ElI5: how is each CPU unique even tho they’re labelled the same? For example a Ryzen 9 5900X can be better than another Ryzen 9 5900X.
In: Technology
When they make the chips they aim for them all the be the top tier in the class they’re batching. But the scale and complexity is so enormous that microscopic defects can degrade quality.
So they test them chips.
All the chips that meet the criteria for Tier 1 gets batched into Tier 1
All the chips that meet the criteria for Tier 2 get batched into Tier 2, etc.
This might also include disabling cores or other functions.
There is a small range between tiers, so it’s possible for a chip to *just* fail Tier 1 to be batched with one that *just* passed Tier 2.
You can also end up with uneven defects – eg two cores might be fucked on a chip but everything else is near perfect – they’ll disable the cores and batch it with the chips with the reduced cores/performance, even though the rest of the chip is a higher quality.
You can also get weird cases where they’ll intentionally downgrade working cores/chip ratings when there’s good demand for the lower tier versions and they don’t want to flood the market with the higher tier versions and push down the prices of the whole product line.
Like a couple people have mentioned before, its due to binning most of the time, there is a video done by one of Linus’ sub-channels that explain the concept extremely well, and should clear things up for you: https://youtu.be/8AQPIBfIqMk
Because of binning. (?) For instance, a ryzen 7 might have originally been a ryzen 9 but had some cores disabled and clock frequency reduced because of instability. They then label and sell it as a lower performing chip, meaning it may or may not have headroom for running faster than an “identical” chip model.
At least that’s what I think.
Depends on a number of factors and mostly external to the chip.
While there could theoretically be some differences in the quality of silicon, you’ll see performance differences from cooling and motherboard architecture.
For all intents and purposes the chips are identical.
The chip label represents the minimum performance that the manufacturer will guarantee. Parts are often built with extra circuits, so that if one fails during manufacturing the part can be salvaged. This reduces waste (and saves all the money spent making that “not quite perfect” part). These features are called yield enhancement, because they allow more of the not-perfect parts that get made to be sold.
However, that minimum isn’t the maximum you might be able to get with excellent cooling and careful overclocking. The manufacturer has to allow for only average cooling and enough margin that they don’t have to spend a lot of money on replacement parts.
When you’re dealing with things on such a small scale and pushing them to their limit the smallest variance becomes relevant and impossible to control. Tiny variations on material quality and manufacturing quality impact performance but the difference is so small they can only determine the difference once the chip is done and tested.
It’s called the silicon lottery. You may get a super weak CPU, a normal cpu, or even a MLG pro cpu in the same sku. But now days you either get a normal cpu or a more powerful cpu. I think they just toss the weak, underperforming dies now.
AMD did this back with the phenoms. 4 core die only 3 cores work now they sell it as a triple core. But they were also taking perfectly fine 4 cores and disabling the 4th core to make it a triple core. Multiple MB manufacturers had a option to enable these cores. But it was a lottery you might have a disabled core or you might have a dead core.
The issue with the ryzen 5000 series, is that the CPUs use an advanced “speed boost” technology, where the CPU selects the best possible speed, taking into account current conditions, as well as the manufacurting variations that happen when making a CPU.
There is natural variation when making chips – the manufacturing scale is so small, that there will always be some variation in the production process. Some chips, or circuits on a chip, might not work. Some chips (or individual circuits, like a CPU core) might be less energy effieicnt, and some might be more energy efficient. Some might run faster, some might only manage a slower speed.
Traditionally, CPU manufactures would test the chip under worst case conditions, and if it passed, then that would be the chip’s speed. If it didn’t it would be sold as a slower speed; if it couldn’t meet the slowest spec in the series, it would be scrap.
Manufacturers realised that worst case stress testing wasn’t always realistic, and under other types of work, there is less heat and power consumption, which means that the chip could be pushed a bit harder when running “easier” work, as long as it slowed down again when running “hard” work. This was a concept called “turbo” or “boost”. The early generations of boost technology worked on the basis of workload and temperature
The Ryzen 5000 series has a 2nd generation boost system, which will push the chip right to its limits at the current conditions, taking into account not just the program workload and temperature, but also the speed profile and energy efficiency of each individual core in the CPU (which is measured at the factory, and programmed into the chip), as well as the capability of the CPU’s power supply.
On the 5900X with 12 cores, if it is running maxed out on all 12 cores, the limiting factor is the power supply. The CPU slows down the cores as needed, taking into account the energy efficiency of each individual core, and the nature of the program, so that the total power consumption of the CPU, is close to the CPU’s maximum allowed power capability, without going over.
What this means is that if 1 specimen of a 5900X is more energy efficient, then it will be able to run faster, because the boost system will always adjust the all-core speed to be just at the limit of the power supply.
Latest Answers