I am sorry to tell you that HSW will actually only go to 4C/8T on the mainstream platform.
Well, I hope the i5 equivalent will have 4c 8t then.
I was talking Haswell.
That's really so sad. Intel introduced desktop quad-core CPUs at 65nm in 2007. Surely, by 22nm and 2012, it might be possible to squeeze another couple of cores in, don't you think?
There is no proof about anything Haswell yet (other than the new instructions), just speculation based on Intel marketing trends up until now.
So there "might" be a 6C/8C desktop part, or even a 4C/8T i5, but I would say that is less than 5% chance of happening.
Intel is pretty rigid with how they market their products. AFAIK, i3 = 2C4T, i5 4C4T and i7 4C8T and beyond. That is referring to the desktop parts which are quite consistent with what I said but the mobile parts are a little bit confusing at times because we could have an i5 part with 2C4T and the same can be said for the i7 mobile.I was talking Haswell.
Why would they? There is no need to do so yet. Software has not caught up to the "more cores" race of the last decade. I would say that 90% of PC users can live with 4 cores. And if you add HT to that, then all the better.
Then the remaining 10% of users could spend a little more and get the SB-E line with 6 cores.
I really don't see why people keep crying about "mainstream" not moving past 4 cores quickly. There are othe options.
Why would they? There is no need to do so yet. Software has not caught up to the "more cores" race of the last decade. I would say that 90% of PC users can live with 4 cores. And if you add HT to that, then all the better.
Then the remaining 10% of users could spend a little more and get the SB-E line with 6 cores.
I really don't see why people keep crying about "mainstream" not moving past 4 cores quickly. There are othe options.
Here's to hoping that PD & Steamroller are worth a crap...
Because I want them. Because what I do with my computers will use every thread they have without issue. Because these things need to be "mainstream" so they can be a cheap bundle @ microcenter.
Because I want my cake and I want to devour it too.
Here's to hoping that PD & Steamroller are worth a crap...
I'm glad intel is focusing on making better chips and not just MOARRR CORESS. More cores is akin to the GHz race in the late 90s/early 2000s. It's not just the quantity of cores but the QUALITY. So yes, the future definitely will be more cores, but lets keep improving those cores instead of just adding more. BD has how many cores and yet can't even deal with a 4c/4t i5-2500k ??
For the mainstream, the focus is on lower TDPs, more integration, and higher IPC. For high end, you get more brute cores.
Because I want them. Because what I do with my computers will use every thread they have without issue. Because these things need to be "mainstream" so they can be a cheap bundle @ microcenter.
Because I want my cake and I want to devour it too.
Here's to hoping that PD & Steamroller are worth a crap...
What's your point? AMD's cores are so weak and slow that it doesn't matter if they have more of them.
You have Intel, which has the Core i7 with four cores and SMT. You have AMD, which has eight integer cores and four floating point cores in a CMT architecture.
To set the baseline, one AMD core is 1.0x. In comparison, one Intel core is 1.5-1.6x. On top of that, when you stress the processor completely, resources need to be shared on the AMD core. That means each AMD core goes to being 0.8-0.9x, while Intel with SMT is able to improve performance by .2x, or 20%.
No wonder Intel is faster even in multi-threaded even with half the number of cores.
What's your point? AMD's cores are so weak and slow that it doesn't matter if they have more of them.
You have Intel, which has the Core i7 with four cores and SMT. You have AMD, which has eight integer cores and four floating point cores in a CMT architecture.
To set the baseline, one AMD core is 1.0x. In comparison, one Intel core is 1.5-1.6x. On top of that, when you stress the processor completely, resources need to be shared on the AMD core. That means each AMD core goes to being 0.8-0.9x, while Intel with SMT is able to improve performance by .2x, or 20%.
No wonder Intel is faster even in multi-threaded even with half the number of cores.
I am going to put forth that is only due to Intel's epic lead in manufacturing/process node maturity.
Intel currently has an approximate 18 month lead time on its smaller rival when it comes to moving to new process technologies, while its gross margin is 16 percent higher. When it comes to research and development, the gap between the two companies is enormous. Intel spent $3.9B on R&D in Q2; AMDs total revenue for the same period was $1.57B. What this means in the real world is that Intel can build more CPUs per silicon wafer than AMD can, earns more money per sale, and has a research fund thats nearly 2.5x AMDs quarterly income.
Historically, AMD has ignored these factors and designed cores that were meant to go toe-to-toe with Intels best in terms of single-threaded performance. This has proven to be an ineffective strategy *. AMD has never been able to retain any performance crown it took from Intel, and the cost of attempting to do so nearly killed the company. Bulldozer breaks with these trends. Its explicitly designed to lower AMDs manufacturing costs, play to the companys strengths, and help it achieve competitive parity with Intel over the long term. Doing so necessitated some short-term tradeoffs, but theyve been made in an intelligent manner. The result is an x86 processor thats different from anything weve seen from either AMD or Intel before.
Really? LOL_WUT indeed. I am saying that 4C w/HT from Intel is not enough after three or four years of that being a mainstream (not and Extreme Edition) part. WTF does AMD have to do with anything?
We aren't starting another whacked out digression with funky math you throw out there here. Your baselines are in your own freaking head. No, don't post anything here to attempt to back them up, I am not interested in discussing them with you, save it.
Tocks are never as exciting as ticks I think most would agree. A die shrink, some new features, better thermals and overclocking. Pretty cool but not that much of a thrill. But a tick brings pages and pages of new architectural improvements with lots of cool diagrams and hours of reading to try and figure it all out (for me anyway) before you see the actual IPC improvements in various applications in the benchmarks. We're always looking for the next release to be as dramatic a performance improvement as we witnessed in going from P4 to Conroe. That was nearly a doubling of performance per clock. It spoiled us.
All that said I'm grabbing one of these as soon as I can.
Says who?
And the math is easily available, and AMD comes into the picture because you brought it up and insinuated that AMD's approach would be better in the long run. Sandy Bridge has 50% higher IPC than Bulldozer, and when you use resources on both cores on an AMD module, per-core performance goes down another 10-20%.
You don't need more cores when with a small architectural revision ("tick", i.e. Sandy Bridge to Ivy Bridge) enables you to get 5% higher IPC and a major architectural revision ("tock", or Westmere to SB) gives you 10-15% higher IPC on top of that. In the end, you're getting 15-20% more performance at the same clocks every two years, which is nothing to be whining about.
Simply adding more cores while not improving the performance of each one makes for lazy, craptastic engineering, which is what AMD achieved with Bulldozer.
Says me? Intel has to get me to part with my $$$? It's the only vote I even have in the matter.
It's a good thing SB has that IPC lead, otherwise they would look silly. If it was the other way around, and Intel was pumping out 5Ghz Bulldozers @ 95W on a very mature process vs 2.7 ghz leaky Sandy Bridges from AMD the shoe would be on the completely other foot.
I should be happy with 15 to 20% every two years? Why? That is pathetic. Thuban gave me 50% more power over Deneb for the same Watts in ~2 years. Intel did the same with Core 2 Quad. Given the increasing density of the chips, that should the norm. According to me.
Chasing IPC is a fools errand at some point, you end up with what you were telling me I was supposed to be happy with, that's why we don't have uber-wide single core processors today. Scalling out is the other obvious choice to increasing performance.
To each their own, man, but I'll keep expecting more at this point. Those transistors that should be cores are being blown on IGP Until they offer compute capabilities. Then their wastage becomes more open to interpretation, IMHO.