Ryzen 7 5700G Memory Scaling Benchmark, DDR4-3000 vs DDR4-4400

[Music] welcome back to harvard unboxed now recently i checked out the ryzen 7 5700g which is currently only available in oem systems prior to its retail release next month so in order to get one early for testing we had to purchase an hp system and then had that shipped to australia so if you haven't watched our review then i suggest go checking that out first as today's video is a follow-up that looks more closely at memory performance a few harbour inbox members requested a memory scaling video and i felt this tied in pretty well with one of the biggest criticisms we saw from that review it seems a few viewers were upset by our choice in memory claiming that we were leaving a lot of igp performance on the table and some even went as far as to say we were deliberately gimping performance and this was despite us meeting the amd specification by using ddr4 3200. in fact the ddr4 3200 memory that was used generally outperforms less premium ddr4 3600 and 3800 kits as its low latency cl 14 samsung bdo memory which starts at around 130 us for a 16 gigabyte kit and that's quite a premium given the ddr4 cl 16 memory generally starts around 70 us so we're talking about almost twice the price not only that but we're using two kits which sees the memory operate in a dual ranked configuration and this is not to be confused with dual channel memory and if you want to learn more about this i do have a video specifically covering this topic and i'll link that in the video description but the point is we're using pretty expensive high quality ddr4 memory already so how much more performance are we really leaving on the table well today we're going to find out as i'm re-testing the igpu performance of the 5700g using a range of memory configurations but before we get into the data it is worth noting that we typically review cpus using ddr4 3200cl 14 memory as it's proven to be a pretty reliable indicator of sweet spot memory performance closely matching the performance of lower latency more affordable ddr4 3600 and 3800 memory moreover most of these current generation cpus officially support ddr4 3200 and there's no guarantee what kind of memory frequencies the imc will support beyond the official spec so for example showing ddr4 4000 performance by default could end up being very misleading if only half the chips in circulation are capable of operating at that frequency so with that out of the way let's talk about the testing for this comparison i am going to include the geforce gtx 1063 gigabyte as a reference simply because it can be purchased second hand along with a brand new core i5 10400 f processor for at the same price as the ryzen 750 700g now normally we don't compare new and used hardware this way but we're living in quite strange times now and it really only makes sense to buy secondhand for lower tier gpus that said if it were more normal times i feel things would be even rougher for the 5700g as we could use an 8 gigabyte rx 580 as these were selling for around 850 us back in early 2020 now for the memory i've tested half a dozen different kits some are single rank while others are dual rank and that information is listed on the graph along with the primary timings finally all were configured to run at the one-to-one ratio with the infinity fabric i also messed around with the memory allocation for the igpu often referred to in the bias as the uma buffer size and found as was the case with the 2400g 3400g it made no difference to performance even when allocating eight gigabytes okay let's get into the blue bar graphs starting with assassin's creed valhalla we see that the review configuration allowed for 33 fps on average and it turns out that's a pretty good result for the 5700g the best result i was able to produce using higher frequency memory was 36 fps so a nine percent improvement over the review configuration and it was achieved using g skills ddr4 4000 cl 16 memory using a dual rank configuration though be aware it does cost 220 us for a 32 gigabyte kit i also have a ddr4 4400 cl 16 kit from g skill which uses the same primary timings but it's single ranked and despite clocking 200 megahertz higher it was actually slower for this test offering just a single frame over the review kit freemium 3 performance uplift and this memory costs 150 us for a 16 gigabyte kit that said you can buy 16 gigabytes of ddr4 400 from g school for just 125 us when buying from their ripjaws v series that kit uses looser cl18 timings and as you can see here igp performance match that of our low latency ddr4 3200 review kit with just 33 fps on average so simply throwing high frequency memory at the ryzen 7 5700g won't solve the problem timings are important as well as is the memory rank i also tried a ddr4 3800 dual rank cl 18 kit and found performance was on par with the review kit then using single rank ddr4 3800 that dropped performance slightly while cl 18 ddr4 3200 single rank was slower again though we're only talking about a six percent decrease from the review configuration then running really cheap underspec ddr4 3000 memory drop performance by 9 when compared to our review configuration so it does appear as though the review configuration is a pretty good guide delivering comparable performance to reasonably affordable ddr4 3800 memory but let's move on and check out a few more games horizon zero dawn is certainly more favorable towards the higher clocked memory kits showing a 15 performance improvement with ddr4 4000 and 4400 configurations all of which appeared to max out the integrated vega graphics at 31 fps but again the review configuration does do a good job of mimicking the kind of performance you receive from affordable ddr4 3800 kits as we're looking at just an extra frame here or a four percent increase the margins between the majority of memory kits are much smaller when testing with watchdogs legion again the review configuration basically matched the ddr4 3800 kits with no more than one fps in it which amounts to a four percent variation meanwhile the single ranked ddr4 4400 kit was up to seven percent faster so a pretty small gain there for what really is quite expensive memory then at the top of the graph we have the very expensive ddr4 4000 kit which did offer a 14 increase over the review kit so a decent performance gain there but we're still only talking about 32fps 1080p using the lowest possible quality settings testing with f1 2020 shows up to a 14 performance increase using ddr4 4400 cl 16 memory which allowed for 67 fps on average though only a very small improvement one percent low performance over the review configuration then we're only looking at an eight percent improvement with the expensive ddr4 4000 memory while the 3800 kits were again comparable to the review configuration then for those of you using cheaper ddr4 3200 memory you can expect around a 7 reduction performance compared to what i showed in my review then finally we have tom clancy's rainbow six siege which was one of the few games tested that ran really well on the 5700g at 1080p and we got away with using the medium quality settings the review kit allowed for 65 fps on average and again that saw it match the higher clocked ddr4 3800 memory at best i was able to boost performance by a further 11 percent when using the more expensive ddr4 4000 memory while the 4400 kits were at best eight percent faster so pretty mild gains there so as you'd probably expect faster memory does help improve performance though it might not be to the degree that you'd hoped but given what we just saw i do feel even more confident about our choice and memory to test the ryzen 750 700g which is fortunate as it was kind of set in stone quite some time ago the premium ddr4 3200 memory that we used is certainly comparable to cl 18 ddr4 3800 and we found in the past that the same is also true for cl 16 3600 memory it's also not unrealistic in the sense that it's too good as the cheap 3200 kit was typically only four to six percent slower using the expensive ddr4 4000 memory we did show that you can often improve the performance of the igpu by around 15 which is a nice performance uplift though you will be paying significantly more than 15 for the faster memory that said you can tune cheaper memory which does help improve performance but it can be a complex process and achieving stability can be a real challenge which is why most people just leave memory timings alone it's also well worth noting that for those of you looking at using the 5700g for gaming and aren't constrained to a tiny mini pc pairing the apu with premium memory just doesn't make sense this is because the gtx 1060 will deliver the exact same performance shown in this video using ddr4 2666 memory as it's in no way cpu limited and doesn't require system memory to function well that is assuming that you don't flood the three gigabyte vrm buffer and you likely won't at 1080p using low quality settings at least in most games so again if you're just gaming and not trying to justify buying something like an asrock desk mini you're far better off ignoring the ryzen 750 700g and instead snapping up a core i5 10400 and pairing it with a cheap secondhand graphics card something like the three gigabyte gtx 1060 for example as that will net you over 100 more performance but if you're hell-bent on purchasing the ryzen 7 5700g when it hits shelves next month my advice is to buy some affordable ddr4 3600 or 3800 memory whatever the sweet spot happens to be in your region and that is going to do it for my 5700g memory scaling guide if you liked the video you know what to do you can also subscribe for more content we will be looking i guess officially at the 5600g and 5700g early next month when you know we can post those reviews when samples are provided by amd so there'll be more apu content coming in the near future so yeah if you want to see that make sure you are subscribed also you can join the harvard box community over at floatplane or patreon you will get access to our live stream that till myself do that's actually coming up i think next week or later this week depending on when this video goes soon that's coming up soon we also have q a stuff behind the scenes content and a really awesome discord server for harbour and box community members also right now we do have a second run of the harbor unavailable merch because it was very popular so we are doing a second run so if you want that make sure you get it because not sure there'll be a third run wasn't sure there'd be a second but who knows anyway that is going to do it for this video thank you for watching i'm your host steve i'll see you again next time [Music] you

Channel: Hardware Unboxed

Views: 180,501

Keywords: hardware unboxed

Id: mi4eAWWWxGc

Length: 11min 27sec (687 seconds)

Published: Thu Jul 22 2021