Important Starting from the second half of 2019, most cards ship with a newer XUSB FW Version ID, preventing almost every single BIOS in the TechPowerUp BIOS Collection from being flashed to the card through NVFlash, so be aware when purchasing a brand new card, you might not be able to flash a higher power limit BIOS on it. Example: MSI Gaming X Trio purchased in Q3 2018 shipped with the 0x70060003 firmware (Build Date: 2018-08-29), the same MSI card purchased a year later in Q3 2019 shipped with the firmware 0x70100003 (Build Date: 2018-12-28), this newer card also had a different EEPROM (WBond instead of ISSI). Note: The BIOS and firmware can be overridden forcefully using external programming/test clip, bypassing NVFlash altogether.
Question: What does Non-A mean? Answer: There are two GPU variants: TU102-300 (1E04) and TU102-300A (1E07). Factory overclocking is prohibited on the former, it has a boost of 1545 MHz and a maximum power limit of 280-310W. The latter chip has varying factory overclocks and power limits up to 520 W, flashing a 300A based BIOS onto a 300 GPU and vice versa is not possible. Manual overclocking is possible on both.
Question: How do I know if I have a 300 or 300A card? Answer: Check the list above, but it may be incorrect as the early batches of the ASUS Turbo card had the A chip, recent reports indicate later batches are the correct 300 chip since the card was always advertised as not having a factory overclock. To be certain, check the Device ID in the main window of GPU-Z, if it shows 1E04 it is the 300 chip, if it shows 1E07 it is the "binned" 300A chip.
Question: How do I know what power limit my card has? Answer: In GPU-Z, click the Advanced tab, then General and finally choose NVIDIA BIOS, under the Power Limit section you will see Default and Maximum.
Question: What does the power limit actually do? Answer: Several years ago we had full voltage control, then NVIDIA introduced a power and temperature limit. Once a 300 card reaches its maximum 280-310W power usage it will restrict the voltage: for instance, my 280W card could not exceed 0.913v running a game in 5K resolution, this meant it could not reach a higher than 1860MHz core clock although the GPU has a hard limit of 1.093V, therefore it's far from its true potential.
Question:How big is the performance gap between the two? Answer: Simply put, a higher power limit allows for a higher voltage, which in turn results in a higher overclock. While my 280W card could only reach 1860MHz at 0.913V, my other 330W card achieved 1980MHz at 1.018V, both at 75c running the same game in 5K resolution.
+18% power limit → +11.5% voltage → +6.5% core clock.
In further testing, trying to score as high as possible in ‘UNIGINE Superposition 1080p Extreme’ at watercooled temperatures, the 280W card (0.913v at 43c) got a score of 9188 (69 FPS), while the A card flashed to 380W (1.068V at 51c) got 10306 (77 FPS).
+36% power limit → +17% voltage → +12% performance.
At the time of writing this, the least expensive A GPU has a mere 4.4% price premium, using air cooling with a 330W power limit one can get up to 6.5% better performance, thus justifying the slight price bump. If you plan on running the card on water or with air cooling at the highest fan speed, the performance increase scales up to 12%.
Question: Is it safe to run the card at 380W? Answer: Yes, as shown above, with the higher power limit the voltage could not exceed 1.068V, the hardware limit is still 1.093V. In order to reach that voltage and unlock the full overclock potential it would require to surpass 400W. There is also a common misconception that the 8-pin connectors are capped at 150W each, in actuality the port does not define the power cap and is capable of more than 300W, making it over 600W for every 2x8-Pin card.
Question: Is it safe to run the reference PCB at 1.093V? Answer: Yes, as pointed out by Buildzoid during his analysis of the reference PCB: “If you're on water cooling or air cooling you have nothing to worry about, there is not much room for board partners to make a better PCB, for daily usage you can just buy a reference card and knock yourself out, really solid Vcore VRM, massive overkill on the Memory VRM, this is a really, really, really solid card”. To put things into perspective, the reference board can handle above 600W, at which point the GPU will have already hit the 1.093V limit a long time ago so no damage can be done.
Question: What about shunt mod? Answer: Shunt modding your card will increase the power limit, bypassing the software limit. This effectively means a guaranteed voltage of 1.093V, but is a high risk procedure, possibly voiding your warranty. I have no experience in removing a shunt mod so I cannot say how difficult it is. In short: not recommended unless you know exactly what you’re doing.
Question: Do the custom PCB cards such as the ROG Strix and FTW3 overclock better than the reference PCB cards? Answer: For daily use, the answer is no, the power limit is still the main limiting factor with overclocking. Long before you would've gained higher overclocking from the increased amount of power phases, the 1.093V limit would have already been reached.
Question: What about the temperature limit? Answer: This is where the custom PCB cards actually shine, not due to the extra power phases, rather the air cooling capacity. For instance, the massive MSI Gaming X Trio, being the second largest card with its 327mm length, 2.7 slot form factor and 3 fans, managed to keep a temperature of only 51°C (124°F) just above 1900MHz, locked to 0.900V with the fans set to 100%.
Meanwhile, the Palit Dual, which is a 292mm 2.5 slot dual-fan card, barely managed to keep it at 60°C (140°F), even after replacing the TIM with TG Kryonaut.
Thus we can conclude that a temperature delta of 9°C (48°F), which is a step higher on the clock curve, meaning 15MHz faster (not even 1%), does not matter for daily use, yet it's the polar opposite for attempting a high benchmark score.
The real utility I found for these massive coolers, is allowing for near silent gaming, the MSI Gaming X Trio managed to stay at 75c, running 1980MHz (1.043V) on the core while gaming at 5K resolution, with the fan speed at a mere 45% (1530/1220 RPM).
It made less noise than my 4x 120mm 850RPM radiator fans cooling the CPU, thus we’re getting watercooling noise levels from an air cooled card, running at a higher clock than the fastest factory overclocked card! If that is not impressive, then I don’t know what is.
Question: My card reaches 2200MHz at just 1.000V, am I lucky? Answer: No, since when the GPU is not fully utilized, it can freely go up to 1.093V, as it is not close to the power limit. To really find out your max clock, for your cooling (temperature), you need to run the most demanding benchmarks or games in up to 8K resolution which will result in low FPS and high VRAM usage.
I have seen many people running Time Spy, complain about reaching the power limit in Graphics Test 2 specifically, which is due to Time Spy being a 1440p benchmark that does not max out the GPU throughout the tests.
To really put your card to its knees, run Time Spy Extreme (4K), which is far more demanding on the hardware, but is sadly locked away behind a paywall. My recommendation is to instead run UNIGINE Superposition 8K Optimized preset.
Question: How much performance will I gain by water cooling my card? Answer: A lower temperature allows the GPU to increase the core clock in steps of 15MHz, roughly at about every 10°C (50°F), so keeping your card at 40°C (104°F) instead of 75°C (167°F) is likely about 4 steps, therefore 60MHz higher, which translates to 3% higher performance. This is of course only possible if you do not hit the power limit first!
Question: Are the more expensive cards binned (hand-picked)? Answer: There is no evidence supporting that. For example, the Galax OC Reference PCB 380W card could very well overclock higher than a EVGA FTW3 Custom PCB 373W card, it’s mere chance. This is why there is no point in choosing an FTW3 over the cheapest A GPU if you are going to water cool it, they will have near identical performance if you use the Galax 380W BIOS on both.
Question: So which card should I buy? Answer: The first thing to look at is the power limit, which rules out any Non-A GPUs as they are all locked to 280W and cannot be modified. Another thing to look for is the cooling, the obvious answer would be a hybrid variant, featuring a closed liquid cooler, unfortunately they tend to be far more expensive.
When looking for a GPU to cool with a water block it does not really matter what you get, as long as it is a reference PCB variant, there is no need for a custom one.
Finally, we have the air cooled GPUs, the bigger the better generally, triple-fan is recommended over dual-fan, but price is also an issue here, the large custom PCB cards tend to cost a lot more.
I would say that a good compromise would be the 2 to 2.5 slot triple-fan reference PCB variants, such as: Gainward Phoenix GS, Gigabyte Windforce OC, Inno3D X3 OC, MSI Duke OC, PNY XLR8 OC, Zotac Triple Fan and Zotac AMP.
The GPU with the highest power limit out of these is the Gigabyte Windforce OC, at 366W, but all of them can safely be flashed with the Galax 380W BIOS.
08-20-2018 11:23 AM
I will order Asus or Ref :)) Hope i get more then 40+% then my 1080ti on non ray-trays games)
08-20-2018 11:34 AM
I got my Founders Preorder in!
980 Ti needs to retire
08-20-2018 11:35 AM
08-20-2018 11:38 AM
guys off price is 1199?? or 999?
08-20-2018 11:39 AM
Pre-ordered the EVGA on Amazon.
I'm custom loop though, so will need to wait for a block. I'm almost thinking of cancelling and waiting until a block is actually available to buy but then I'm worried about the damn thing not being available anywhere.
ANY info on blocks for these things yet?
08-20-2018 11:41 AM
Originally Posted by Madness11
guys off price is 1199?? or 999?
FE variant is OC version and it is a 1199$, "non FE" Ti variant is 999$.
08-20-2018 12:00 PM
I officially have a pre-order for an nvidia founders edition.
08-20-2018 12:00 PM
Pre-ordered the MSI GeForce RTX 2080 Ti Gaming X Trio. But might switch to an EVGA if I can find the right model.
Edit: Pictures after getting the card. Click here or here for many more!
Anyone else find it interesting that NV-Link is supported for these cards? Will this have any significant benefit over SLI apart from bandwidth? Or is it something developers will still have to explicitly support?