Hi,
are there any benchmarks that show how much newer features, architectures and istructions improved the final modern cpu speed at the same clock with or without supporting all those new features (sse etc), at the same clock?
I'd like to understand how better new cpus are compared to the older one (let's say after 2000) when not helped by multicore or newer techs.
Thank
The tl;dr is that yes there are benchmarks, but there doesn't exist a universal one. Each CPU is good at doing their own thing in their own slice of time. Early 486 and 386 CPUs were 2D beauties. supporting anything that existed within a flat plane. Commander Keem and Civ welcome! The DX2/4 CPUs were good at early 3D, with my DX4-100 (OC to 120Mhz) running Duke 3D at fairly decent settings.
The Pentiums were great at running later 3D games like Quake and Tomb Raider, due to their excellent FPU performance compared to the previous generation. The 3dfx cards started to make a difference, with ATI, NVidia, Matrox, and S3 hopping on at different points in the race, each with their own added performance, meaning the CPU was less of a factor when considering gaming performance.
Newer and better memory and hard disks were a thing. Prices went way down, and the overclocking game really took shape. Pentium 4s started the first go into x64, a vast improvement over the previous PAE technology.
There really is no definite answer to which numerical aspects of a CPU makes it better, and benching is not always that great. There exists no universal benchmark for single core CPUs and their features, and most later benchmarks take the CPU in as a second number instead of the main deal.
Phil does have a video comparing the 450Mz Pentium 2 and 3, which does give some idea of what this stuff is like.
image in this article:
https://www.nextplatform.com/2016/05/05/long- … eon-processors/
says that broadwell core is about 1.8x faster than dothan core
wrote:The tl;dr is that yes there are benchmarks, but there doesn't exist a universal one. Each CPU is good at doing their own thing i […]
The tl;dr is that yes there are benchmarks, but there doesn't exist a universal one. Each CPU is good at doing their own thing in their own slice of time. Early 486 and 386 CPUs were 2D beauties. supporting anything that existed within a flat plane. Commander Keem and Civ welcome! The DX2/4 CPUs were good at early 3D, with my DX4-100 (OC to 120Mhz) running Duke 3D at fairly decent settings.
The Pentiums were great at running later 3D games like Quake and Tomb Raider, due to their excellent FPU performance compared to the previous generation. The 3dfx cards started to make a difference, with ATI, NVidia, Matrox, and S3 hopping on at different points in the race, each with their own added performance, meaning the CPU was less of a factor when considering gaming performance.
Newer and better memory and hard disks were a thing. Prices went way down, and the overclocking game really took shape. Pentium 4s started the first go into x64, a vast improvement over the previous PAE technology.
There really is no definite answer to which numerical aspects of a CPU makes it better, and benching is not always that great. There exists no universal benchmark for single core CPUs and their features, and most later benchmarks take the CPU in as a second number instead of the main deal.
Phil does have a video comparing the 450Mz Pentium 2 and 3, which does give some idea of what this stuff is like.
It would be nice to see how a single modern core clocked at let's say, 500Mhz will perform on an old game let's say Quake II with just software renderer to a time correct 500Mhz cpu and with a time correct 2D only card. 😉
http://kva.pl/index.php?link=artykul&id_a=6
This could be interesting for you.
I remember doing a 1GHz comparison between a Coppermine PIII and a San Diego Athlon 64 years and years ago. On average, the K8 @ 1GHz was around 1.9x faster than the Coppermine.
I've also done some single-core comparisons between a Socket 939 Opteron 185 @ 3GHz and an i7-4930K at 4.5GHz. Taking into account the difference in clock speed, the i7 is roughly 2.8x faster than the Opteron, clock for clock.
Assuming perfect clock speed scaling, a single-core 4930K @ 1GHz would be 5.3x faster than a 1GHz Coppermine.
94 MHz NEC VR4300 | SGI Reality CoPro | 8MB RDRAM | Each game gets its own SSD - nooice!
wrote:http://kva.pl/celeron_vs_celeron/
This could be interesting for you.
Interesting, thank! Too bad that they all seems to be modern software that probablyhave the cpu features optimization (sse_x) that help more than the basic architecture. Also I don't understand if the ram speed was clocked at same clock or not. If one had DDR2 @ 533Mhz it would be nice to see it clocked down to the equivalent 133Mhz of the SDRAM.
A very old (but still heavy) game like without any 3dnow,sse,etc support would be a nice thing to understand if that impressive speed difference would be similar.
wrote:wrote:http://kva.pl/celeron_vs_celeron/
This could be interesting for you.
Interesting, thank! Too bad that they all seems to be modern software that probablyhave the cpu features optimization (sse_x) that help more than the basic architecture. Also I don't understand if the ram speed was clocked at same clock or not. If one had DDR2 @ 533Mhz it would be nice to see it clocked down to the equivalent 133Mhz of the SDRAM.
A very old (but still heavy) game like without any 3dnow,sse,etc support would be a nice thing to understand if that impressive speed difference would be similar.
Not every tested program was using SSE etc. For example Super Pi doesn't use any. Also BOINC test shows results with and without modern instructions. It is not possible (and would be pointless for at least a few reasons) to test this CPUs with the same memory type or speed. I wasn't able to make any 3d tests because Celeron 220 mono doesn't have AGP 🙁
wrote:wrote:wrote:http://kva.pl/celeron_vs_celeron/
This could be interesting for you.
Interesting, thank! Too bad that they all seems to be modern software that probablyhave the cpu features optimization (sse_x) that help more than the basic architecture. Also I don't understand if the ram speed was clocked at same clock or not. If one had DDR2 @ 533Mhz it would be nice to see it clocked down to the equivalent 133Mhz of the SDRAM.
A very old (but still heavy) game like without any 3dnow,sse,etc support would be a nice thing to understand if that impressive speed difference would be similar.Not every tested program was using SSE etc. For example Super Pi doesn't use any. Also BOINC test shows results with and without modern instructions. It is not possible (and would be pointless for at least a few reasons) to test this CPUs with the same memory type or speed. I wasn't able to make any 3d tests because Celeron 220 mono doesn't have AGP 🙁
But with a PCI good old video card they would be at the same level of graphic speed and memory too. I know it would not be at their best but in a theoric CPU only test it would be interesting. 😀
I have bought a GT 8500 Sparkle PCI few weeks ago so maybe I will do part II of this comparison some day. Now I am working on Pentium Pro 256kb cache L2 vs Pentium Pro 1MB cache L2 in the same set of tests as used with Celerons. This is taking some time cause for example single BOINC run takes almost ten hours. And I need three of them for every CPU...
