual channel ddr3 200 memory with the exact same timings I found our real difference in performance between reserving 64 megabytes or 2 gigabytes of system memory for example I did test half a dozen different modern titles that all called for around 2 to 3 gigabytes of vram at 1080 P using the low to medium quality settings but rather than look at the half a dozen tiles that I've tested all of which showed the exact same thing I'm just showing the battlefield one results at 720p and 1080p along with some additional testing I did with Metro last light as you can see with battlefield 1 all the results are within the margin of error for the 3 run average and we're not just talking about the average frame rate but also the frame time performance as well I've heard a few reports that Metro last light of all games saw a massive performance uplift when going from 512 megabytes to 2 gigabytes I stopped testing with this title about three years ago now so I thought that would be odd if true but decided to check it out anyway as you might expect from this rather old game it does use very little memory and we saw no difference in performance when comparing the average frame rate at 720p and 1080p and the same is true for the frame time results as well given what we discussed previously before getting into all the results this probably shouldn't be that surprising regardless of whether the Vega GPU is accessing data via the allocated memory or not it's still using the same system memory and therefore it's limited the same bandwidth and in our case that was around 35 gigabytes per second in fact we can look at this a little more closely using the ADA 64 GPGPU benchmark tool we can measure read and write performance between the CPU and GPU effectively measuring the performance the GPU can move data in and out of its own device memory into the system memory this is called device to host bandwidth but more importantly and this is what we're going to focus on you can also look at copy performance this test measures the performance of the GPU memory by copying data from its own device memory to another place within the same device memory so in the case of the rx 550 that would be the onboard gddr5 memory but in the case of the Raven Ridge use that is our ddr4 system memory here we can see that was 64 megabytes of RAM allocated to the Vega at GPU and the 2200 G we have a throughput of 33 point 4 gigabytes per second when copying data from within the system memory and that's pretty much in line with a 35 gigabytes per second the CPU cores have when accessing the dual channel ddr3 200 memory if we increase the allocation size to 2 gigabytes this has no impact on bandwidth in fact based on an average of 3 runs we saw a slight decrease but of course these results are within the margin of error given how long this test takes to run it's safe to assume we're transferring well over 2 gigabytes of data so we're not just benchmarking within the allocated buffer then if we look at the rx 550 which as I've mentioned many times in this video has a theoretical peak bandwidth of 112 gigabytes per second we can see that in this test it's good for 88 gigabytes per second then for comparison I tried out Vega 56 and we hit 321 gigabytes per second and this model has a theoretical peak bandwidth of 410 gigabytes per second so all the evidence suggests that setting the I GPU allocated memory buffer beyond 64 megabytes is pointless and on systems with limited Ram even a bit foolish that said while I've tested quite a few games and applications now I haven't tested all of them of course I'm just not able to simply aren't enough hours in the day the only reason I can think of why you might want to increase the reserve memory buffer is for games that need or detect a certain amount of VRAM before they'll actually load we've seen this in the past with games that have these built-in safeguards that won't allow you to load them without meeting a minimum hardware spec it's very annoying as the game developer isn't really saving anyone from anything certainly no real harm instead they're just inconvenience gamers that probably have acceptable hardware but they're just waiting on a driver update to improve detection I expect that AMD will always deliver new drivers to solve these issues if and when they arise but we might see situations where in the meantime gamers can increase the allocation to meet the vram requirements and that will allow them to load the game so short of any potential compatibility issues I can't really think of any reasons why you'd want to sacrifice more than 64 megabytes of memory to the GPU but maybe you guys have some ideas that I have thought of also please note 64 megabytes might be an extreme example maybe err on the safe side and set it to 512 megabytes or you could just go with 64 megabytes like I am and wait till you run into some kind of issue and if you do run into an issue please let me know about it because I would like to investigate if you do happen to run into an issue all you have to do is reset the system increase the allocation size and then boot back into Windows and you should be right since most APU users will be using 2 4 gigabyte memory modules for an 8 gigabyte capacity especially those buying the incredibly good value risin 320 200 G for a hundred and ten dollars u.s. those users will want to save as much memory as possible and by telling them to lop off one or two gigabytes for the vaguer GPU seems like really bad advice based on my findings so finally circling all the way back to when I was discussing trying to make sense of why AMD was offering up to a two gigabyte frame buffer and there's even been talk that AMD is pushing board partners to offer a 4 gigabyte option so I guess I have to ask the question why I am D why give you customers the ability to degrade their experience on your very impressive Raven Ridge ap use well after a bit of thinking I think your answer to this question is probably quite obvious it just has to do with marketing aim D has to play the game and that's the game of numbers big numbers bigger is always better for example 2 gigabytes or up to 2 gigabytes of graphics memory sounds a wee bit better than 64 megabytes and that said AMD doesn't actually appear at this stage to be advertising the vega 8 or vega 11 memory spec and maybe once they push for that 4 gigabyte allocation we'll start to see this pop up don't get me wrong though it's certainly nice to have this option should you need it for whatever reason one of those reasons I haven't yet thought of but in my opinion aim D really needs to educate buyers on the issue and probably reviewers as well so we can inform you guys on which settings you should be using and why I Raven Ridge is far more appealing in the low end if you can have almost all 8 gigabytes of your system memory when using non gaming applications in Windows rather than just being fixed with 6 gigabytes for example and that's where I'm going to end today's video if you enjoyed the video please hit the like button subscribe for more content if you appreciate the testing we do here we're unboxing consider supporting us on patreon where you'll gain access to our discord chat and monthly livestream thanks for watching I'm your host Steve see you again next time