Gadgetory

clock for clock we wanted to know the difference between shader count and vegam 56 and Vega 64 where 56 host 3584 shaders and 64 host 4096 that's a 14 percent increase in shader count to be 64 but the gains aren't linear we were able to cross Vega 64 reference performance with our heavily modded Vega 56 card and that was done using power play tables and liquid to jump to 17 42 megahertz clock speeds by going less crazy about overclocking and limiting clocks to match speeds we can reveal the importance of shader count or lack thereof before that this coverage is brought to you by the core G 21 enclosure from thermal take a $70 case with 2 4 millimeter thick tempered glass side panels and a power supply shroud with top mounted SSD sleds learn more at the link in the description below and given that we now have about 1% of the Vega 56 cards that have shipped to the US there's another one in the test bench in the other room it seems like a good time to run this test so the biggest issue with the benchmark is that AMD's newest cards vega cards use a similar boost functionality to Nvidia's pascal line so it's it's more or less equivalent to boost 3.0 and that means that the clock speed jumps up and down based on various parameters those include power thermals voltage things like that and basically determines on its own if it has enough Headroom and either one in any of those parameters to boost clock further or if it needs to drop clock down so that makes it difficult to do a clock for clock comparison but we can control for it if long as you're careful and measure everything while testing things like that so the big pin here unfortunately that is a difficulty the upside is that theoretically these boost parameters with Andy and NVIDIA mean that out of box without any overclocking or any user interaction whatsoever the card should more or less be performing at their maximum potential within all of the various parameters defined for safe operation so that's the idea of it but it does make the job of overclocking more difficult so we have controlled for that and we've solved for power Headroom Thermal headroom and more or less voltage we ended up clocking up to around 1590 megahertz on both cards for that frequency bounced around based on the game with vegam 56 and 64 cards we applied power offsets of 80% with power play tables just to ensure there was no power limit issue and then we imposed a 1.2 v core which was checked on the back of the card with a digital multimeter HBM two speeds were set to nine forty five megahertz on each card meaning that we've brought the V 56 up to be 64 stock memory speeds and so we're really just looking at two cards that have the same speeds the same voltage very similar power limits and the only difference now is the shader count will show some frequency comparisons on the screen now while talking through this next part we ended up retesting this three different times thanks to the new clock behavior in Vega the final two tests logs frequency for each test and each game which was new for us and then we checked those averaged frequencies and frequency over time for each test this allowed us to better determine if the clocks were actually roughly matched or if one card had more boosting Headroom than the other obviously we want as little movement as possible in the clock the Vega 64 card often needed to be set to around 1632 megahertz to get 1590 megahertz in some games for example which can make things confusing and other games like ashes would be was higher than we configured so the only solution was logging and manually checking it we ultimately ended up with a range of about 2 to 10 megahertz difference for most tests generally sticking around 2 megahertz and this in mind our margin of error is going to be a little wider than previously with perfectly matched speeds but to be fair 2 megahertz is generally 0.1% different so not a big deal the biggest deal is that if it boosts higher and one scenario than another you might see that reflected but the best solution is just to control and testing some other games we're a bit more variable than this but we've got all that data let's start with 3dmark firestrike here's the clock comparison chart for 3dmark fire strike at 1080p as seen here we're roughly equal in clocks averaging the frequency during actual tests though not between tests that doesn't matter we ended up at 1580 5.6 megahertz for Vega 64 and 50 83.5 for bacon 56 or 0.1% offset that's close enough to be effectively the same averaged our graphics score for Vegas 64 lands at 2276 one with v56 at 22 7 to 4 and that's a difference of 0.1 6% which coincides with our clock speed deficit on Vega 56 and as well within margin of test variance on 3dmark applications the scores are effectively the same between B 56 and V 64 when matching clock speeds showing no advantage for higher shader count on Vega 64 to put those scores into perspective here are the FPS scores 100 11.3 for Vega 64 100 and 11.2 7 for bacon 56 looking at GT 1 GT 2 puts us at 89 9 and 80 8.854 64 and 56 respectively though it doesn't really matter which order they are at this point just to make sure there's not some major advantage in higher resolutions we tested a few games at both 1080p and 4k here's the clock chart for for honor where we see effectively equal clocks the average clock for Vega 64 was fifteen seventy eight five megahertz with bacon 56 at 1580 dot four megahertz sticking to our 0.1% difference at 4k we fall within our error margins and tested test variance they can fit these six technically plots higher but this is a statistically insignificant difference that lead could just be because Vega 56 happened to hit a higher clock and a more abusive scene or maybe something slightly different happened on screen during that particular test pass there's no statistically significant difference in for honor at 4k between these two cards when they are clock matched at 1080p we see similarly in significant differences we're at roughly 137 FPS average for each GPU the shaders provide no clear advantage in any of our three measured metrics hell-blade at 4k and very high had both the vega 56 and vega 64 cards at around 34 FPS average with low is similarly matched in this particular test there is not an advantage from increased shader count from the way we tested or measured at 1080p we see the same we're at about 81 FPS average for each GPU so once again no real difference from the shader count ashes of the singularity with dx12 and GPU focus testing lands an average V 64 frequency of 1600 six megahertz with v56 averaging 15 98.9 megahertz so about one different close enough measured performance has vega 64 at 58 FPS average with vega 56 at 56.7 FPS average this is close enough to be within our error tolerances for this particular test especially because Ashes does kind of jump around a bit and framerate lows are largely the same so no real difference here Ghost Recon wildlands at 4k had Vega 56 and 64 both operating at 42 FPS average with lows between 37 and 38 1% and roughly between 36 and 37 fps 0.1% low is 1080p wasn't much different we saw largely the same performance once again with this test as with all the others there might be applications where the shader difference is more noticeable but it's not in any of these games maybe in some kind of production application or something we didn't catch there's a possibility there's a bigger advantage there but for the games we've tested which serve as an analog for all games hopefully although there are always outliers we controlled for a frequency on the core controlled for frequency on HBM controlled for thermals for power target controlled for voltages and checked carefully on the back of the card and performance in gaming is effectively the same once we equalize for all of those things and we only have the one difference being shader count or CU count if you prefer so this further sort of puts the nail in the coffin for Vega 64 for gaming as we've said since the start Vega 56 is more or less well so here's the thing with the pricing situation now what I'm going to do is rather than reference any specific price numbers we're gonna just reference them relative to each other if they get 56 can stick near the 10 70s price it has been and still is a pretty good buy it is and these strongest argument for the Vega line 56 competes rather directly with the 1070 where 64 just doesn't make a lot of sense in terms of value and the biggest downside is the boosted power consumption on 56 versus the 1070 so that's not of concern to you there's not a lot of a downside there if you are looking at overclocking 56 definitely is worth buying over sixty-four there's one potential downside which is that there might be a case where Vega 56 is binned is a lower bin of HBM than 64 in that case you would be more limited but so far I've only heard of one card personally that can't reach 945 mega HBM to speed and that one was hitting 930 but in theory you should be okay with the ones we've tested it boosts high enough with an overclock maybe not as high as 64 but it's still pretty good and either way you can hit 64 performance ignoring the shader difference with gaming that we've tested so looking good there the only real reason Vega 56 would underperform versus Vega 64 is the power limit 56 comes with a lower TDP allowance in its stock bios than Vega 64 which is Andy's way of limiting the power permitted to the core and the HBM and everything so your overclocks will be more limited on 56 than 64 however that can be solved you can solve it with a registry table or as we've recently learned you can solve it by flashing a be 64 air BIOS onto a V 56 and as long as you don't modify that BIOS it'll work it won't work if you make any modifications it won't work if you take VF EE but 64 on 56 works and the most critical thing there is that it unlocks a higher power limit so you can do everything you want to deal with 56 within whatever Silicon lottery may apply so as for the shaders it looks like there's not a big difference in the games be tested again there might be an application where it matters but it's not in these games and these games largely serve as an analogue for other games so fair to say Vega 56 remains a far better buy than Vega 64 and remains competitive with a 1070 particularly if you actually tap into its overclocking head room and start doing some modding so as always thank you for watching you go to patreon.com/scishow and exit stops out directly or a gamer taxes dotnet for the full article subscribe for more we'll see you all next time you