Slashdot Mirror
4 Cores? 6 Cores? Do You Care?
An anonymous reader writes "Intel has updated its processor price list earlier today. Common sense suggests that Intel may not care that much anymore whether its customers know what they are actually buying. One new six-core processor slides in between six-core and quad-core processors – and its sequence number offers no clues about cores, clock speed, and manufacturing process. If we remember the gigahertz race just a decade ago, it is truly stunning to see how the CPU landscape has changed. Today, processors carry sequence numbers that are largely meaningless."
Some combination that measures both how many operations per second, and how much power it's going to take to do said operations (i.e. Watts/computing unit). I don't know if even FLOPS is sufficient anymore to describe current computing tasks. Heck, I'd be happy with any sort of standardization.
The sending of this message pretty much inconveniences everyone involved.
What benefit is there in confusing your customers as to which product they should purchase? When I, as a consumer, feel overwhelmed or confused about a product choice, I usually respond by simply purchasing nothing at all. And I'm sure I'm not alone in that.
If the chip can't run all the cores at full speed due to heat/power considerations and therefore either throttles back each core's speed or disables some cores under heavy load, than core counts are really just a deceptive pissing contest, aren't they?
I've abandoned my search for truth; now I'm just looking for some useful delusions.
Have you considered that the reason the processor numbers tell you nothing is that ALL the chips are fabbed with 6 cores and the ones that have one or two bad cores in testing have 2 cores disabled and are sold as quads?
I've abandoned my search for truth; now I'm just looking for some useful delusions.
The problem now is that you have to do a tremendous amount of research before you buy now. It used to be much simpler: Pentium 60, 66, 75 or 100, pick one. Later it was still simple with Celeron or P2/P3/P4, as you are picking bigger cache and faster bus speed. Now to get the highest return on partially defective silicon, they offer too many models, many that overlap each other, and can be very confusing, with some dual core models that outperform quad core, etc. A year ago I finally settled on a Q9550 but it took reading 50 articles to figure out that it was, at the time, the best bang for the upper middle buck. So yes, the average consumer will get boned.
Tequila: It's not just for breakfast anymore!
The program? I dunno about you, but I run plenty of programs at once. And having 4 cores means that I have a few on standby whenever I feel like doing input, even when the machine is busy processing stuff.
The real issue I see is memory access. Even with a single core did we run into memory bandwidth/latency bottleneck; with 4-6 cores those are 4-6 times as much. In the long run we have to give up Neumann architechture; it simply can't scale to our needs. A NUMA might be an acceptable compromise, but in the long run we need to change to a dataflow architechture, and that also means a step beyond C/C++ and other Algol-descended languages which have dominated our thinking these past decides.
We need to switch to a system with lots of cores, all with their own local memory, and able to send each other messages. As an added bonus, such a system is also a natural fit for artificial intelligence.
Forget magic. Any technology distinguishable from divine power is insufficiently advanced.
Actually, I (sometimes) use my quad-core to run a virtual machine on two cores, and the native OS on the other two cores. That means that both OSes can potentially run one crappy application and neither becomes unresponsive.
Any fewer than four cores, and it's iffy, for exactly that reason.
I would hope that the scheduling would get better, actually, but even if the engine is only optimized to take advantage of four cores, it would probably run better if it could actually have all four cores to itself, with the OS and everything else running on core five.
I suppose it depends how much overhead there is.
Don't thank God, thank a doctor!
Its doubtful AMD will ever regain the performance lead again, Intel was lazy, lost one round and learned they had to bust their ass cause AMD was going to push them.
From here on out, barring complacency by AMD, the best you can expect is that AMD will be close to Intel in performance for most things, better at a select few, and almost invariably cheaper resulting in more performance for a given cost, but not being capable of producing the fastest raw speed or the lowest power draw. Intel will win around the board at the raw numbers and will continue to only occasionally have AMD do some things better.
I hope the two of them continue doing exactly what they are doing for at least 10 more years. They are a duo-oply(? spellcheck failure!@$!@$!$), but one that competes and so far appears to be providing benefits to consumers rather than price fixing with AMD and ripping us off while they sit on their laurels.
Persistent Volume manager for Kubernetes - https://github.com/dwimsey/openshift-pvmanager
> My current system is a C2D 1.8GHz E6300 that's now pushing 4 years of age,
> yet according to all the benchmarks I've seen by Anntech, Tom's Hardware and others, my performance results are less then 20 percent below the latest/greatest CPU's.
While you probably don't need to upgrade your CPU, I don't see how your CPU can be only 20% slower than the latest and greatest. Even for single-threaded stuff.
See: http://www.anandtech.com/bench/Product/61?vs=142
Note: I'm even comparing the 2.33GHz C2D to the latest and greatest, since the 1.8GHz one isn't listed. But I'm sure the 2.33GHz C2D should be a bit faster than your 1.8GHz C2D.
For graphically intensive games, though the difference in the average fps would not be as high, the difference in the minimum fps might be, and that might be more important in many real-world scenarios.
In many ways it's quite impressive what Intel has done with the x86. The equivalent of a hypersonic flying pig beating the less "ugly" MIPS and Alphas ;).
Assuming nothing breaks, my next upgrade is more likely to be an SSD than CPU, GPU, RAM or HDD. I'm just waiting for the prices to go down to more reasonable levels (and the number of bug reports to dwindle as well ;) ).
Everything I see shows that modern OSes not only don't have an overhead with more cores, it helps things. Reason is what OSes really have is a heavy context switching overhead. If a processor is doing something, and the OS needs it to do something else, it has to generate an interrupt, push everything on to the stack, switch to the kernel, switch to the net process, etc. It is a hefty overhead. However that all goes away if instead multiple things run at the same time on hardware. They don't switch contexts, they just keep running.
This is the reason why web/DB heavy servers like to have lots of cores, even if less powerful. Sun's new chips are designed with that in mind. Each core can handle 8 threads in hardware, meaning it acts like a 64-core CPU though only having 8 actual cores. Why? Context switching. The tasks it normally deals with are not high load, but they switch around a lot. The more than can run side-by-side from the OSes perspective, the less overhead and the more efficient use of processor resources.
In a desktop the tasks are more intense so it is less useful to have lots of threads/CPU (currently 2 is the highest in the Core i3/5/7 series) but more cores are still quite useful. It allows for more things to happen at the same time, from an OS perspective, and lowers overhead.
You notice too, using a multi-core, multi-threaded system. Things are damn responsive.