5 ms·

Take with a grain of salt. Comparing a single benchmark against an aggregate isn't particularly meaningful (although getting within a stone's throw definitely i

Take with a grain of salt. Comparing a single benchmark against an aggregate isn't particularly meaningful (although getting within a stone's throw definitely is). You can go through and cherry pick benchmarks and find plenty that obliterate that measurement. Better article title would be, "3900x appears comparable to 9980XE in single threaded workloads."

walshemj 7y ago
But single treaded was really the only area Intel used to have a lead in - which is interesting
- vbezhenar 7y ago
  Those nanometers are not just numbers. TSMC 7nm is ahead of Intel's 14nm. It's not really surprising. Even if AMD can't catch up with frequencies, they have more transistors, so they can increase many things. Better comparison would be with 10nm Intel, but Intel failed there, so I expect AMD to lead until Intel will release mainstream 7nm CPUs.
arcticbull 7y ago
The 3900X already crushes in multi-threaded. That stone-throw multi-core result result is 12 AMD cores on vs 18 Intel cores. That's 50% more cores and 300% more cost for basically the same results. Single-threaded is the only thing Intel's been touting as they get steamrolled. This is particularly important because a big market for this class of CPU is desktop gamers and for better or worse games tend not to be particularly well multi-threaded so single-core performance is king in that market. Obviously this will change over time as engines continue to adapt. To your point though it's better to look at an overall trend of AMD vs Intel over the last couple of years, and that's equally devastating.
- holy_city 7y ago
  What I mean is that it's a sample size of 1. I can go through the 9980xe results and cherry pick a dozen iterations that crush this single sample. Even looking at the leaked 3600 benchmarks listed on Geekbench and there's some pretty gnarly variance. The benchmarks are extremely promising, and I'm likely going to be an AMD customer in a month, but these articles are not grounded in statistical significance. Wait until the embargo is lifted before posting stuff like this, imo.
- deogeo 7y ago
  > Obviously this will change over time as engines continue to adapt. They're already fairly well adapted - Xbox One and PS4 both have 8 cores for a reason.
mrb 7y ago
Calling Geekbench a "single benchmark" is misleading because it runs 25 different types of workloads. So AMD doing well here means it does well compared to Intel in 25 different areas when they are all averaged: • Cryptography Workloads - AES • Integer Workloads - LZMA Compression - JPEG Compression - Canny - Lua - Dijkstra - SQLite - HTML 5 Parse - HTML 5 DOM - Histogram Equalization - PDF Rendering - LLVM - Camera • Floating Point Workloads - SGEMM - SFFT - N-Body Physics - Ray Tracing - Rigid Body Physics - HDR - Gaussian Blur - Speech Recognition - Face Detection • Memory Workloads - Memory Copy - Memory Latency - Memory Bandwidth Source: https://www.geekbench.com/doc/geekbench4-cpu-workloads.pdf https://www.geekbench.com/doc/geekbench4-cpu-workloads.pdf
- holy_city 7y ago
  Sorry for the confusion, I meant "single benchmark" as in "single run of Geekbench." As in comparing one measurement versus an aggregate of measurements isn't particularly meaningful, you need to compare aggregate to aggregate to get something valid. It's still a sample size of 1, even if the sample has 25 dimensions. (I don't know if Geekbench uses an average or what they're weighting is e.g. with memory speeds and operating systems, which is why I'm saying "aggregate").
  - MegaButts 7y ago
    I'm all for reproducing results, but given the relatively deterministic nature of computers (especially for new hardware) I feel this is an unwarranted criticism. How often do people get significantly different results when running identical tests on identical hardware?
    - holy_city 7y ago
      Just looking at the first page of results for the 9980xe, it's pretty common, especially for multithreaded benchmarks: https://browser.geekbench.com/v4/cpu/search?q=9980xe https://browser.geekbench.com/v4/cpu/search?q=9980xe But the big qualifier is "on identical hardware." The geekbench score in the article is not a measurement on the same hardware, it's an aggregate of measurements across lots of hardware (different memory clocks, motherboards, etc). That's why a single run is not particularly useful, it just tells us what that user got on that day with that hardware. We need lots of data points to tell us more or less how it will compare on most builds. Again, not to diminish how promising the score is for the product. Hats off to AMD. I'm just waiting until there are more data points to be confident.

walshemj 7y ago

But single treaded was really the only area Intel used to have a lead in - which is interesting

vbezhenar 7y ago
Those nanometers are not just numbers. TSMC 7nm is ahead of Intel's 14nm. It's not really surprising. Even if AMD can't catch up with frequencies, they have more transistors, so they can increase many things. Better comparison would be with 10nm Intel, but Intel failed there, so I expect AMD to lead until Intel will release mainstream 7nm CPUs.

arcticbull 7y ago

The 3900X already crushes in multi-threaded. That stone-throw multi-core result result is 12 AMD cores on vs 18 Intel cores. That's 50% more cores and 300% more cost for basically the same results. Single-threaded is the only thing Intel's been touting as they get steamrolled. This is particularly important because a big market for this class of CPU is desktop gamers and for better or worse games tend not to be particularly well multi-threaded so single-core performance is king in that market. Obviously this will change over time as engines continue to adapt. To your point though it's better to look at an overall trend of AMD vs Intel over the last couple of years, and that's equally devastating.

holy_city 7y ago
What I mean is that it's a sample size of 1. I can go through the 9980xe results and cherry pick a dozen iterations that crush this single sample. Even looking at the leaked 3600 benchmarks listed on Geekbench and there's some pretty gnarly variance. The benchmarks are extremely promising, and I'm likely going to be an AMD customer in a month, but these articles are not grounded in statistical significance. Wait until the embargo is lifted before posting stuff like this, imo.
deogeo 7y ago
> Obviously this will change over time as engines continue to adapt. They're already fairly well adapted - Xbox One and PS4 both have 8 cores for a reason.

mrb 7y ago

Calling Geekbench a "single benchmark" is misleading because it runs 25 different types of workloads. So AMD doing well here means it does well compared to Intel in 25 different areas when they are all averaged: • Cryptography Workloads - AES • Integer Workloads - LZMA Compression - JPEG Compression - Canny - Lua - Dijkstra - SQLite - HTML 5 Parse - HTML 5 DOM - Histogram Equalization - PDF Rendering - LLVM - Camera • Floating Point Workloads - SGEMM - SFFT - N-Body Physics - Ray Tracing - Rigid Body Physics - HDR - Gaussian Blur - Speech Recognition - Face Detection • Memory Workloads - Memory Copy - Memory Latency - Memory Bandwidth Source: https://www.geekbench.com/doc/geekbench4-cpu-workloads.pdf https://www.geekbench.com/doc/geekbench4-cpu-workloads.pdf

holy_city 7y ago
Sorry for the confusion, I meant "single benchmark" as in "single run of Geekbench." As in comparing one measurement versus an aggregate of measurements isn't particularly meaningful, you need to compare aggregate to aggregate to get something valid. It's still a sample size of 1, even if the sample has 25 dimensions. (I don't know if Geekbench uses an average or what they're weighting is e.g. with memory speeds and operating systems, which is why I'm saying "aggregate").
- MegaButts 7y ago
  I'm all for reproducing results, but given the relatively deterministic nature of computers (especially for new hardware) I feel this is an unwarranted criticism. How often do people get significantly different results when running identical tests on identical hardware?
  - holy_city 7y ago
    Just looking at the first page of results for the 9980xe, it's pretty common, especially for multithreaded benchmarks: https://browser.geekbench.com/v4/cpu/search?q=9980xe https://browser.geekbench.com/v4/cpu/search?q=9980xe But the big qualifier is "on identical hardware." The geekbench score in the article is not a measurement on the same hardware, it's an aggregate of measurements across lots of hardware (different memory clocks, motherboards, etc). That's why a single run is not particularly useful, it just tells us what that user got on that day with that hardware. We need lots of data points to tell us more or less how it will compare on most builds. Again, not to diminish how promising the score is for the product. Hats off to AMD. I'm just waiting until there are more data points to be confident.