Comparison
Compare
Two to four records side by side, one comparability group at a time. A multiplier appears only where the rules allow one, and where they do not, the reason is written out.
Prepared comparisons
-
2 records · 3 multipliers
Two guidance computers, one clock
The Apollo Guidance Computer and the Saturn Launch Vehicle Digital Computer ran at the same 2.048 MHz and were not the same speed.
-
3 records · 4 multipliers
A decade of Commodore clocks
Commodore 64, Amiga 500 and Amiga 1200 in their PAL forms. Their clock rates rose fourteen-fold across ten years, while their memory capacities remain separate measurements.
-
4 records · 9 multipliers
Four consoles, four clocks
PlayStation, Nintendo 64, Dreamcast and PlayStation 2 form the widest comparison the catalog supports. It shows a metric that survives four columns while its neighbors do not.
-
2 records · 3 multipliers
What a pre-launch specification costs
PlayStation 2 against PlayStation 3 compares consecutive consoles from one manufacturer. Their headline figures line up, but most of the PlayStation 3's were only ever announced.
-
2 records · 3 multipliers
One processor or two
The same SGI Octane in its single- and dual-R10000 configurations: identical clocks, twice the memory, and four SPEC results that still cannot be divided.
-
4 records · 0 multipliers
From GeForce 256 to RTX 5090
Four NVIDIA reference cards across twenty-six years: memory bandwidth and GPU clocks in their own rows, with no invented performance line through FLOPS or power declarations.
-
2 records · 2 multipliers
Forty years of home computing
A PAL Commodore 64 and an 8 GB Raspberry Pi 5 across forty-one years: valid CPU-clock and whole-system-memory multipliers that remain separate facts, not a performance score.
-
2 records · 3 multipliers
PlayStation 2 to H100
The PlayStation 2's Graphics Synthesizer and NVIDIA H100 SXM module: a valid 70× theoretical memory-bandwidth comparison, kept to one metric, method, and scope.
-
4 records · 0 multipliers
From 8086 to Core i9
Four representative Intel PCs across forty-five years: clocks and fitted RAM in full view, while provisional Dhrystone results do not support a speed score.
How a comparison is put together
A comparison holds between 2 and 4 records. They are either whole systems or components, never a mix, because a machine and one of its parts answer different questions. Figures from the machine and from the parts fitted to it are both shown, since a console's clock is recorded against the console while a workstation's benchmark result is recorded against its processor.
Every row is one comparability group: one metric, one method, one benchmark version and one device scope. Two figures obtained differently are two rows and never two columns of one row, however similar they look. Sections group like with like. Nothing combines across them, so there is no total or overall score.
Multipliers are measured against the first record and computed by the same versioned formula that produces the catalog's published derived claims, so a comparison assembled here and one recorded in the data are the same arithmetic under the same rules. Read how derived claims work.