Slashdot Mirror
AMD Launches Fastest Phenom Yet, Phenom II X4 980
MojoKid writes "Although much of the buzz lately has revolved around AMD's upcoming Llano and Bulldozer-based APUs, AMD isn't done pushing the envelope with their existing processor designs. Over the last few months AMD has continued to ramp up frequencies on their current bread-and-butter Phenom II processor line-up to the point where they're now flirting with the 4GHz mark. The Phenom II X4 980 Black Edition marks the release of AMD's highest clocked processor yet. The new quad-core Phenom II X4 980 Black Edition's default clock on all four of its cores is 3.7GHz. Like previous Deneb-based Phenom II processors, the X4 980 BE sports a total of 512K of L1 cache with 2MB of L2 cache, and 6MB of shared L3 cache. Performance-wise, for under $200, the processor holds up pretty well versus others in its class and it's an easy upgrade for AM2+ and AM3 socket systems."
I'll be waiting for the dust to clear with Bulldozer before I make a commitment for my next build. No reason to buy a $200 Phenom II X4 980 now when there is no application that needs that much power. If you buy a Sandy Bridge or a higher-end AM3 board/processor now, your average gamer or office worker won't be able to max it out for years -- unless he does video editing or extensive photo shop or if he has to get his DVD rips down to a 10 minute rip vs a 15 minute rip per feature film...
Might as well wait for the dust to clear or for prices to fall.
6 core is slower per core than 4 core simply because of thermal envelope.
6 core is superior if you need to use more than 4 cores at same time.
This [arstechnica.com] thread has some interesting information on possible BD performance.
.....
This is 301 posts with back and forth that looks basically to be speculation. Prove me wrong by quoting specific statements of those that have benched the [unreleased] bulldozer. Because otherwise, this link is basically a bunch of AMD fanboys fighting against Intel fanboys. But prove me wrong...
Once the GPU is maxed-out, there's nothing more for the CPU to do. If you're running at 30 FPS at high-res, the CPU might be at 30%. At that point, any number of different CPUs will have identical benchmark results. When you drop the load off the GPU, the CPU hits 100% usage and you can compare 150 fps to 160 fps, for example. This is a very simple and typical way to benchmark CPUs for gaming perf. Reviews and reviewers (such as myself) have been doing this for 10+ years, since the very first 3D accelerators came to the gaming market.
So clock speed means everything when comparing different CPUs and not their raw performance. Got it.
Furthermore, there is no 10 year old CPU that runs at 3ghz unless you did some absurd overclocking.
The Phenom II x 6 core chips already run at 3.7 GHz when 3 or fewer cores are in use (that's what the automatic turbo feature does), so the 980's ability to run 4 cores at 3.7 GHz is only a minor improvement since it basically has no turbo mode. The x6 will win for any concurrency workloads that exercise all six cpus. Intel cpus also sport a turbo mode that works similarly.
The biggest issue w/ AMD is memory bandwidth. For some reason AMD has fallen way behind Intel in that regard. This is essentially the only reason why Intel tends to win on benchmarks.
However, you still pay a big premium for Intel, particularly Sandy-Bridge chipsets, and you pay a premium for SATA-III, whereas most AMD mobos these days already give you SATA-III @ 6GBits/sec for free. Intel knows they have the edge and they are making people pay through the nose for it.
Personally speaking the AMD Phenom II x 6 is still my favorite cpu for the price/performance and wattage consumed.
-Matt
Moores observation was about transistor count, not mHz, corecount, speed, wattage, flops, bogomips, or anything else.
I won't talk about Intel's system, but AMD is actually relatively straightforward:
First comes the family name. For desktops, this is usually either "Athlon II" or "Phenom II". The only real difference between them is the amount of cache.
Then comes the core count - X2, X3, X4 or X6. Completely self-explanatory.
This is followed by a number that essentially stands in for the clock speed. Higher-clocked processors have higher numbers, lower-clocked processors have lower numbers.
Finally, certain processors have "Black Edition" appended, which simply means that the multiplier is unlocked, greatly easing overclocking.
Not exactly, but close for single-core performance. The "MHz Myth" is largely a myth itself. As this table shows, per-MHz single-core performance between the infamously bad (even at the time) P4 and the current best (Core i7) has only improved by a factor of less than 2.6, since October 2004! (When the Pentium 3.6 EE was released).
Perhaps more importantly, the ratio between the most productive (per-mhz) chip from 2004 (Athlon64 2.6) and the most productive on the chart now is a mere 1.6! That's a 60% improvement in almost 7 years!
That is a joke. For reference, we went from the Pentium 100 (March 1994) to the Pentium 200 (June 1996) - approximately a 100% improvement in a little over 2 years.
So, no, improvements in instructions per cycle are not even close to keeping pace with what improvements in MHz used to give us. (And if you looked at instructions per cycle per transistor, it would be abysmal - which is another way of saying Moore's law is hardly helping single-threaded performance any more).
Prime95, in this context, is for convincing 0v3rcl0ckz0r kiddiez that their massive overclock is stable even though it's a terrible stability test. A prime number search program is not exactly the world's best method of achieving full test coverage of a CPU, no matter what a billion leetboy forums may tell you.
Eh... Prime95 is a darned sight better then a simple memory test, because it actually *does* stress the CPU and L1/L2 cache as well as the RAM. Plus it keeps track of whether the calculations are correct.
Which is the exact same tactic that you'd better take if you're going to "do scientific calculations which have to be right". You run the calculation and either you have built-in checks or you do the calculation twice, on two different machines and compare the results. (Surprise surprise, guess how Mersenne.org checks that the turned-in results are correct?)
I've been using Prime95 ever since it came out. I've personally seen it find RAM that is slightly dodgy on timing where other tools like MemTest86 gave the RAM a free pass. In one case, the RAM was GEIL and was mislabled as a faster CL value then it actually could handle (naughty GEIL, or might have been counterfeit). Let Prime95 run for 24-48 hours with no errors, and you've got a pretty good assurance that there are no issues with timings or the memory / CPU. (Doesn't do jack to test the disk / video, but there are other tools for that.)
Now, you complain that it's not a comprehensive tool. Have you *ever* seen a case where a CPU was bad / dodgy where Prime95 did not throw an error that you caught in some other manner? That was specifically something wrong with the CPU / cache / RAM?
And frankly, there have always been those who think product X is a magic bullet. Your rant is misplaced.
Wolde you bothe eate your cake, and have your cake?
I've had prime95 catch errors on overclocks that passed -everything- else.
Know what, in every one of those instances, it was right. If I kept running at the speed that passed prime but failed everything else, I'd eventually run into random errors, sudden unrepeatable crashes, or other mysterious problems.
I've never had any issues with any overclock that passed 24 hours of prime, including distributed computing projects where they'll yell at you if you're returning bad data (i.e. aren't passing the redundancy tests).
Don't forget that IPC isn't the be all and end all. If you're stalled due to cache misses, then IPC goes out the Window. Modern CPUs have much more cache and much faster buses to main memory than we had in 2004. That is a large reason as to why they're faster. They also have additional instructions that can do more work per instruction - so comparing IPC from CPUs released today to CPUs released last decade is even more meaningless.
I run: Windows, OS X, Linux, FreeBSD. Just because you have a hammer, doesn't mean everything is a nail.
Alternatively, even though intel has won, he is acknowledging that for 99.9% of people, CPU performance no longer matters as much as it used to.
In general use for example, I see no difference between the core i5 in my work machine, and then Pentium D in my oldest home box.
Gaming? Sure, however even that is becoming more GPU constrained.
Both AMD and intel are on notice. Hence both are putting more focus into GPUs. In terms of CPU, it won't be that long before some random multi-core strongarm variant is more power than any regular user will ever need, and they absolutely kill x86 in terms of performance per watt.
The focus is no longer absolute CPU performance, it is shifting towards price, size, heat and power consumption. Computing is going mobile and finding its way into ever smaller devices. Rather than building one large box, distributed computing is the new wave (the cloud, clustering, etc).
AMD's CPU business might have a tricky path ahead, bu thent so does x86 in general, barring niche markets. If AMD are doomed, then intel's traditional dominance using x86 won't be too far behind them.
I run: Windows, OS X, Linux, FreeBSD. Just because you have a hammer, doesn't mean everything is a nail.
And AMD supports ECC memory in "non-server" cpus.