Gadgetory

putting Andes at Vega frontier edition card underwater taught us a lot about the cards Headroom for overclocking its severe power limitations and how solving for power leakage can improve performance in non-trivial ways with this architecture in many ways it's a callback to the pure e^x especially when we look at the liquid cooling aspect of it we have better tools to test video cards now than we did back then today we're looking at the results for our Vega frontier Edition a hybrid mod which mounted a liquid cooler to the card and then used via RAM heat sinks separately for cooling the FETs before getting to that this is brought to you by our patreon backers you can go to patreon.com/scishow and join the patreon backers and if you join there you can join us on discord where we have a patreon back or chat with GN staff or if you prefer store gamers nexus net has our shirts available for purchase if you watched our Vega frontier edition hybrid model blog it was not the most straightforward hybrid mod that we've ever done we had to take an intel bracket CPU bracket and drill some extra holes in it so that allowed us to get the 64 by 64 spacing on Vega I believe the fury X also has that spacing that we ended up doing this instead because the fury X has other issues like choke clearance and then you have to start separating parts of the cooler that you really don't want to start pulling apart so instead we did this we took the liquid cooler we already had weren't using drilled some holes in the bracket and mounted it to the card and we've got b-roll of all that and it worked out pretty well for the vrm cooling we had to get rid of the plate so that did not keep the base plate on there instead we replaced it with just some thinned heat sinks and honestly these aren't even necessary the MOSFETs on this card it's a 6 by 2 so it's a 12 phase fix doubled basically and it's exposed copper anyway so if you just point a good fan of it you'll be fine and we sort of did that anyway but that's what the solution was and a few reasons for that one we didn't want to drill into the base plate and start making modifications in a way that we couldn't return this card to this condition because ultimately when it is tested in the future for regression testing we want it in its original state to represent the card as it's sold and also just it works fine this way so why not in fact it worked quite well so we're going to go over all the thermals and noise today one big thing here to keep in mind with these hybrid mods the goal isn't just to get more performance in terms of frame rate or frequency or points and synthetic benchmarks the point is to see one is there some kind of limitation and thermals and if there's not that's fine because the next thing you look at is what's the noise improvement what are the acoustic differences with liquid versus there when you can run this thing really if you wanted to at 70 decibels at 20 inches away at 100% fan speed or you run something like this at 43 and see that this will perform better so that's a big part of it and keep that in mind that it is about acoustics as well and then of course the bigger picture here is that we do this to learn about the card and one of the things we learned about Vega Frontier Edition is that power leakage actually impacts your overclocking the Headroom quite significantly this was the same with the fury X so the fury X when it launched there was a lot of discussion by review sites ours included about why do you do a closed loop on your video card part of that answer was you at lower power leakage on HBM and on the GPU itself and that gives you either just lower power consumption overall at the same clock speeds and voltages or more Headroom to boost the clock and voltages and so that's what we see again here it's at this time this particular model is air-cooled although they do have a liquid-cooled variants as well it will be sold in smaller quantities so we're focusing on the air-cooled variants with our own mod instead let's get the overclocking numbers on the screen first since these will be shown in the rest of the test the Vega Frontier Edition hybrid model actually posted reasonable gains an overclocking over-the-air cold variant this is due to more aggressive clock scaling the versus the power consumption similar to what's seen on some other architectures we were able to max out our stock card air-cooled at around 16 60 megahertz overclock with an 1,100 megahertz HBM to overclock this we ultimately found governs performance more heavily than the core by the way a hybrid card with the help of some fans point of um components that they were unnecessary I was able to overclock to 1705 megahertz completely stable no issues at all with an 11:05 megahertz 8 p.m. to OC this sustained through all the fire strike tests all the gaming tests and spec view per we end up running our published numbers that you'll see in this video at 1700 megahertz core and eleven hundred megahertz memory and we'll be revisiting with higher clocks later because we think there's some room to go a bit beyond 1705 and 11:05 anyway with some better tuning as for power consumption our overclocked Vega Fe hybrid mod had us around thirty three point three amps max during our tests at the rails not counting the PCIe interface power and that's for GPU power consumption not total system power so GPU power consumption at the PCIe rails from the power supply reads out at four hundred Watts thirty three point three amps is about four hundred Watts twelve volt power that's GPU power again not a total system power this is the power limitation for Vega and is the reason we can't OC higher even though we have plenty of thermal Headroom to do so and is something we'll look at more in the future hopefully to unlock a higher TDP either way the most notable element here is what's not shown in temperature plots noise output the hybrid mod outputs a constant noise level of 43 DBA because we're maxing the radiator fan and that's what the radiator fan only the Maglev fans boost that noise level a little bit but we're still below what the stock card does when you're in the overclock settings in order to sustain its lower overclock the 16 60 megahertz one the air-cooled stock card was whining at about 67 DBA output the perceived at noise level difference here as significant we're at more than a two times perceived reduction with the hybrid mod given that this is a logarithmic scale that is it's more than 2x it's not a linear scale and this next test is new it's also pretty cool we're looking at power leakage here using a current clamp to measure power consumption at the 12 volt rails from the power supply we're not measuring PCIe slot power yet which can push up to 75 Watts under the spec but maybe in the future this touch requires the card to operate at a fixed reliable speed without any fluctuations as observed in the stock air Gold Card which never really attained or held at 1600 megahertz anyway as such we're running the core clock at 1,300 megahertz for both the liquid cooled and air cooled variants since that can be sustained without change and a memory clock of a thousand megahertz which seems to help stability as well the power offset is at plus 50% just to further ensure there are no fluctuations the result is on the screen now with the air cooled card we start encountering power throttling that had us bouncing between 18 and 25 amps once the GPU hit 85 C and so we are forced to increase the fan rpm manually from its Auto configuration this fluctuation by the way is indicative of small throttling occurring at a level that doesn't always show up in clock measuring software there is performance loss from the high thermals but it doesn't necessarily show up increasing the fan rpm to a 51 DBA and sustaining about 80 to 83 Salty's on the core we see that the power consumption is now around 26 amps sustained from the power supply after some burnin time whereas the hybrid card's power consumption even after burnin time and with temperatures of around 42 BC sits closer to 23.4 amps for the power mark this is also largely unchanging from start to test and while the air-cooled card scales from about 22 to 26 ABS over the period of the test this is a 2.6 amp overhead from the hybrid mod resulting in a power leakage of about 31 watts due to heat so measuring with this we're drawing 31 watts more with the air-cooled card versus the liquid cooled card with the fan keeping vanes at around 81 to 82 Celsius when we were hitting 85 C it was starting to throttle enough in the power consumption metric that adjust it was a much harder task to actually be comparable but if you manually increase the fan outside of its auto settings to push a bit harder we're seeing a power loss basically through leakage of 30 watts and that's where the additional Headroom comes from when we got the liquid cooled variants of this card about 40 mega Hertz higher on the core so that's the answer of where your Headroom emerges and it's in the extra power now the extra thermal is necessarily the certainly helped by way of reducing power leakage one more note here's a look at the temperatures under this particular workload that we use for power leakage measurement our hybrid mod is able to keep the MOSFETs around or below 50 C for this entire test the stock card is hitting the upper 80s and 90s in this load especially again with the plus 50% power offset with a fan configuration we use now these numbers are not comparable to the initial review tests because they use a completely different method and a different software benchmark and even some different positioning for the measuring tools but overall we're within spec it's certainly just contributing to power leakage and boosted temperature of the GPU core given that they ultimately share the same cooling solution and are all neighbors this verifies some of what we already knew or at least what we thought we knew and that is that Vega frontier Edition is bumping against power more than anything else for limitations so the power limit appears to be about 400 watts to the PCIe cables connected to the card you've got some additional power that comes in through the PCIe slot on the motherboard which we can't currently measure but we are looking into it and that's the limiter now as we get around that you would need a BIOS update or BIOS flashing tool and a hex editor which is something we're looking into and trying to figure out we may end up using something like a Raspberry Pi to flash the BIOS because no flashing tools currently exist ATI flash doesn't work on Vega right now it's something we're looking into no promises on if it'll work which means we wouldn't deliver content on it but if it does we'll certainly be back and try and boost TDP to something like 500 watts and see what does an extra hundred watts get you out of the clock assuming that the card can survive at that point which should be feasible if you look at builds BRM analysis on our Channel the vrm can take it it's just a matter of can we get it to happen so that's the next step we are power-limited more than anything else but the thermal limitations sort of existed in that you had a an indirect conflict for boosting clocks and that was because of the power leakage because of thermals just to illustrate the clock rate stability with these our clocks quickly here's a look at frequency versus time and our testbench as at red just through an automated series of 3dmark benchmarks getting from ultra to extreme than normal and followed up with times by both clocks are fairly stable in this particular run but that's because we've boosted the speed on the air-cooled card - Brent throttling we can also plot the stock air-cooled card with no overclock and auto fan speeds and that's this next line that's being drawn on the screen now moving on to power versus thermals in a stack view prayer workload which cycles load as the test executes and simulates production and tests we see the overclocked hybrid configuration drawing about 600 watts from the wall again from the wall we're back to wall draw test and not the current clamps testing so 600 watts during the heaviest load of this cycle to compare versus the stock cars this is an increase from around 440 watts loaded from the wall originally and that's just measured that way because we started running those tests before we got the current clamp despite the increased power draw of the hybrid overclock we're actually way down in thermals from the air-cooled card the liquid cooler also soaked thermal changes and high thermal load spikes with greater finesse ultimately topping out around 47 Celsius or the stock card was hitting 81 C pretty close to its limiting 84 to 85 C throttle range finally here's a quick study state thermal chart to show the temperature performance of various board components when the card is put through a power virus scenario we see significant improvement not only on the GPU temperature but the two hot spots on the vrm even without additional fans providing direct airflow the vrm components are hybrid mod is well within any component heat specification most of which are well above 100 C above 125 C even and we're only showing Vega on this chart right now for comparative thermals and noise normalized thermals check our initial Vega frontier Edition review which also contains a Titan XP and other cards getting into benchmark numbers now this chart shows the specular performance improvements in various workloads with our stock hybrid and overclocked hybrid mods 3ds max posts the performance improvement of about 6 percent when overclocked of the hybrid mod with equal performance for the stock air and hybrid Vega F II cards kadia posts an 11 percent performance gain from the overclocked and liquid with creo at a 5 percent gain followed by the energy test at a 15 percent gain and then the Maya test with a 6.8 percent gain going through these pretty quickly the medical benchmark is responsive to the overclocked actually decently responsive posting nearly a 19% improved with the ocm liquid with the next past showcase at a 13% improvement SN x o2 at 9.5 percent improvement and s w o3 at an 8 percent gain from the overclock only a few of those were particularly exciting the rest don't really feel like they were worth the effort kind of a lot of effort for no big gains and some of those tests but the medical one as a standout example does look better for the 100 megahertz boost on the core and sets the stage for a next round of tests which are something gaming workloads and a few things to note here this card is clearly not really gaming ready at launch it does have a gaming mode driver it is supposed to be able by am these own definition to sit between the gaming and the professional audiences something like we said in our initial review might be occupied by a game developer despite this not being fully ready through drivers or enabled or disabled features we're still testing it for gaming and this still provides a good look at clock scaling even with the driver limitations or shortcomings and even with some of the features potentially disabled so we still get a look at the clock performance uplift which is an important metric nonetheless and just another note like we've been making in all these videos we're not talking RX Vega performance yet and suggest that you refrain from attempting to speculate or extract numbers from these on how Rx Vega will perform just wait for our review to see how that does starting with fire strike with a pen ATP test our graphics core on the hybrid modded an overclocked score of twenty four thousand nine hundred ninety eight points versus the twenty one three five five score of the stock air-cooled card that's a gain of 17 percent with the overclocked hybrid with the overclocked air card we were at twenty four five five four or a gain of 15 percent over the stock card the boost strictly from our extra 40 megahertz on liquid isn't tremendous over-the-air boost only a couple of percent but again this is about thermal and noise performance more than anything and the noise is way down with the hybrid mod firestrike ultra has our overclocked versus stock air gain at around seventeen point seven percent so that's the 4k benchmark once again providing some consistency between 4k and 10 dp4 the other fire strike in times five test check the article linked below we're going to only show a few games next not too many and all the rest of the benchmarks can be found in that article linked below started with doom at 4k with Vulcan and asynchronous compute and yes disabling anti-aliasing does still enable asynchronous compute as this was changed with more recent patches to the game we see the overclocked vega card operating an average FPS of 80 with lows at 66 and 60 is a staggering performance gain of 25% in doom specifically with the overclock and the liquid cooling and as the title in which we saw the biggest Swain note that doom and Sniper Elite both tend to favor Andy so this is something of a best-case scenario for the overclocking result you cannot and should not extrapolate this 25% number to all the games as well show more momentarily regardless that's an impressive boost and not to take away from it the Vega EFI card on liquid and with our significant overclock is now performing in between a gtx 980 reference card and a gtx 980ti reference card sitting about 11% behind the 1080 TI f e or if you want to look at an overclocked model from an AIB partner 19% behind the 1080 TI gaming X model now big note here this chart doesn't contain overclock numbers for the 1080 simply because it's not really the focus right now and where you haven't rerun them on this suite yet which is a brand new test suite so you'd see some performance up with there regardless Vega FPS gains are noteworthy in Doom we're next moving on to Ghost Recon wildlands with DirectX 11 very high settings and 4k resolution in this title the hybrid overclocked configuration outperforms the stock air configuration by 17.5% certainly not as impressive as in doom but still a good boost that boost is that the cost of course of 400 watts of power consumption through the 12-volt rails coming out of the power supply a hybrid OC card is now effectively tied with the 1080 F II card under stock clocks but the difference between them is within our test and test variant so they are effectively equal aside from the power metric which is quite different the hybrid OC is behind the 1080i F II card by about 20% or behind the 1080 TI gaming X card by 24% if you want the rest of the gaming results including Sniper Elite for Ashes singularity and for honor you can check the article linked in the description below that'll have those and will prevent some of the copying and pasting of the words it's not a gaming card in the comments below so you can check that out there if you want to see it a few things here this card still shouldn't be bought for gaming it never should have been we've said that since the initial review and the same is true kind of for its direct competition the Titan XP but today we're focusing on Vega versus Vega liquid as we've created it anyway so that's the first thing to note the drivers still need polish not just for gaming but for everything some of the programs need polish there are still issues with blender and some of them appear to be driver side potentially in a more kernel failure area of the drivers so there's a lot of things that Andy needs to improve before the card is really bought in general by the mass audience this was a rush launch despite being the way we've already talked about that the Russian was largely in the form of drivers but rx Vegas coming up and on the note of rx Vega the same disclaimers here try not to extrapolate too much on performance instead of speculating wait for our review we'll have one and doom I'm just mentioning this because I know it will be posted as a seminal example and stripped out of all context of the rest of the review and people just share the dune hoppers doom looks like a really good uplift from the overclock on this card because Vega as an architecture will still be used in our X Vega we could expect that overclock in our X Vega provided there is reasonable Headroom with our X Vega would also see pretty good results from Dumas specifically that said again I would caution you to just not speculate what the performance of our X beta will be based on rumored clock rates or card specifications are for our X Vegas just just wait for that so back to what we're supposed to be looking at Vega versus the hybrid mod it's an interesting question of was it worth it because it was a hell of a lot more work than normally thrilling the holes in the bracket certainly added a bit of time but then you've also got to do things like deal with the base plate and if you don't want to modify the base plate or you don't want to start cutting them to your liquid coolers then you have to find another solution like this or just like running more fans over the prm's so we got pretty decent gains in thermal performance we got pretty decent we got excellent gains in their own performance excellent gains in acoustics and then in gaming some of the performance uplift from overclocking was to the range of 17% which if you look at it versus a narrow overclock only a couple percent more but way quieter and not screaming from a blower fan so those are all decent and worth note I am not sure yet what the difference is of vega's liquid-cooled officially liquid cooled card versus the air-cooled card theoretically it has a bit higher power target and that's probably because of the reduced power leakage it's not from something else like this modified bios but from what we can see here there are decent gains to be had with a good hybrid mod whether or not you should buy the card is a different question you can view our review for that if you're curious about the cards positioning overall but that's what the hybrid model ike uplift is good in some titles is boring in a few of the spec you perf tests but overall a very fun test to run and we learned a lot about Vega about power leakage and about how to use new tools for our benchmarks so thank you for watching as always you can go to patreon.com/scishow and access if you like these types of tests I would like to help us do more of them buying tools like this not necessarily cheap but definitely worth it because we get a lot of cool new data out of it so as always subscribe for more we'll see you all next time you