I would be interested in a benchmark of a 2004 Pentium 4 3GHz HT against the single core performance of a 2016 i7 3 GHz. (To make it compareable, the memory modules have to be as slow as in the 2004 era, also the hdd.)
Well this isn't exactly a fair comparison, still it seems to indicate that overall (not for specific applications eg. graphics acceleration), the performance improvement is not that massive:
http://www.cpubenchmark.net/compare.php?cmp[]=1081&cmp[]=107...
Thats not really a good comparison as Intel was back in the game then (Core 2 processors were popular, and the Pentiums were just lower end versions of that keeping the brand alive).
The Pentium 4 519K from 2004, part of the Prescott (90nm) family, would be a good comparison. I think it was the first CPU that was easily overclockable to 4GHz, or in extreme cases 6Ghz (with enough water cooling).
Sadly it's not on that site (or at least I can't find it on mobile).
Since I can't edit my post anymore - the closest thing I could find was Cinebench R10 32 Bit (Single Core) comparisons (unfortunately without deactivated turbo boost and adjusted clock and RAM speeds):
It's sad to see that AMD could compete in single core performance until Sandy Bridge (i7 3930) came along. Would be nice if Zen can at least get Sandy Bridge numbers paired with good efficiency so we get more competition in the x86 space again. Numbers taken from: http://cpuboss.com/cpus/Desktop-CPUs-best-Overall-Performanc...
Yes, absolutely the price and price/performance from AMD is good (I would compare it against the 2600k (4 core/8 threads) though, which was 300 but still faster, or even the 2500k (4 cores/4 threads) which was ~230 but still faster).
Unfortunately as merb already mentioned the performance to power ratio is not really good for AMD, which I guess contributes quite a bit to AMDs lost marketshare in the server segment (plus it doesn't look good in reviews which is bad for AMDs mindshare, which already is far behind Intel and Nvidia). And until Mantle/Vulkan/DX12 the performance in quite a few games was noticable behind Intel because AMDs single core performance was not on par which was bad because DX11 is basically single threaded. [1]
Overall I guess AMDs biggest problem is the mindshare. Wheras Intel or Nvidia can allow themselves one or two not so great generations, AMD tanks as soon as they have problems. AMDs HD 5000 generation was out sooner than Nvidia, faster and more energy efficient, but it only helped them gain 5% marketshare (around 45% at the time). AMDs 7000 generation could compete with Nvidia, but used more energy and bang, Nvidia regained more than the 5% percent that they lost, and that continued with Maxwell (they now have ~75% marketshare even though AMDs cards can compete on performance). [2] The same with Intel, when AMD could compete, and even beat Intel in performance and efficency over many years, they gained not that much marketshare (in part because of Intels shady manipulation, but in part probably also because of mindshare). Intels Conroe could regain some of the marketshare Intel lost thanks to its Pentium 4 (which reduced AMDS ~48% to 35%). Then came Bulldozer, which was underwhelming outside of multithreaded tasks and late in part thanks to GloFos 32nm problems and now AMD has aroun 20% marketshare). In the server space it's even worse for AMD with less than 1% marketshare compared to 25% around 2006-2007.
Lies, damn lies, statistics, and benchmarks. I'm curious as to how much of the jerkiness and slowness in that test was due to comparing the integrated graphics on the P3 machine with dedicated graphics on the P4. Hooray for rigged demos.
The benchmark is definitely misleading because the Powerpoint workload is playing on a continuous loop. This means that when the test is over, the Pentium III ends up performing more overall work than the Pentium 4.
Any time you have operating systems running in a test of this kind, you're dealing with the same thing, a difference in work done the longer the test runs.
I do not think there was graphical acceleration for videos back then. It is probably doing well from the SSE2 instructions that Intel's marketing made certain the video player was using for decode. If it were not for that, the Pentium 4 would not have done as well.
There was. Way back in the day, I even had an mpeg-2 accelerator card that passed through a signal to do its magic. The card used in the P4 was a Geforce 2 GTS, which did have mpeg acceleration. The onboard graphics of the P3 accelerated certain primitives, like motion compensation, but not to the degree of the P4.
I think he meant that the P4 (netburst) architecture wasn't very good. It was optimized for high clock rates und utilized a long pipeline for that. I had some fellow students back in those days that overclocked their P4s to 4Ghz and beyond, which was quite insane compared to other architectures (P3 and AMD). However as a tradeoff the architecture was not very efficient.
With the Core2 line Intel then moved into a completely different direction with higher efficiency and lower clock cycles. Core2 was considered to be a massive success compared to P4.
The Pentium 4 architecture was a godsend for the overclocking scene. IIRC The P4 record is at 8.2GHz, with a Celeron 352 being in third place worldwide with 8.5GHz
I replaced a P4 with a C2550D4I for a friend last year. Where the P4 struggled with modern YouTube videos at 480p, the Avoton (an Atom) did 1080p without much trouble. It is odd that The P4 did so well in Intel's marketing video, but did not in 2015. My guess is that the decoder for the more modern video codecs require more than the P4 is able to do. It is also possible that the OS made a difference. His P4 ran Windows XP while the Avoton runs Gentoo Linux. Gentoo Linux should have had a decoder that could use SSE2 while Windows XP might not have.
That said, I was pleasantly surprised by the performance of the puny Aspeed graphics card with kernel mode setting. There is minimal tearing and the frame rates are smooth. The P4 had used a GeForce 3.
Conroe was almost two times faster than Netburst and Ivy Bridge was about 50% faster than Conroe. Skylake is another 15-20%, so I guess it would be about 3-4 times faster.
