Skip to content
TheChatArchiveThe Chat Archive
gpt-4.1-2025-04-14· OpenAIHugging Face LMArena Human Preference 140K

how fast do I have to go back in time for my iphone to outperform the best super…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Googlegemma-3-27b-it’s answer.
  1. User

    how fast do I have to go back in time for my iphone to outperform the best super computer of the time at supercomuting workloads (whatever those were at the time)

    don't just look at raw flops numbers, consider the workload

    if my phone can't do something because it lacks the memory and/or memory bandwidth, than don't count it

  2. Assistant

    Great question! To answer this, we need to:

    1. Pick a modern iPhone (let's use the iPhone 15 Pro, with the A17 Pro chip, as of 2024).
    2. Define "outperform": Your iPhone must complete the same supercomputing workload faster than the best supercomputer of the time, without being bottlenecked by memory or bandwidth.
    3. Consider real workloads: Not just peak FLOPS, but actual tasks supercomputers ran (e.g., scientific simulations, weather modeling, etc.).
    4. Account for memory and bandwidth: If the iPhone can't fit the workload in RAM or can't feed the processor fast enough, it doesn't count.

    Let's break it down:


    1. iPhone 15 Pro (A17 Pro) Performance

    • CPU: 6-core (2 performance, 4 efficiency), up to ~3.78 GHz.
    • GPU: 6-core Apple GPU.
    • RAM: 8 GB LPDDR5.
    • Peak performance: Apple claims up to ~35 TOPS (trillion operations per second) for the Neural Engine, but for general-purpose FP32:
      • CPU: ~1 TFLOP (estimate, FP32)
      • GPU: ~2.5 TFLOPS (FP32, estimate)
    • Memory bandwidth: ~100 GB/s (LPDDR5)
    • RAM: 8 GB

    2. Supercomputing Workloads

    Supercomputers run large-scale simulations (climate, physics, molecular dynamics, etc.), often requiring:

    • High memory (tens to thousands of GB)
    • High memory bandwidth
    • Parallelism (thousands of CPUs/GPUs)
    • Interconnects (for multi-node communication)

    But, for small enough workloads (that fit in 8 GB RAM), we can compare.


    3. Historical Supercomputers

    Let's look at the top supercomputers over time, focusing on those whose best-case workloads could fit in 8 GB RAM and not be bottlenecked by bandwidth.

    A. 2010: Jaguar (Cray XT5)

    • Peak: ~2.3 PFLOPS (LINPACK: 1.75 PFLOPS)
    • Nodes: 18,688, each with 2x 6-core AMD Opteron, 16 GB RAM/node
    • Memory bandwidth: ~10 GB/s per node
    • Workloads: Large-scale, but per-node, smaller jobs possible

    B. 2005: Blue Gene/L

    • Peak: 0.28 PFLOPS (LINPACK: 0.13 PFLOPS)
    • Nodes: 65,536, each with 2x PowerPC 440, 512 MB RAM/node
    • Memory bandwidth: ~2.8 GB/s per node

    C. 2000: ASCI Red

    • Peak: 2.4 TFLOPS (LINPACK: 1.3 TFLOPS)
    • Nodes: 9,632, each with 4x Pentium Pro, 128 MB RAM/node

    D. 1997: ASCI Red (initial)

    • Peak: 1.06 TFLOPS

    E. 1993: Numerical Wind Tunnel

    • Peak: 124 GFLOPS

    4. Comparing to iPhone 15 Pro

    A. Raw Performance

    • iPhone 15 Pro: ~2.5 TFLOPS (GPU, FP32)
    • ASCI Red (2000): 2.4 TFLOPS (entire machine)
    • Blue Gene/L (2005): 0.28 PFLOPS (280 TFLOPS, but per node: ~4 GFLOPS)
    • Jaguar (2010): 2.3 PFLOPS (2,300 TFLOPS, per node: ~120 GFLOPS)

    B. Memory

    • iPhone 15 Pro: 8 GB RAM, ~100 GB/s bandwidth
    • ASCI Red node: 128 MB RAM, ~1 GB/s bandwidth
    • Blue Gene/L node: 512 MB RAM, ~2.8 GB/s bandwidth
    • Jaguar node: 16 GB RAM, ~10 GB/s bandwidth

    C. Workload Fit

    • If the workload fits in 8 GB RAM, the iPhone can run it.
    • If the workload needs more RAM, the iPhone can't run it.

    5. When Does the iPhone Outperform?

    A. Node-to-Node Comparison

    • iPhone 15 Pro is much faster than a single node of any pre-2010 supercomputer, both in compute and memory bandwidth.
    • For small workloads (e.g., a simulation that fits in 8 GB RAM), the iPhone 15 Pro would crush a single node of ASCI Red, Blue Gene/L, or even Jaguar.

    B. Whole-Machine Comparison

    • ASCI Red (2000): 2.4 TFLOPS (entire machine)

      • iPhone 15 Pro: ~2.5 TFLOPS (GPU)
      • If the workload fits in 8 GB RAM, the iPhone 15 Pro could match or beat ASCI Red's entire machine on compute-bound tasks.
    • Blue Gene/L (2005): 280 TFLOPS (entire machine)

      • iPhone 15 Pro is much slower than the whole machine, but per node, it's vastly superior.
    • Jaguar (2010): 2,300 TFLOPS (entire machine)

      • iPhone 15 Pro is much slower.

    6. Realistic Workloads

    • Supercomputing workloads are often memory-bound and parallel, but for small enough problems (e.g., a small molecular dynamics simulation, a small weather model), the iPhone 15 Pro could run them faster than the entire ASCI Red (2000) or any earlier supercomputer, as long as the problem fits in 8 GB RAM and doesn't require massive parallelism.

    Conclusion

    If you go back to the year 2000, your iPhone 15 Pro could outperform the world's fastest supercomputer (ASCI Red) on any workload that fits in 8 GB RAM and doesn't require more memory bandwidth than the iPhone provides.

    • Before 2000: Your iPhone easily outperforms the best supercomputers, even for whole-machine workloads, as long as they fit in RAM.
    • 2000 (ASCI Red): Your iPhone is roughly equal to the whole machine in raw compute, and much faster per node.
    • After 2000: Supercomputers quickly outpace the iPhone in aggregate compute, but for small workloads, the iPhone is still faster than any single node until about 2010.

    So, you need to go back to the year 2000 (ASCI Red) for your iPhone 15 Pro to outperform the best supercomputer of the time on supercomputing workloads that fit in 8 GB RAM.


    Summary Table

    YearSupercomputerPeak Compute (TFLOPS)RAM (per node)iPhone 15 Pro Outperforms?
    1993Numerical Wind Tunnel0.124~1 GBYes (easily)
    1997ASCI Red (initial)1.06128 MBYes
    2000ASCI Red (upgrade)2.4128 MBYes (if fits in RAM)
    2005Blue Gene/L280512 MBNo (whole machine), Yes (per node)
    2010Jaguar2,30016 GBNo (whole machine), Yes (per node)

    If you want your iPhone to be the fastest supercomputer in the world for any workload that fits in 8 GB RAM, you need to go back to the year 2000.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.