R-Batches, 13900KSS and IMC regressions? – Intel Core 14th Gen Binning Results from almost 600 CPUs

1 - Procedure and Disclaimer 2 - i9-14900K and i9-14900KF 3 - 6 GHz VID for 14900K(F) vs. 13900KS – Which CPU is better? 4 - i7-14700K and i7-14700KF 5 - i5-14600K and i5-14600KF 6 - i9's from 2022 and 2023 - Which vintage is better? 7 - IMC-Tests, SP vs. Cinebench and Conclusion

i9-14900K and i9-14900KF

Let’s start with the new top SKUs i9-14900K and i9-14900KF, but that doesn’t mean that the next few pages will be boring – quite the opposite. we got our hands on 154 chips of the K variant and 70 of the KF version – without iGPU – for a total of 224 i9s. The worst overall SP is 90, the best 112 and on average we have a value of 98.6.

While the KF SKU had a slightly higher SP in the last binning of the 13th generation, this time it is exactly the opposite with the 14 chips. Could this mean that Intel is not holding back high-quality silicon with iGPU for a KS SKU this time? Ultimately, our test quantity is too small to draw any really reliable conclusions, but the inverse ratio is interesting nonetheless.

For the SP values of the P Core, we start at 99 and end at 121, a spread of 22 points analogous to the total SP. The average here is 107.7. As the P Core SP is weighted in a ratio of 2:1 to the E Core SP, the distribution in the diagrams of P and total SP is very similar. However, the KF SKU ends with the highest value at 115, while the K SKU adds another 6 points. In other words, 19 of the 14900K CPUs are better than any other 14900KF CPU, although we only had just under half of these as a test set.

With the E core SP, the distribution in the diagram is most similar to a Gaussian curve, whereby the 14900K has more outliers at the top and bottom and the KF variant is more in the middle. The worst E cores of an i9 CPU have a Silicon Prediction value of 68, the best 95, a spread of 27 points. The average is relatively close to the mean value at 81.5.

Before we delve deeper into the science of SP calculation, I would like to mention the tested batches. The structure of the batch numbers has remained the same, explained using the example of X338Q171:

• one letter for the production location:
  • X: Vietnam (in this case)
  • V: China
  • L: Malaysia
• 3 numbers for the year and week of production (in this case year 2023, week 38)
• a counter consisting of one letter and 3 digits

This batch number and a 5-digit numerical PCB number can then be used to individually identify each CPU within an SKU. These 3 pieces of information, i.e. SKU, batch and PCB number, are encoded in this QR code, which you can find on the IHS and on the PCB or package of each current CPU.

Because only bar charts are boring, I have now thrown the whole thing into a bubble chart, where the size of the bubbles represents the number of tested CPUs from this batch and the position along the axes represents the quality based on the SP values. The top line of the label is the actual batch number, the second line is the number of CPUs tested in the batch.

we had 6 batches of the 14900K or KF SKUs available, each with between 14 and 68 CPUs. As experience shows that the PCB numbers within a batch range from single digits up to approx. 5000, it can be assumed that a complete batch contains approx. 5000 CPUs. We are therefore only likely to have tested a maximum of 10% of a single batch, which in turn makes the comparison of batches with each other even less meaningful.

Nevertheless, one observation is unmistakable in our data. Batch X337R173, i.e. where the counter starts with an “R”, is significantly better than all other i9 batches, by over 6 points in the overall SP. To put it anecdotally: after half a day of binning, the first CPU from this batch was better than all the other almost 100 i9 CPUs that I had tested that day. This phenomenon, where one “R batch” is significantly better than all the others, will also happen with another SKU. But I don’t want to say too much in advance – you’ll have to scroll through the pages for that.

Of course, we can’t say for sure whether batches with the letter “R” in the counter are really particularly good, or whether a letter as late as possible in the alphabet indicates a higher silicon quality. Despite everything, our test set of CPUs is too small for this and we would simply need a lot more data. So if you happen to have such an “R” CPU yourself and can read out the VID or SP values, please share your results. In neighboring German hardware forums, the “R” batch scavenger hunt is already in full swing! 😀