Gadgetory

welcome back to harbor unboxed now normally on the channel we provide buying advice we try to steer our viewers towards the best buy whether that be the best $200 graphics card or the best to be 450 motherboard we try to ensure that you're getting the most bang for your buck today's video though isn't buying advice it's more of a facets type test if you will back when the GTX 1080 was first released many viewers claimed that it was probably memory starved of memory balance starved so I decided to investigate despite stellar performance the claim seemed reasonable enough after all the previous generation gtx 980ti supported a 384 bit wide memory bus and when coupled with 7 gigabits per second gddr5 memory packed a peak bandwidth of 336 gigabytes per second the GT x 1080 was limited to a bandwidth of 320 gigabytes per second despite using the fastest gddr5 x a memory available at the time and this allowed in video to get away with using a narrower 256 bit wide memory bus still given the gtx 1080 sported vastly superior shader power to that of the gtx 980ti it seemed reasonable to question how a reduction in memory bandwidth would impact performance and this is why many were suggesting a memory bottleneck could be an issue anyway long story short it turns out Nvidia knew exactly what they were doing here and the GT X 1080 had enough memory bandwidth to achieve maximum performance and we worked this out simply by underclocking the memory while leaving the core frequency alone still I found that testing interesting and although it wasn't a particularly popular video I've decided to do it again with the new GeForce RT X series again I'm not expecting this video to be all that popular but I think our core audience will appreciate the testing generally if I find this stuff interesting you guys do as well which is nice as we get to create the content that we enjoy anyway the focus of my testing has been on the new flagship model the RT X xx atti using gigabytes gaming OSI model but I've also done some testing with gigabytes RT X 2080 gaming OC and the MSI RT X 2070 armor OC so having said all that let's check the results out first up we have Assassin's Creed origins stock out of the box the 28th eti average 60 for FPS at 4k and we see the same 64 FPS when overclocking the memory to a transfer speed of 15 a half gigabits per second under clocked to 13 gigabits per second we do see a single frame dropped then two more when going to 12 and a half gigabits per second and then a further two frames at 12 gigabits per second and that's as far as we could under clock the GDD are 6 memory we find slightly different scaling results with battlefield 1 we're also dealing with much higher frame rates here as well right away dropping down to just 13 and 1/2 gigabytes per second results in a single frame lost and this data is based on a 6 run average so while 1 FPS is still within the margin of error we consistently saw a single frame dropped with the slower memory interestingly as the memory speed is wound down we don't see the average frame rate impacted that heavily but we do see a rather bigger hit to the frame time performance when going from the stock 14 gigabits per second configuration down to 12 gigabits per second we see a 14% hit for the one percent load but just a 6% decrease for the average frame rate meanwhile overclocking the memory didn't improve frame time performance but it did increase the average frame rate by 2 fps again we see a bigger hit to the 1% lower result than the average frame rate went under clocking the memory this time when testing with far cry 5 that said the discrepancy between the average frame rate and 1 percent low isn't as extreme as we saw previously with battlefield 1 interestingly overclocking the memory had a much bigger impact on the 1% low performance boosting the minimum frame rate by 3 FPS the memory scaling performance seen when testing with strange Brigade is similar to what we've seen in the previous two tests performance isn't degraded too heavily from 14 bits per second to 13 gigabytes per second but after that performance falls off a cliff particularly the frame time results overclocking the memory was surprisingly beneficial to both the frame time and average frame rate performance the last game are going to look at is shadow of the Tomb Raider here we see no loss in performance with the 13 and 1/2 gigabits per second memory and almost no change when dropping to 13 gear bits per second however beyond that performance does start to fall away quite substantially overclocking the memory to 50 and a half gigabits of a second also Nets us a few extra frames running shut off the Tomb Raider once again but this time with the RT X 2080 the non TI model we see a pretty minor but still consistent drop-off in performance with each underclock step the frame time and average frame rate performance is impacted fairly evenly that said we only see an extra frame for an 11% increase in memory throughput with a frame dropped for a four percent bandwidth decrease then I rerun shut off the Tomb Raider again with the RT X 2070 and we see no real change performance by overclocking the memory and then when we down clock at two thirteen and a half gigabits per second we also see no change below that though we do see a frame dropped every point five gigabits per second reduction in throughput for the last test conducted I reinstalled the r-tx 2082 I decided to run shut off the Tomb Raider with some core overclocking testing with 13 and 15 a half gigabits per second memory configurations to my surprise despite the memory bandwidth bottleneck overclocking the cores to maintain a clock speed of 2,000 50 megahertz the 2082 I jumped ahead of the stock core configuration using 15 and a half gigabits per second memory then with the memory boosted to 15 a half gigabits per second with the cores overclocked we saw a 6 percent increase in frame rate performance opposed to overclocking them at all that's it with the cores overclocked we only see a 5 percent increase in performance for what is an almost 20% increase in memory throughput there we are again seeing a situation where the improvement for the frame time result is less significant and this is primarily where we are seeing the slower memory limit performance some interesting results there and as we expected the new GeForce r-tx series very much needs the support of GDD are 6 memory the 14 gigabits per second stuff now for the most part that proved optimal though we did see a few instances where the factory overclocked model from gigabyte their gaming OC card that did work a bit better with the fifteen point five gigabits per second memory we saw some nice boost to the frame time performance in games such as far cry 5 and strange Brigade now that said other titles such as Assassin's Creed origins and battlefield 1 proved that the 14 gigabits per second spec is optimal on average we saw a 13% drop in frame time performance when going from the stock 14 gear bits per second memory down to 12 gigabits per second and that is in line with the 14% decrease in throughput given that gddr5 X memory was never SPECT higher than 11 gigabits per second there's just no way in video would have been able to use that memory with a 352 bit wide memory bus for the RT X xx atti or a 256 bit wide bus for the 2080 and 2070 in fact had they used 11 gigabits per second gddr5 X memory the 20 ATT I would be limited to a memory bandwidth of 484 gigabytes per second and that's 21% less bandwidth than it actually has it also be 8% less bandwidth than the 12 gigabits per second configuration that we tested in this video using 14 gigabits per second GDD r 6 memory the RT X xx atti has a memory bandwidth of 616 gigabytes per second that's the peak memory bandwidth now this figure is worked out by multiplying the memory clock rate which is 17 hundred and 50 megahertz by the 352 bit wide memory bus so 1750 x 352 and then divide that number by 8 which changes the figure from bit to bite then multiply that figure by the memory clock multiply which 4g ddr 6 is 8 and that gives you the memory bandwidth in megabytes per second therefore in order for video to achieve the same memory bandwidth using 11 gigabits per second gddr5 X memory they would need to make the memory bus around 30% larger and with the chirring die already measuring a massive 754 millimeter squared they simply couldn't afford to waste the silicon real estate so that's that really like I said this was just a science type test not really buying advise of any description I just found it interesting and I hope you guys did as well if you did enjoy the video well be sure to subscribe if you're not already because we do this kind of content from time to time and if you appreciate the web widget of our box then consider supporting us on patreon because as I noted earlier we really just do these types of testing for you guys I don't generate a whole lot of attention for the channel I don't expect you had too many views on this one it's just something that yeah our core audience seems to enjoy and if you'd like to request other tests like this some sort of for science type test then consider joining us on patreon because you will gain access to our private discord chat and that's generally where passionate viewers requests these types of tests anyway thank you for watching I am your host Steve and I will see you again next time you