holyhyperion
Carol1134
Ran a Time Spy test in stock settings and gets 21461,only 2%-3% better than my KPE 3090,and the clock was 2070(2043avg)
Would they allow me to start an RMA?I'm thinking I received a very bad simple.
Perhaps Sajin can comment on this, but no, I don't think you will be allowed to RMA your card for this. If you look on the 3090 TI KPE the EVGA website under the "Specs" tab, EVGA only guarantees a 1950 MHz boost clock. Anything higher than the guaranteed 1950 MHz boost clock is technically overclocking or running the card "out of spec." EVGA does not guarantee overclocking capability or capacity in any form. By your own admission, your card at "stock clocks" is clocking a boost of 2043 average, which is almost 100 Mhz higher than the guaranteed EVGA boost clock of 1950 Mhz on this SKU. No way EVGA would allow an RMA on that.
This is called playing the Silicon Lottery. Some cards are binned extremely well. I believe Sajin's 3090TI BLACK (not even FTW3 Ultra) can hold OVER 2250 Mhz in PR/TSE. On the other hand, some 3090 TI KPE's can only do 2050 Mhz.
My own 3090TI KPE can only hold 2175-2190 Mhz on PR and Time Spy Extreme. It can do +1000 Mhz on the memory. I have a separate Asus ROG Strix LC 3090 Ti that can consistently hold 2205 Mhz on PR/TSE and +1400 MHz on the memory. Technically, my Asus ROG Strix LC 3090 TI performs slightly better than my KPE 3090 TI.
In short, we both played the Silicon Lottery and we "lost." I.E., we got a card that seems to clock at average or slightly below average.
It will be extremely difficult to truly know what the bins are without taking a wide survey with a large enough sample size (at least 50). Who knows, maybe the "true average" bin of the 3090 TI chip is 2150Mhz for maximal overclock potential. Also, there is probably selection bias for the card-owners willing to take such a survey, as most people who purchase these cards don't use MSI afterburner or PX1 and don't bother posting on forums. They just run the cards and play their games or use their GPU accelerated applications.
I can guarantee you 100% however, Carol1134, that EVGA absolutely does NOT bin these chips for the different SKU's. When you purchase a 3090 TI KPE, you should be fully informed about your decision. You are NOT purchasing a better binned chip, 100%, full stop, period. The difference between the cheapest SKU (FTW3 Black aka Sajin's card) and the most expensive SKU (KPE 3090 TI) is only 90 Mhz in the boost clock. Your extra $500 (FTW3 Black at $1499 vs KPE 3090 TI + PSU at $1999) is going to the special custom PCB on the KPE. The KPE has the dip switches, the extra 12-pin GPU Power plugin for an extra 450W of PSU power, and the ability to use the card for XOC without having to apply any shunt mods. The KPE PCB is ready to "plug and play" for LN2 XOC without having to do any additional modifications. The LN2 bios allows you to bypass any temperature safeties. All you need to do is just strap on an LN2 pot and you're ready to go. Before the EVGA discounts, you could also argue that part of the extra $500 was going to the AIO cooler, but that argument is now void given you can purchase an FTW3 Ultra Hybrid 3090TI for the same $1499 price point.
In short, your KPE 3090Ti has a special PCB that makes it much easier and straightforward to do XOC. You also get the KPE Edition "e-prestige." But please be aware that EVGA absolutely does not bin these chips for performance. Virtually every 3090TI chip can do 1950Mhz boost clock, which is all that EVGA guarantees you in terms of performance.
OK so I've seen this theory parroted around. I'm going to share what I know about this after talking with several people who are in the know. EVGA absolutely DOES bin the chips in their Kingpin models. Generally they just check V/F scaling and take the top couple (5-10%? exact % is dependent on how many they want to make) of chips and put them in KP. The problem is the silicon Nvidia sends to AIBs is not all of equal quality. Some batches are significantly weaker than older batches, and in general GPU dies trend weaker with time. This is a well known phenomenon in XOC. The best cores generally come very very early on in the production cycle and again trend weaker over time. Some of this is down to Nvidia Keeping the best V/F curve dies for themselves to put in their Quadro/A6000 GPUs, and some of it is likely due to yields and manufacturing tolerances etc. But the point is, EVGA does indeed bin these cards. But they can only bin what Nvidia sends them. If the batches are weak, then even the best dies in those batches might still be weaker than GPUs from stronger bins that didn't make the binning threshold.
If you want to know your GPUs general bin Go into precision X1 after a reboot and without touching anything else hit the boost lock button. The higher the frequency displayed, generally the better the card is in terms of silicon quality. You can also look at your V/F curve in MSI Afterburner and find the the highest frequency point that the GPU is set to achieve by default. You can think of this similar to how you might think of an SP rating on Intel CPUs with ASUS boards. However (like SP ratings) there is still definitely alot of variation so that's why its important to also test max overclocks and performance.
However bear in mind that this "Boost Bin" number this is affected by temperature, so you need to be aware of which temperature thresholds cause Nvidia's GPU boost to kick in and downclock the cards so you can isolate for temperature. Bios can also play a slight factor in boost clock but there is no general rule to follow to isolate and account for this. This Bin also only works for core quality. Memory is another thing entirely and has to be checked independently. But if you look at Gamers Nexus video when he tore down the 3090 Ti he notated that his FTW3 had a 2040 Bin as notated on the die by EVGA, and confirmed by EVGA Jacob on the phone. I personally have Several EVGA cards that have their Bin numbers hand written on the side of the die. Its clear evidence that EVGA is indeed "Binning" the GPUs. But buying a Kingpin does not guarantee you an absolute golden sample. There is still A LOT of variation but ALL the dies on KP cards are binned and are FAR above the average die within the particular batch the Nvidia sent EVGA at the time of production.
I have yet to see a 3090 Ti KP that isnt atleast a 2070 MHz binned chip. And 2085+ MHz is VERY common on them. Those are VERY strong cores compared to the current 3090 Ti Class of silicon being sold by AIBs. It seems as though the initial batch of 3090 Ti Silicon was very strong. Likely Nvidia had been binning the dies and building up stock before the official release of the product. Also it seems the initial batch of new GDDR6X memory chips were also had much better scaling than the newer batches. As a result the initial 3090 Ti cards overclocked MUCH better (2190+ on PR) than modern ones, many of which cant even sustain 2145 MHz. The memory on those early cards also overclocked much better (+1400-1500 was achievable on good samples) where as current cards can start artifacting with as little as +700.