Slashdot Mirror
Phenom IIs, Core I7-920 Win Out In Value Analysis
An anonymous reader writes "We've all seen processor benchmarks, but how do today's enthusiast CPUs look when you account for performance per dollar? Using a smorgasbord of charts, scatter plots, and performance tests, The Tech Report attempted to single out the highest-value offerings out of 16 popular Intel and AMD processors. The results might surprise you: AMD's 45nm Phenom IIs (both triple- and quad-core) prove to be strikingly competitive with Intel's Core 2 Quads. And, on the high end, Intel's $266 Core i7-920 turns out to be a compelling step up despite the higher costs of Core i7 platforms in general."
Really, who doesn't know that AMD is higher performance per dollar.
http://CryoLANparty.com/ A lan I'm staff on!
which one is more more secure?
TFA says that the Core i7-920 is $284; the chip below it (The Core 2 Quad Q9550) is $266. It's still up there on the performance/price scale, though.
Its about your investment.. For me Phenom II was a no brainer because of AM2+ compatibility. Once newegg put those suckers at 200 bucks i jumped. Its like i have an entire new PC and that was upgrading from the 9600 quad core.
Oddly enough i didn't have complaints about the performance of the 9600.. i just figured encoding times and processing times would be reduced enough that it would accelerate my work and well, for 200 bucks its done so and more so than i expected.
i7 is a nice platform but i'm penny pinching right now and looking for better ROI vs bragging rights.
It is wrong to compare performance/price because this assumes price scales linearly with performance, which is clearly false. Nobody expects to get 50% more performance when they pay 50% more. But if there is a $100 process having a performance of 1000, then we would normally consider it an excellent deal if we could pay $150 for a performance of 1300. The value for your money therefore scales in a non-linear way, and it's better to just have everyone look at the scatter plot and choose their own price point based on their personal internal scaling function. The core i7 has the greatest discontinuity in jumping ahead of the rest of the crowd in this regard.
Until electricity is free, comparing CPUs based on up-front cost of the CPUs alone ignores a major part of the expense of owning and operating computers, particularly if you're running servers.
But that's okay, Slashdot. I understand that you live in your parents' basement and you don't pay for electricity anyway.
What stood out to me is that AMD seems to have a fairly consistent price:performance ratio. Is this policy?
Most of their offerings fall pretty close to a line (not quite a zero crossing, but close). If this holds true for all their current and future offerings, you don't have to have test metrics for every processor. You can use price as a reasonable estimate of performance. i.e. Double the price gets you twice the performance.
Intel on the other hand, you can't trust price to indicate performance. A lot more research is involved. OR else you have to assume there's a high likelihood that the AMD offering for the same price will be better.
The world is made by those who show up for the job.
Buying a faster chip is a lot cheaper and faster than rewriting something to be multithreaded.
I recently had to make the tough choice of a Phenom 2 vs Intel Core Quad. I went with the Intel because I somehow came to the conclusion that they run cooler.
You see, I'm building a recording PC, so I want to have as few fans as possible. I plan on having a huge heatsink with NO fan. Most reviews, if they focus on heat, focus on the overclocking aspect.
If wattage correlates to heat like I think it does, I may have been better off with a Phenom 2. But, then again, the wattage test was only run during one task in this review. I read another review where it was different.
There just aren't enough review sites out there for... ahem... "grown ups". Maybe I should start one that takes a look at performance with DAWs like REAPER.
In the end, I don't care about best performance per dollar, or wattage per dollar. I care about performance per degree of heat, because heat = noise. Performance of modern CPUs is good enough these days.
Oh well, that's my rant of the day.
Buckle your ROFL belt, we're in for some LOLs.
0/0 gives me a headache. Have to keep carrying the zero.
The world is made by those who show up for the job.
They're just using that chart as an example to illustrate their ranking system. You see the little next page button at the bottom of the page?
But buying a faster multi-core (as in, 3 or more cores) chip isn't going to do you any good if your application only runs on one or two threads.
The article is missing the best CPU value for the money, in my opinion. The AMD Kuma 7750 AM2+ processor. It's dual core, but at around $60 shipped (Newegg) nothing else touches it from a performance to dollar perspective. They should have included the 7750 in the comparison rather than the Athlon X2 6400+ (the 7750 is K10 architecture vs. K8 for the 6400+, has 2MB level 3 cache, is not discontinued, etc.)
But buying a faster multi-core (as in, 3 or more cores) chip isn't going to do you any good if your application only runs on one or two threads.
Very true if your system only runs that single application. However, everyone I know runs multiple applications just by booting their OS.
There is no "I disagree" mod for a reason. Flamebait, Troll, and Overrated are not substitutes.
I just redid my system line by putting an e8400 in my desktop, where I mostly game. I switched out a Phenom 9600 (cause the tlb erratum for vista x64) to my file server/media center (which runs Ubuntu). and really as this graph will tell you: fast dual-cores are going to blow away slow quads in gaming because most games are not programmed for multiple threads and take advantage of a higher clock core. However, for most other tasks I do, like compiling the Linux Kernel (I run gentoo side by side with vista), the Quad Core Phenom 9600 seems to be much faster. Plus, I had a hard time overclocking the 9600 to anything past 2.6 ghz whereas the 3.0 ghz stock e8400 easily clocks up to ~4.0ghz on air. I should also note that I picked the e8400 over the q8200 because of the virtualization tech as I do alot of virtual systems for testing.
Live Free
Which is even more the case if you virtualise an operating system.
But but but but ....
Any number divided by itself = 1
THEREFORE 0/0 = 1
See, maths is easy!
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
multithreading is not that difficult, and there are ways to multithread an app without rewriting. Granted it's hackish, but then so is almost all of software.
The Kruger Dunning explains most post on
However, everyone I know runs multiple applications just by booting their OS.
But not everybody knows what their system is doing. I do: I always keep a CPU load chart in my KDE taskbar, and for interactive usage I probably use one core less than 3% of the time, and both cores about 0.1% of the time.
If I am transcoding a video, then one core gets pegged. However, I would never notice load on the processor on even with a single-core CPU if I just ran the transcode under "nice". It might take 3% longer to run because it waits for the interactive stuff, but that's insignificant.
But I don't run transcoding on my workstation anyway. Why? Because all the I/O continuosly flushes out my disk buffers for other processes. That makes my interactive apps seem slower than crap anyway because they have to hit the disk every time some of their data or program image gets flushed out. My dual cores do nothing to address that issue. I run transcode jobs on a server box where they won't bug anybody.
The only place dual core would really help most people with typical single-threaded apps is if they run at least two copies of programs heavy on number crunching but light on I/O and memory bandwidth, like $FavoriteCause@Home. Other than that, people will have to wait for multithreaded user apps to get much real-world benefit out of multicore CPUs.
But but but but ....
Zero divided by any number = 0
THEREFORE 0/0 = 0
See, maths is contradictory!
( Those mods that didn't #DE or #Z already will now #MF )
I have discovered a truly marvelous proof of killer sig, which this margin is too narrow to contain.
There is another barrier that we will eventually hit. The current process schedulers in use on modern operating systems have a problem. Attempting to use more than roughly 38 logical processors will result in the additional processors either waiting to run the process scheduler, waiting for a memory access, or waiting for I/O. Currently this is sidestepped on mainframes using virtualization and low latency I/O. I have a hard time seeing how virtualization or lower latency I/O could be adapted for use in desktop computers. Running Windows Aleph-Null, MacOS X 12.8, and Linux 2.8.1853 would probably not have much appeal to average users that would still need the power a desktop computer offered at that point. Intel, the main driving force of the PCISIG keeps pushing newer versions of PCI-Express that do not address the interconnect's inability to play nice with multiple masters, providing any type of packet routing, or deal in any way with its absurdly high latency, which is over 100ns even with PCIe 1.1, and gets worse with each newer version. Some sort of low latency sideband channel would work, but figuring how to maintain backward compatibility with current PCIe cards and motherboards is not easy. Instead Intel has added only DRM features, but no actual security for the computer's user on the bus itself (think Firewire and writing to whatever memory locations you want).
Bad form ahead:
If embryonic stem cell research does not make you uncomfortable, you have not thought about it enough. --James Thompson
Better:
If embryonic stem cell research makes you uncomfortable, you have thought about it far too much. Try researching the actual potential feasibility of the scare stories, and consider that adult stem cells have never been made totipotent, only pluripotent.
Yeah sorry about that.
Impersonating Tycho from Penny Arcade since before there was a PA.
Slashdot's seemingly-ridiculous problems with non-ASCII characters are simply a safeguard against displaying the nullity character, which would cause the universe to implode.
But I don't run transcoding on my workstation anyway. Why? Because all the I/O continuosly flushes out my disk buffers for other processes.
Which is exactly the reason why posix_fadvise(2) exists.
Attempting to use more than roughly 38 logical processors will result in the additional processors either waiting to run the process scheduler, waiting for a memory access, or waiting for I/O.
[Citation needed]
SGI sold systems with 128, 256, 512 and even 1024 processors running a single Linux image.
SJW n. One who posts facts.