The scores on the Passmark CPU are noticeably different now compared to before.
The scores on the Passmark CPU are noticeably different now compared to before.
Many people have observed a notable shift in CPU benchmark ratings on Passmark. Over recent years, the i7-4790k—once a strong performer—has seen its score drop from around 11,000 to about 8,600 as of early March 2020. This represents a significant change compared to its stable performance over nearly six years. I captured the original results in December 2019 to illustrate the contrast. It’s quite disheartening since this metric helps customers evaluate whether upgrading older hardware is still beneficial. Anyone have insights on whether Passmark has updated its testing methods recently? Additionally, it appears newer Intel and AMD models show much smaller score variations (4%-15%), while older generations see larger swings (22%). You can view the original data here: [link] and the archived version here.
I also noticed this. I've been looking for used CPUs and found several that caught my eye before they changed much. But when I checked them again yesterday, many were performing much worse than expected. I haven’t found any clear reason for this. It doesn’t seem to be happening all at once either. My i7-3820 was rated at 8990 in October, and I remember thinking it was about 7400 yesterday—though I don’t have proof. Now it’s around 6000.
I've been observing my own performance and it's becoming quite unsettling, to say the least. Much like rogerswb, I've been examining various used CPUs to start some builds—Xeons, older i3/5/7 cores, even some AMD models. My current FX-8350 serves as a benchmark for comparison. Just a week ago, it was rated at 8949, but now it stands at 6246—a drop of about 30%. The Xeon E5620, which I recently acquired, was scoring 4770 in December 2019 and has fallen to 3103 today, a 35% decline. The Xeon X5570, previously at 5615 on Dec 11, 2019, now sits at 3640, marking a 35.2% drop. The E5-2620, which was 7904 on December 12, 2019, is down to 5138, a 35% decrease. i3-550, once scoring 2829 in late 2019, now has 1687—a 40.4% reduction. The Core 2 Quad Q6600, which reached 2946 on Dec 11, 2019, is now at 1590, a 46% drop. The Core i9-9980XE, previously at 29,576, has slipped to 28,160—a 4.8% decline. Interestingly, the AMD Athlon 200GE shows a much smaller change: from 4952 on Dec 13, 2019, it's now 4482, just a 9.5% drop. The Threadripper 3970X, which scored 61,454 on March 12, 2020, is down to 60,980—a barely noticeable 1% change. On the other hand, the Ryzen 7 2700X improved from 16,949 to 17,970—a 6% increase. What’s puzzling is the inconsistent patterns across these machines. It seems like numerous benchmarks have been updated with suboptimal configurations. This raises questions about the reliability of these scores and their true significance.
passmark evaluates CPUs based on specific performance metrics rather than averaging scores across similar products. It doesn’t simply take the average of all comparable SKUs. As new CPUs are released, their scores naturally rise, which can affect averages without necessarily reflecting real-world reliability. People still rely on passmark because it provides a clear benchmark, though newer guidelines like Plex’s suggest considering factors beyond just raw scores.
Me too; the numbers really changed. I’ve been working mainly with Xeons and saved a lot of scores in spreadsheets. The slight shifts in single-thread results and the minor sample count differences caught my attention. I initially thought benchmark standards might have shifted, pulling old scores into newer ranges. Newer chips shouldn’t affect older ones much, unless there was some manipulation. For instance, the E5-2687W dropped from 14,341 to 7,526, while the E5-2690 fell from 13,723 to 9,970. That’s huge—almost a fifty percent drop for the 2687W. It’s strange seeing such large variations even in lower-power models like i5s, with changes seen in generation 2, 3, and 4. I’m wondering if we should reach out to Passmark for clarification.
Before now it worked enough for my needs (remember I don’t run any gaming systems, just the charts), but this oddity requires clarification. Of course, I avoid any "gaming" mechanics or similar tricks. As I’ve mentioned before: "I figured out how much I lost on computer games—just a quarter of a cent—because I played Pong in an arcade back in the early '70s." I’m more into real life...
It's true that consistently testing a system was always difficult. Outcomes shifted with each update to match modern processors and software. This might be an educated guess, but it helps explain the recent scoring changes. HWBot has adjusted this frequently, though most benchmarks remain stable... The main change was the point system itself.