Gadgetory

twenty-seven hours is how much time we had to review and these are active eiga 56 their direct competition to the gtx 1070 by nvidia the busy six is a $400 card we're looking at it from a lot of angles today including under volt in overclocking BIOS modding HBM vs core overclocking and we'll try to get into thermals power consumption and gaming performance though some of that may have to be saved for a separate video because we're about nine hours away from embargo lip before getting to that this video is brought to you by synergy the software that lets you share a keyboard and mouse between multiple systems if you have limited desk space and multiple computers to command that synergy removes the need for separate peripherals or a KVM and works as over the network software use our link below to get 50% off the basic or pro version we won't be going over any of the very basics today if you need pricing information specs things like that check the article below or check other content creation outlets we're focusing on data because that's what we have time to do so first off overclocking notes on it and issues we encountered Vega 56 was very exciting to us when it was first announced specifically because as a $400 card with these specifications it carried and with what we knew about EFI 56 looks like it could overclock to outperform a Vega 64 card or at least perform equally if you know what you're doing or even if you don't you just kind of hack at it but we were let down or if you know builds IDE we wanted to set forth turn Vega 56 into effectively a Vega 64 says the CU difference really would become less significant as frequency increases we quickly found that 56 is limited to 300 watts of power by bios that is a hard limit despite the fact that the vrm can easily handle more it is after all the very same vrm as on vfe and vega 64 the core and the vrm can handle the wattage but we're not given the allowance stock 300 amps really shouldn't be a problem 360 watts shouldn't be a problem but we're stuck at 300 so what's the solution well you flash bios that's not only the way it's done fortunately these cards the ante reference ones do have a dual BIOS switch so there's really no risk in doing it as long as you don't flash both in flashing bios using the newest version of ati flash which is Vega compatible we found that there is a security block on the bios that prevents any kind of modifications to bios and that includes changing things like the name of it so what we ran into was a successful flash it applied correctly rebooting you get a black screen at no display initialization and an error code on the motherboard that indicates as such so to fix this you boot back in you flash back and you're good to go unfortunately this is because of an these security features that they've added with Vega and AMD tells us that they are to comply with the secure boot protocol set forth by Microsoft and other vendors unfortunately that protocol does in friends on enthusiasts demand to be able to play with the video card but that's not seems to be taking the backseat here as has been the case lately so this was a big letdown to both build lloyd and myself we were really hoping for a lot more out of AMD a company which traditionally has advocated openness and permitted whether explicitly or not some level of modding that their competitors don't necessarily permit at least not at the same level after rolling back we next tried a registry hack this registry hack works with a Vega frontier edition to the point where you can push 400 watts through it if you want to so it's known to work it's just that it's a power table tack you install it you're good to go with Vega 56 the power tables mod available for efi doesn't work at all so either a new one needs to be written or it's just not going to work we think that it might be possible at some point but today is not that point so the registry mod that works for this does not work for 56 and the BIOS flashing works but you can't change it so there's no getting around to that 300 watt limit which means that as you'll see later in the performance section there's a whole lot of performance left on the table for Vegas 56 because there's just no overclocking Headroom which is a power choke moving on here we also proved the HBM voltage so HBM to voltage looking at it with a DMM we found it to be 1.3 volts it's a little bit lower than our VF e this partly contributes to the lower achievable HBM to clock on Vega 56 as it's not getting as much voltage B core now maxes out at 1.2 volts as well rather than 1.25 volts on VF furthering the overclocking limitations on Vega 56 anyway despite being disappointingly limited in overclock and we can still use Vega 56 to prototype a few veins under volting is one of them and then we can play around with the overclocking or HBM 2 vs cores login' is what we're about to do in this content a couple of notes here overclocking Vega just like with Fe has a lot of bugs along with it and the acknowledges these at least openly to the media and if you want to overclock as well you're going to have to keep an eye out for some of these although in two weeks once this thing comes out hopefully they're resolved but 64 buyers will need to keep an eye out for things like misreported frequency and misreported voltage where occasionally the voltage readout in watt man and I'm not sure about other tools but wot man especially will not match what you should see if you probe the back the card and also frequency will sometimes report things like 12 50 megahertz while still performing at 1600 megahertz levels of performance ie 1070 levels despite reading about 4 300 400 megahertz lower let's call it 300 to be more accurate so those are items of note for overclocking a lot of the software is wrong don't trust it just do the overclock run a quick fire strike test if the score improves great if it doesn't there's a problem and that's the best way to validate this right now not by looking at clocks as you OC let's start off with HB m2 versus core clock overclocking the behavioral outcome of HB m to burst core clock is going to change based on the application and the card so these results will not apply evenly across all gaming card some are more memory intensive than others but this is just kind of a quick look at things fire strike 1080p graphics scoring increments as we overclock the card stock we're at 18 8 1 6 points for an average FPS of 90 and gt1 increasing power targets by 50% boosts up to 21 188 no other changes so just power offset now our power consumption goes from about 196 watts to about 300 watts at the PCIe rail that's not the total system draw that's PCIe cable output talk more about power in a moment though anyway that's a gain of about twelve point six percent in this benchmark from the power target offset it's not linear to all games of course but it is significant here if we overclock HB m2 and offset the power our target we have about 3.6 percent boosted at nine fifty megahertz HP m2 over just the power darken offset it's not a bad game from HP m2 only overclocking to nine 80 megahertz HB m2 with a 10% offset on core because manual input didn't work with these percentages for the most part that boosts us to six point four percent over the power offset with no HBM to overclock or nearly nineteen point six percent over stock the memory overclock and power offset alone get us 19.2% overstock and the power offset gets us twelve point six percent over stock the takeaway here is that first of all power target offset does a whole lot for you and unfortunately it increases power but we have a solution to that it's under volt and we'll get there aside from this the next takeaway HBM to overclocking in our quick testing appears to be more relevant for performance than core overclocking if you had to pick one pick HP m to push it as high as you can and roll forth without noticing much of a loss we were at a couple percentage points max from overclocking the core alongside HBM 2 so that's the main takeaway and a lot of this limitations the reason we stopped where we do is because we're starved for power you can tell this by just looking at the current clamp while it's drawing never goes over 300 watts so we're clearly starved and this is a BIOS limit it's unfortunate that it's there but hopefully AV partners will have some flexibility to do custom BIOS or in the very least remove some of those security features although that seems unlikely so that you can push it further if you wanted to because the card is capable of more it's got a great vrm in it this is one of the best reference cards that we've ever seen and yet we can't really tap into the full potential of the vrm so that's unfortunate anyway the overclocking was disappointing let's try the opposite let's undervolt it here's a chart showing Vega 56 power as measured at the PCIe tables of the current clamp running twelve point three volts an absolute stock configuration and during a half-hour burnin with fire strike the V 56 consumes on average about 180 watts of the PCIe tables not accounting for the other 28 watts or saw the PCIe slot which is responsible for fan power at 2.4 amps and 12 volts here's what the power draw looks like when we increase the offset by 50% remember earlier we saw that increasing power target alone could stabilize clock and improve upwards of 12% and fire strike when comparing to the stock config of course doing this also increases power and lands us at 270 watts sustained power consumption at the cables rather than nearly 100 about a 100 watts lower here's where it gets interesting when under volting we managed to stabilize at 10 25 millivolts with DPM states 6 and 7 set to 1652 megahertz digging out here this is the clock output as we see it in software the real clock output is about 15 24 megahertz so you have to put in a higher number than you receive because AMD now has a dynamic clock anyway we've dropped from 1200 millivolts to 1025 that's huge the increase in power target of 50% along with this increases the speed overall in frequency and the result is clear the first 300 seconds or so that you see it going crazy is when we were trying to work with an these software issues at which point we gave up on watt man and resorted to watt tool which is a fantastic solution for getting this type of work done doing this made under volting to work and as we see when the line levels out around 210 watts we overall reduce power from the 50% offset at a slight increase over the baseline note also that the line is nearly perfectly flat now meaning that we've controlled from fluctuations in power delivery and clock frequency results has smoothed out performance on the whole while drawing 55 watts less power than the offset V 56 but 30 watts more than the stock card ultimately we get away with better performance let's look at a frequency chart next check this out we're at higher clocks than just the 50 percent offset 1524 mega study as a sniper versus 14 75 mega Hertz the stock card reports 1300 megahertz here resulting in a 224 mega Hertz boost for 30 watts more power not a bad trade at all there's plenty of more room to play to if you were to get serious about it we kind of cut it off here and called it a day for now big note there's a bit of reporting bug with Vega right now sometimes the frequencies look a bit funny it's always the same level of funny though seems to be an accurate miss reporting in software so when you're looking at frequencies it's kind of like if you were to take k-type thermocouple do you have a known range of variants maybe 2.2 Salty's or something like that but that thermocouple is always that same amount of measurements off it's always 1 Celsius off always in the same direction so it's kind of like that here where the frequency appears to be oddly read out in the same fashion across all tests so in the very least it's accurate in that regard although we don't know the true frequency because of AMD's new way they've set a Vega and the way they've set up the software it's hard to know but we did talk to them at length about this finally here's a look at the temperatures the fan was left alone for all these tests so there's obviously maneuverability for users willing to speed up or slow down on the fan with v56 that 50% off set we're at 84 Celsius by the end of the test quite warm the stock card and under-voltage card both fluctuated around the study for 275 see mark under bolting gave us more clock speed a middle-of-the-road power metric and didn't impact their most negatively that's a big improvement more or less across the board power is not so bad either just as a very quick aside here's a PWM to noise response chard we're at about the same noise level versus vfe within variance anyway as it's the same cooler nothing new to discuss here our auto speed on be 56 times V 44% rpm which outputs about 48.8 DBA and that's when you're in heavier operation the fan will sometimes sit closer to 40% technically 39 for lighter workloads which is about 45.8 DBA certainly neither quiet nor efficient as it is a blower fan the coolers just not very good but that's always been the case for these types of designs we really have to look out for AIB partner cards later to look for something with improved cooling that says the vrm is good so maybe a good h2o candidate if you want to dodge the bullet of the AIB partner cards where you pay for something they're not going to use anyway there was much talk of power consumption earlier here's a comparative chart showing idle power consumption at the wall again at the wall not at the rail so we switched how we're measuring from earlier wall draw the total system operates at 76 watts with Vega 56 which places it and here it's RX 580 predecessor the gtx 1070 fe operates at 67 watts idle or 12% lower Ghost Recon wildlands provides a gaming workload that places the Vega 56 at 332 watts power draw adjacent to our 10 ATT is e2 and just past the rx 580 org lock systems again systems here the gtx 1070 system has about 27 percent lower power consumption than the RX they got 56 system the overclocked 1070 consumes about 283 watts for the total system for honor shows power consumption at 313 watts for Vega 56 with the GTX 970 EFI stock at 232 or 26% lower power consumption finally 3dmark fire strike puts us at 303 watt system draw with the gtx 1070 efi at 212 watt system draw again that's the difference of about 30% the RX 580 operates about the same here once we've overclocked it and Vega efi system both 381 watts note that because of timing for filming this review and running tests there may be some additional Vega numbers or other numbers on these charts that aren't verbally mentioned they'll be there but we're kind of running a lot of tests alongside filming the video next some brief discussion on thermal starting off with a look back at Vega from to your Edition including our hybrid mods and 40 DB a noise normalize testing the Vega 56 card runs a GPU core temperature of about 74 to 75 see under stock conditions this requires a 38 to 44 percent fan speed depending on how the application enumerates the clock and impacts the via rams and things like that GPU temperature is therefore lower than VF e is stock but not all that much it's about what you'd expect given that the same cooler in a different GPU just slightly fan PWM to RPM response isn't too different here either the MOSFETs also measures similarly for the right side of hot spot where we observe a 63 C measurements on the Vegas 56 card this is completely within reason for a video card and for vrm and can even be run without a baseplate or VR on fan at all why not the best but we've tested it and you can actually do it at least on vfe so no trouble there measuring the backside hot spot on the PCV if hosing the top inductors the V 56 card reports a 65 cells is for the PCB rear temperature we remove the backplate later and saw that it improved thermals in this department by about 3 to 5 Celsius as the backplate is acting like a heat trap and preventing heat from escaping adequately Andy could probably ventilate this better in the future but it's a reference card so what are you going to do for a comparison to the 1070 F II here's a look at a 3 mark burnin for half an hour the Titans have any system draw about 250 watts at the wall during the burn with the RX Vega system drawing about 300 watts at the wall or about 20% more thermally the 1070 prefers to run a higher core temperature of 78 79 Celsius in favor of a lowered noise level with its fan operating at 53% and 42 DBA output the RX Vega 56 reference card prefers to stabilize at 74 to 75 C but it's been operating louder 44% and 48 DBA I noticed will stop a bit noise but slightly cooler there's room here to decrease noise and increased temperature if you wanted to but really make more sense to go with a partner cooler instead at that point let's get the gaming we're not going to have every single game test with under Bolton we're not gonna have overclock and results on here just yet before people start crying and saying Steve why don't you manipulate the space-time continuum and make it so that there's enough time to do everything in 27 hours that you've had the card we're working on it this this review has a lot of stuff in it already especially given the circumstances so work with me here this is the initial data we have 1070 and v56 performance stock some overclocked numbers and they're not too many and we'll look into it more but let's just kind of hold off on to firmly drawing conclusions just yet especially with the bike lock modifications and things like that that are forthcoming but this will give you a first look in the very least oh and as a note we did work on under volting pins and testing them in games but it's really buggy it sometimes will go down only three amps instead of six amps and sometimes it goes down the full 6 amps and sometimes it doesn't go down at all in power consumption so it's really hard to know when under volt things are actually working and we haven't figured out what causes it to break yet anyway started with Ghost Recon at 4k the RX Vega 56 performs at 32 FPS average lows at 29 and 28 the gtx 1070 reference card and FC cards both stood at around 35 FPS average setting Vega to a plus 50% power target gets it to 39 and overclocked in 1070 SC ties it with Vega at this point these are within variance and are effectively equal there's not a so-called Victor in a 0.3 fps difference they're the same at 1440p the RX Vega 56 card operate - the so not because average placing it behind the 1070 reference card by 7.6% or behind the SC by 8.8% overclocking the SC lands it at 68 FPS with Vega 56 adjacent when set to plus 50% power targets for about 65 FPS average overclocking HP m2 and the core will help here further in theory that will test that more soon in this configuration we're measuring 25 amps or 300 watts at the PCIe cables on the V 56 at 1080p the v56 operates at 74 FPS average or 84 FPS when set to the plus 50% power target the gtx 1070 runs in 83 FPS average stock 85 FPS for an AI BSC and 90 FPS when overclocking let's see we're ghost recon tends to show an Nvidia Advantage sniper attempts to show an AMD advantage so it goes the world of games or everyone optimizes differently at 4k v 56 operates an average FPS of 53 followed by the overclocked 1070 SC also have 53 FPS average though with marginally less consistent frame times the gtx 1070 SC at 49 fps is 8% behind here with the reverence 1070 at under 50 FPS at this point note also that vega frontier edition comes off working pretty poor in the scenario given its tied performance with v56 we're all we've been told by rob of tech gauge check them out friend of a site the vega 64 and 56 cards that still possess decent professional performance actually very good in the 64 instance in some cases and sort of makes BFE look even odder in our opinion that it already did but check the review out for more on workstation stuff ashes of the singularity and dx12 runs v56 at 67 FPS average with the GTX 970 SC trailing 5% behind at 63 FPS reverence card operates a 61 FPS average overclocking levels things out that we haven't fully overclocked b56 es so Andy still has some room there to booth beyond the 1070 SC check back hopefully this week for more of that with for honor at 4k the V 56 runs 38 FPS average against 210 Sony reference cards 41 FPS arrad Vega FPS it's at 40 FPS average in this title for reference at 1440p the v56 run 7% behind the reference 1070 or 9% behind the SC frame rates here are in the 70s for Vega 56 with ten 70s stretching to the low 80s 1080p illustrates the gap with most of the GTX 170 F EE at 123 FPS average versus one 11 of the b56 SC operates at 127 FPS average so these are 11 to 14 percent I had in this title but to any strong gain there will be countered by doom-doom at 4k post v 56 at 61 FPS average with the 1070 SC at 54 like sniper Andy tends to receive favor in doom from Vulcan which is what we're seeing here the lead over the 1070 SC is about 13% in this particular title countering the last one for NVIDIA for one of the more remarkable boosts over the tens only baseline finally we just added hell-blade to our bench so it presently only has a 1070 SC and vc6 on the charts 4k yields about 30 fps or both sets of devices so skip that at 1440p in very high hell-blade positions with 1070 SC at 57 FPS average and the Vega 56 is sort of nearby and before anyone cries that the reverence 1070 isn't down here yes yes we're aware no we can't manipulate time and space there's more work to be done we know but the limited time and be provided for this bench means that for now this is where it stands so that's it for now this was a lot of work already we had like I said something like 27 hours to do it all and there's plenty more to be done this is a fairly conclusive review in some aspects thermals and power but there's a lot more to learn in gaming check out other sites as always to pick up multiple sources of information because everyone's gonna be looking at different stuff right now because we all had no time most people had one two three days and this was sandwiched between thread Ripper components so it's an interesting choice by Andy to decide to launch Vega right after one of their biggest cpu launches of the year of the decade and give everyone two to four days to work on Vega not sure why they did that but that's why you're going to see a lot of reviews today that are probably going to be along the lines of here's the data I've collected so far we're going to have to look more head to it to really make conclusions which is basically what I'm saying here so we've got conclusions on something's power clearly v56 cause Moore's anywhere from twenty to thirty sometimes a bit more twenty to thirty percent more power in the same scenario as the GTX 970 in game that kind of plus or minus eight percent depending on what game you tested the Nvidia gave the games that tend to favor Nvidia and the AMD games at the games that tend to favor a and B we're showing basically opposite results so you get wan words plus 13% Nvidia and one words plus 13% AMD just depends on what you're playing you really need to look into the games that you play figure out which ones fit those metrics and that kind of dictates your card to some extent and then the biggest thing here is mining so this is the other reason why I hesitate to do any kind of really firm conclusion right now we have no idea what these cards will be priced at hours after release so we're told $400 is MSRP just like we're told from Nvidia that they're 10 70 MSRP is what but 400 or 380 or something like that so we're somewhere around there but it's clearly not available at that price most the other cars on the market aren't available at MSRP so we'll see how long it lasts if at all for V 56 and then at that point it's not a question of who's MSRP is lower it's a question of who's available price is lower so it's way too hard to tell definitively which card makes the most sense right now without seeing how the prices land but you've got the numbers so that gives you some preliminary information to think about and then from there hopefully stay on top of coverage and maybe watch the prices here where they fall so that's all for this time check back for more subscribe for additional coverage on this card and I think it's 3 a.m. so I've got about 4 hours of editing writing and uploading to go so hopefully this will make it online at embargo lifts subscribe for more patreon.com slash gamers x ourselves out directly there watching and i'll see you all next time I go to sleep now