Xtreme Addict Posted May 4, 2017 Posted May 4, 2017 (edited) Introduction Finally 1080 TI custom PCB cards are out! This time I would like to focus on Asus Strix Gaming 1080 TI card. This guide applies also for POSEIDON card cause they use same PCB. Still 1080 TI chip is pretty new for vendors, and there is no super top highend LN2 card as Matrix yet, but looks like Strix is doing really well. Though we have to remember that in product ranges it's just "low rank" custom design which is supposed to be just better than reference card. It's not designed for XOC in first place, though it doesn't mean it can't be overclocked Testing platforms: i7 6950X @ 4.5 GHz Rampage V Edition 10 4x8 GB Galaxy GOC Edition B DIE HyperX 240 SSD Kingpin Fat TEK 6.66 for GPU & Corsair H100 for CPU Seasonic 1200W Platinum Thermal Grizzly & Kingpincooling "BLUE PASTE" Gallery: Stock accessories. Nothing extraordinary, just useful things - drivers, adapters, power connectors. Stock card with mounted cooler - looks really big and powerful. Back of the card. Unmounted stock cooler. We can see a lot of material used for stock cooler structure, many heatpipes which transport heat from the core to the fins of radiator. It's definately not a "cheap" one - better than stock reference cooler. Stock cooler Fan/RGB AURA connectors, extra RGB connectors and additional FANs connectors. All can be controlled and adjusted by AURA software for RGB enthusiasts. Naked card. Micron "D9VRL" GDDR5X memories. "Missing" GDDR5X memory ICS which distinguish the card from Titan X Pascal. PCB was designed to deliver more features. We can notice soldering pads for BIOS switch in other section of PCB we can notice room for 2x BIOS SMDs, but only one is soldered on the card. ASUS decided to use "IOR" mosfets. I would say it was a very good call - from XOC experience those are most likely the best on the market providing very good parameters of work, low RMA rate, and can handle high load with high voltage. Just as reminder - top models of XOC motherboards or TOP XOC GPUs like Matrix also commonly use IOR mosfets. "uP9511P" - reference VGPU controller, used on reference Titan X Pascal, Titan Xp, GTX 1080 TI FE. Sadly it's a bad news - I am not the enthusiast of not fully digital solutions, cause from experience proper voltage controller design like iChill or IOR is a key to max out CPU, GPU clocks. In my honest opinion best solution is to pair IOR mosfets with IOR voltage controller. "ITE" controller is responsible for AURA RGB features and extra FAN connectors. Not important for us - overclockers. Well known "HOTWIRE" feature. "Missing" 2nd BIOS. My test platform. To sum up design - it's upgraded reference design with extra gaming features, using more phases and better quality, though controllers are still reference. There are a lot of similarities to reference design - for instance card uses "old school" Texas Instrument INA3221 power sensor IC, which is the "heart" of Boost 3 technology, controlling power limit. More advanced/experienced users, especially those who are close to the industry, know that NVIDIA lately is very tough partner in business and tries to control everything even in custom design cards. Currently NVIDIA has to approve "PCB design" and also "cooler design" in order to produce a card. We might have to be sometimes forgiving for vendors cause not everything is in their power to change. AC TIPS Pascals GPUs are mostly limited only by temperature and powerlimit. On stock cooling there is not much we can do. Due to Boost 3 technology Pascals GPUs, depending on load, power target/limit and temperature can reach 2000 MHz in peak moments. Frequency is changing real time all the time. In most of heavy load games we will sadly notice throttling in many moments - our clocks will go down and up all the time influencing negatively efficiency. If we are lucky and have card with high leakage/ASIC our stock card will be working in peak moments with more than 2 GHz GPU. There is not much we can do regarding throttling - we can extend powerlimit/powertarget and improve cooling or change it. Basically I would recommend to follow Roman's YouTube AIR stock guide for GTX 1080 TI. There is everything you need to know about optimizing card for AIR. OC potential on stock cooling/stock BIOS Sadly Pascals GPUs (especially GP102) are pretty maxed out on stock when we are talking about stock AIR cooling. Especially considering custom PCB designs which are already overclocked by vendors comparing to Founder's Edition. We shouldn't expect too much. Thx to ASUS I was able to play with 2x STRIX 1080 TI cards. Both cards on stock cooler and stock BIOS boost to almost 2 GHz, first card was doing 1990 MHz GPU, second 1999 MHz GPU. To determine max stable clocks I used Firestrike Extreme test. Those were max clocks which cards could pass test without freezing or artefacts. 1st card) 2025 MHz GPU/1550 MHz MEM 2nd card) 2020 MHz GPU/1550 MHz MEM Actually overclocking core gives almost no boost, cause it's just 20 MHz. Memory OC headroom is much better and provides significant efficiency gain. Those are pretty typical OC results for BIG Pascal (GP102) considering my experience with Titan X Pascal cards and GTX 1080 TI FE. Some cards of course can reach higher frequency - rarely we can see around 2050 MHz GPU. Best so far I have seen is 2100 MHz GPU on stock cooler on STRIX GTX 1080 TI by smoke and slamms, but card was binned from 10 pieces I am pretty sure for their golden card sky is the limit and we will see some nice WRs coming soon! XOC Firstly - a very good news - we have fully working real XOC BIOS which removes all powerlimit/powertarget protections in NVIDIA drivers. Second great information for many XOC guys who don't like soldering and so on - we can push the card without a single mod all thanks to special tools. A) XOC tools: STOCK BIOS XOC BIOS + NVFLASH FOR WINDOWS B) ISOLATION Everyone has their own way of isolation. Some models of hardware don't like some ways because of freezing or small shorts from vaseline and so on. For ASUS STRIX GTX 1080 TI both cards I used vaseline and paper towels around the pot and on the back - and it was working great for long sessions. I would recommend this method. C) MOUNTING & POT Kingpincooling TEK9 6.66 FAT - FITS! Sadly for some other pots there will be small issue in design of the card because of 2x SMD capacitors so we can't resolder them to the back of the card ? The easiest solution is to find a fitting POT or remove factory capacitors and solder some with long legs, don't forget about protecting the area to remove the risk of potential shorts. EK-SF3D Critical Point GPU POT and der8auer's Raptor 3/4 won't fit the card without a mod. THERMAL PASTE MOUNTING I would recommend "Skylake/Kabylake" way of applying thermal grease and not to move the pot during mounting. Moving it on sides causes thinner layer of paste on parts of GPU which can lead to cracks. I recommend to keep card cool so paste won't be too warm - too liquid. I used Thermal Grizzly and Kingpincooling's "BLUE PASTE" and both worked really well with full pot. D) LN2 TIPS 1) Cards don't have CBB/CB! 2) Don't use NVPMManager software to force 3D mode - with high clocks it will reset drivers before launching the benchmark! 3) I recommend using the newest version of ASUS GPUTWEAKIT for changing clocks - works very smoothly. GPUTweakII Version 1.4.5.2 Don't forget to extend "OC range": 4) GP102 DIE can handle higher voltages than smaller DIE from GTX 1080. Most of the cards can work with 1.5+ VGPU. Sadly clocks are not scaling lineal like in old great GTX 980 TI cards. Though proper voltage/temperature relation is the key to reach the maximum of the card. Sadly uP9511P controler is based on "VIDs" and voltage jumps depends in which Pstate card currently is. On some cards it might cause stability issues, especially during loading the benchmarks - we can fix it by hardmods which are described later in section of MODs. Second important thing - because of Boost 3 technology, when we cool down card below 0*C we are losing the highest BOOST table, so for instance my 2nd card on stock boosts to 1990 MHz on positive temperature, after cooling down maximum BOOST is 1950 MHz. When we lose highest BOOST are stock voltage (which on AIR is around 1.06v) can go down to around 0.9v. So automatically we have to raise voltage or by trimpots or by software. Optimal LN2 voltages for my cards in load which allowed me to get highest clocks: VGPU: 1.52V (stock is 1.06V) VMEM: 1.4-1.5V (stock is 1.35V) VPLL: 1.1V (stock is 1.0V) Though PLL voltage and memory voltage have very little influence. Especially considering memory frequency - adding extra voltage mostly is in order to prevent memories from being too cold, especially when we are playing nearly fullpot. 5) Card stability/behaviour. The hardest part is to start the benchmark or start benchmark for more than 3s. We have to find perfect temperature/voltage. It might be also due to NVIDIA drivers and changing VGPU tables (please check out later - MODs). Sometimes I had to try 3-4 times to start benchmark and pass desired frequency. The good thing is that even after crashing benchmark - drivers come back and doesn't restart, clocks are still fine with good efficiency. 6) Temperature/voltage relation. Each GPU DIE has it's optimal inner temperature of work - which allows for highest clocks. With lower voltage - around 1.1-1.3 VGPU best temperature for most cards will be around -130*C/-140*C. When we go colder - our stable clocks will be lower or card will be unstable. But when we keep adding voltage we can move this barier and go colder which will allow us to clock higher. Temperature also depends from benchmark load. For instance with Time Spy I can go full pot - but FSE or 3D11 need around -160*C to work. Mostly for benchmark start and first 3 seconds we need slightly warmer temperature - for instance for my highest clocks I was starting Time Spy at around -180*C (max idle was -189*C on my Kingpin probe/Fluke) and than going full pot in load which was around -179*C. On my 2nd card going from 1.3v to 1.52v allowed me to move benching temperature from -130*C to -180*C for Time Spy and allowed me to reach 90 MHz GPU more. 7) Cool down PWM. I recommend STRONG delta over the mosfets and on the back of the card. EXTRA MODS - NOT NECESSARY BUT MIGHT HELP IN STABILITY AND CLOCKS POWER LIMIT Because we have fully working XOC BIOS which disables protections in NVIDIA drivers and also disables POWER LIMIT but we still can do oldschool "SHORT" mod to remove POWER LIMIT by hardware mod. FORCE 3D VOLTAGE As I mentioned earlier voltage is regulated by drivers/Pstates and it's fluctuating all the time during start of benchmark which can lead to extra instability. We can Force 3D voltage. After this mod, our stock voltage will be 0.87V in IDLE and will drop in STRESS around 0.1V. After this MOD we have to add more voltages by offsets/hotwire. In order to remove VDROOP in stress we can do additional mod for decreasing load line calibration. LOADLINE CALIBRATION Decreasing LLC will remove vdroop in stress. HOTWIRE HOTWIRE POINT TO GND (GROUND) GPU 20k VR TRIMPOT MEM 50k VR TRIMPOT PLL 100k VR TRIMPOT RESULTS My first card (slightly better on air) was able to hit 2340 MHz GPU /1560 MHz MEM for hard benchmarks and for some lighter benchmarks 2380 MHz GPU, though was really hard to manage and not stable - extra stability mods helped a bit. This one was scaling much worse on voltage - 1.1v to 1.5v gave me around 100 MHz GPU. Second card was clocking much worse on low voltages. At 1.2V I couldn't even pass 2200 MHZ GPU but with 1.52v I was able to bench at 2440 MHz GPU/ 1566 MHz MEM. This one was greatly scaling with Voltage/Temperature "ratio". Additionally second card had no mods at all, not even POWER LIMIT mods. Only flashed XOC BIOS and used XOC TOOL for voltages adjustment. When I refill LN2 I will try card with full mods and full out CPU Summary Card is very nice to play with, easy, no mods required to obtain high frequency. All software/BIOS work great. No CBB/CB also makes fun better As GPU is the part of Pascal family - temperature is the key and voltages give rather slight boost. The key is to find optimal voltage/temperature relation. DISCLAIMER: Do all at your own risk! I am not responsible for any damage, nor ASUS or anyone else. Please note you might void your warranty by overclocking/modding etc.. Edited October 18, 2018 by Xtreme Addict Quote
Alex@ro Posted May 4, 2017 Posted May 4, 2017 Good to see that we can finally bench GP102 without the reference limitations,nice work! Quote
phobosq Posted May 4, 2017 Posted May 4, 2017 Lots of useful info, too bad I will never have a chance to try it thanks Michał for keeping OC spirit of knowledge sharing alive! Quote
Nick Peyton Posted May 6, 2017 Posted May 6, 2017 It's been farrr too quiet on the Nvidia O/C'ing scene over here. For someone as passionate about this as me, this is sad. So I am absolutely delighted to see something fresh and exciting over here. Excellent work! On behalf of my friends and I; we all REALLY enjoyed this article. Hope to be able to join you soon. As I just received (not long ago) my Kingpin pot & Hailea Water Chiller ;-) Thanks again. P.S. @Xtreme Addict I've sent you a PM on here and also on Skype; would really appreciate it if you could drop me a quick reply (just so I know you got it). Nick Peyton aka nrpeyton Quote
Hulio Posted May 7, 2017 Posted May 7, 2017 (edited) Sadly this Bios has worse performance. Edit: Testing done and done some calculations. Kingpin Time Spy GPU Score 13230 @ 2531/1583 LN2 Xtreme Addict Time Spy GPU Score 12316 @ 2440/1566 LN2 Hulio Time Spy GPU Score 11376 @ 2113/1589 H2O 2531/2113=1,198 -> 19,8% higher clock speed 13230/11376=1,163 -> 16,3% higher GPU Score clock to performance scaling isn't linear, no surprise here, but its in line with what we see 2440/2113=1,155 -> 15,5% higher clock speed 12316/11376=1,083 -> 8,3% higher GPU Score it should be more like 12,9 % more GPU Score So with the XOC Bios you are missing ~5% performance in comparison This is exactly the same factor, i'm getting less GPU score in my testing while using that XOC bios with my clocks compared to nvidia fe bios on water cooling. Something is wrong with that BIOS, 5% Performance loss is huge Edited May 8, 2017 by Hulio Quote
Benny89 Posted May 7, 2017 Posted May 7, 2017 XOC Bios seems to now doing what it suppose to do on my 1080 Ti ROG STRIX. I have flashed XOC Bios and I still get Perf Limits: Power and Voltage in GPU-Z and OSD when I run Heaven. Card does not downclock anymore on 40 and 50 temps, it also draw more voltage on stock. However I still get Power Limit perfs and it downclocks my card after I get past 60C. Max my card did draw according HWInfo was 320W so far. I tried flashing stock again, reflashed XOC, and then tried 3 more times but nothing change, still getting Power Limit perfs and downclock because of it after certain temp treshhold. Quote
KraxKill Posted May 8, 2017 Posted May 8, 2017 I'm getting 5% lower clock for clock volt for volt performance on the XOCbios with my Shunt modded FE. My scores at 2164mghz would require 2300mhz+ on the XOC just to break even. Quote
elmor Posted May 9, 2017 Posted May 9, 2017 Are you guys checking what the actual frequency is during the benchmark for the comparisons (reported, not set frequency)? Compared to default Strix 1080Ti BIOS performance is identical. Quote
peterftl Posted May 9, 2017 Posted May 9, 2017 Are you guys checking what the actual frequency is during the benchmark for the comparisons (reported, not set frequency)? Compared to default Strix 1080Ti BIOS performance is identical. Do you mean the reported frequency in MSI AB or GPU-Z? Yes it is always sticked at the max frequency which you set because of unlimited power limit, but the issue is that the performance is not scaling properly with this XOC BIOS, guys who are shunt modded are getting higher performance on the same clocks. Doesn't anybody know if there is going to be any improved version of a power unlimited BIOS with fixed performance scaling? This XOC BIOS looks like an early alpha version to me. Quote
peterftl Posted May 9, 2017 Posted May 9, 2017 Awaiting for the T4 version:D What is the difference between XOC and T4? Quote
nacho_arroyo Posted May 9, 2017 Posted May 9, 2017 Are you guys checking what the actual frequency is during the benchmark for the comparisons (reported, not set frequency)? Compared to default Strix 1080Ti BIOS performance is identical. The bios works great, even better than reference that comes with EVGA and also ASUS on FE editions. Great works XA, very bad about that CAP that kill the GPU POT mounting.... Quote
KraxKill Posted May 10, 2017 Posted May 10, 2017 Are you guys checking what the actual frequency is during the benchmark for the comparisons (reported, not set frequency)? Compared to default Strix 1080Ti BIOS performance is identical. Yes..I log extensively. It's verifiable and reproducible for me without fail. I'm testing by isolating the GPU from the CPU scores using 3D Mark TimeSpy Test 2. I have yet to find a bios that works better than the Vanilla Nvidia FE Bios. I've flashed nearly all of them. It's not just me, there are a bunch of guys on OCN seeing the same delta. My card is shunt modded, so it could be that it's behaving differently than an out of the box FE which would otherwise be crippled by pwr events at lower clocks, but I tested it all the way up to 2200mhz with -10C idles and could not beat my FPS scored on the Vanilla FE Bios at just 2100mhz. The Nvidia FE BIOS is bench stable for me up to 2164mhz at -10C idle temps and about 2100 or so at 20C ambient. I can go all the way down to -20C at 32C ambient with the flip of a switch, but I gave up on the XOC after seeing it underperform my FE Bios at 2100mhz while clocked at 2200mhz. The clocks are higher and are pretty to look at but it's doing little for me when put to a test. After testing numerous others, I thought surely this BIOS will do the trick, but sadly not so. I'm seeing 63fps on the XOC vs 65fps on the Nvidia FE Bios at 2100mhz @ 1.063v on my shunted FE at -10C idle, 4C load temps in Time Spy Test 2 in back to back, clock for clock, volt for volt testing. I have not tested it at clocks below 2100mhz but those that have are reporting it performs about equal to the stock FE BIOS there. Glad it helps some, but not I. I would urge you save time testing the BIOS by running just test one or two exclusively within your favorite 3DMark bench. I was excited for the voltage control and no pwr but something is odd with it. My thinks that the card is simply running out of juice up there. What I really would like to see is a stock Nvidia FE Bios with voltage control up to 1.2v. Doubt will ever see that. Quote
marco.is.not.80 Posted May 11, 2017 Posted May 11, 2017 Thanks for this XA and thanks for the XOC Bios/Tools. I just picked up two of the exact same cards in your photos and just waiting for my waterblocks from EKweb but lots to look forward to now :-) Quote
elmor Posted May 11, 2017 Posted May 11, 2017 I get identical performance here with XOC bios and Nvidia FE bios. Can someone send me the exact bioses and conditions where there's a performance difference? Quote
jaffers Posted May 12, 2017 Posted May 12, 2017 (edited) Fixed, realized that with this bios, adjusting voltage using voltage tool does not report correct voltage, although there is a change Edited May 17, 2017 by jaffers Quote
Gorod Posted May 13, 2017 Posted May 13, 2017 Xtreme Addict , Elmor , Asus : Great BIOS , thank you guys ! :celebration: Working well on Strix Quote
CornerJack Posted May 19, 2017 Posted May 19, 2017 Tests with the XOC bios First comparative performance between stock bios Asus GTX 1080 Ti STRIX OC (86.02.39.00.26) and XOC (86.02.39.00.58) 3DMark 11 stock : 27940 XOC : 28223 3DMark Fire Strike stock : 22625 XOC : 22941 3DMark Fire Strike Ultra stock : 7127 XOC : 7317 3DMark Time Spy stock : 9388 XOC : 9480 The XOC bios is better but max Boost with stock bios is 1987MHz and with XOC, 1999MHz. In aircooling, my 1080 STRIX OC can do 2100MHz. With the XOC bios and voltage, it's not possible to make better in air. I test today in water and after DOD/LN2. Addict : do you have make test in air/water with the XOC bios ? Thanks Asus France, Addict and Elmor. Quote
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.