Slashdot Mirror
How Today's Low-Power X86 & ARM CPUs Compare To Intel's Old NetBurst CPUs
An anonymous reader writes: In trying to offer a unique look at how Intel x86 CPU performance has evolved since their start, Phoronix celebrated their 11th birthday by comparing modern CPUs to old Socket 478 CPUs with the NetBurst Celeron and Pentium 4C on an Intel 875P+ICH5R motherboard. These old NetBurst processors were compared to modern Core and Atom processors from Haswell, Broadwell, Bay Trail and other generations. There were also some AMD CPUs and the NVIDIA Tegra K1 ARM processor. Surprisingly, in a few Linux tests the NetBurst CPUs performed better than AMD E-Series APUs and an Atom Bay Trail. However, for most workloads, the 45+ other CPUs tested ended up being multiple times faster; for the systems where the power consumption was monitored, the power efficiency was obviously multiple times better.
Breaking important news! New CPU's are faster and more efficient than old CPU's! News at... wait... this is news!?
"Be particularly skeptical when presented with evidence confirming what you already believe." -
I would have liked to have seen more AMD processors in the comparison.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
NetBurst CPUs were crappy in their day too. It would be more honest to compare against the Athlon 64 or the Pentium 3.
Every test was either multithreaded or otherwise accelerated (x264 for example). The Pentium 4 chip was one of the slower CPUs from the 130nm era (SL6WT), and Pentium 4 went into the 65nm era. 2.8 GHz versus 3.8GHz is a big handicap. 1 core versus 2+ cores is a big handicap. No built-in video card for x264 encoding is a big handicap.
Would it break the bank to provide even one non-accelerated single threaded comparison? Just one?
I think his article would have been a lot better if he'd put the YEAR OF MANUFACTURE on it. He might recall the details of long gone chips but I don't.
Core i7 3517UE, vs Core i7 3960X + HD 6770, one is 3-4 times faster.... I assume the newer one is the faster one???
I think he's also confused 'low power' with 'small'. NUC and Intel Compute Stick are hardly low power devices, neither runs off batteries and Intel Compute Stick might be sort of small stick shaped, but its got a fan it runs so hot. LowER power than older Intel chips but hardly 'low' power!
Still, the core message it true. I use to develop software on Dells with dual Xeons and 8Gb ram now I develop on a Nettop with an embedded Core i3 and 4Gb.... its good enough to run Eclipse and I don't miss the great big tower whirring into my ear.
Those were awful processors, they were hot, slow and were just produced when intel noticed their design didnt scale at all well at high frequencies.... Its a shame they still worked really, they should have been recycled a long time ago.
I've one of those rare Asus 865+ICH5R that supported Pentium D with just a BIOS update, which I of course did. So I've one (two actually) of the last Pentium Ds able to run on DDR instead of DDR2.
I still *run* a P4 3.8GHz as my main Linux box. Sure it heats up the apartment like a space heater, but it works. Until it stops working, I won't be spending $1200 on a new box... :P
I do not fail; I succeed at finding out what does not work.
I still *run* a P4 3.8GHz as my main Linux box. Sure it heats up the apartment like a space heater, but it works. Until it stops working, I won't be spending $1200 on a new box... :P
$1200 is a *lot* to spend on a new box.
You can get a FX9590 (is that the part? the silly/awesome 220W one) with the stock water cooler, a decent mobo, 16G ECC 1866 RAM, a 500W PSU, a decent enough case, an unimpressive NVidia graphics card and some extra 120mm fans for about 450 GBP or so. Maybe $700.
I know I just bought a bunch (if you look at my story submissions, I asked about a cluster a while back. We got some new money and this is an update).
They are very fast.
Of course, that's a bunch to blow on a new machine, but you can drop the specs a bit and spend rather less.
That said, I fully understand. My home PC is still my trusty eee 900, which is substantially slower than your P4. Recent repair to busted keyboard involved using the dishwasher and now it works perfectly again :)
I also kept on using a 1996 vintage P133 until about the end of 2003.
SJW n. One who posts facts.
Since the 2000's, every box I've built has been specced to about $1200 in parts. One of the reasons this one in particular is going to be so pricey is going for the fastest RAM the mobo will support, the fastest CPU I can afford (in terms of single-threaded performance), and an SSD drive because such a fast CPU would be IO bound on a physical hard drive when running my pet project.
Sure I could go with a lower end CPU and stuff, but the odds are that by the time I'm actually ready to buy, another CPU or two will have come on the market and prices for the CPU should drop to about half what they are now. I expect SSD prices to come down as well, saving a few more bucks. So I'm expecting the actual purchase price in the January-February time frame to be about $900 (Canadian.)
I neglected to mention that this old beast *is* starting to fail, and ever more frequently. RAM tests out clean -- I think the CPU is just finally starting to fail randomly. Recent patches to Ubuntu LTS 14.04.2 have taken care of the display driver/KDE crashes, but it still will periodically just "lock up" completely without producing a white-out screen. Fortunately the lock-ups are only about once or twice a month while under heavy load for an hour or so.
Still, it's far more reliable than Win2K or WinXP ever were -- those boxen had to be hard-booted at least once every day or two. Things *have* improved over the years.
Now if Oracle would only stabilize JDK 8. Update 45 crashes like a fiend on large ant builds, and even the runtime fails so often that I have to do 5-10 runs of one particularly large job before it will run to completion.
Fortunately, time is one thing I have in plenty on disability. Were I using this box for work, I'd have the coin and justification for replacing it *now*. In the meantime, I save as much as I can as often as I can towards that future purchase. :D
I do not fail; I succeed at finding out what does not work.
Motherboard and PSU capacitors going bad.... possibly the CPU going bad but the caps are more suspect.
This is bad value thinking, though.
The fastest CPUs today generally aren't more than 30% faster than a baseline (let's say a $150 Core i5). If all you care about is single threaded performance, then a Pentium G 3258 (overclockable) will get you anywhere you want to go for $70.
That same Pentium G will be 5-10x faster than your 3.8 GHz Pentium 4 with probably (guessing) 1/3 the power usage.
So you can spend $300-500 on a CPU that is 30% faster and hang onto it forever or you can spend $70-150 on a CPU and replace it a little more often (when you'll feel that 30-50% performance difference) and you've pocketed $150-350 straight up. This is even more true for RAM which doesn't have a large performance benefit (maybe 10% at most) but the price may be 50%-100% more.
Buying technology you don't need because you're thinking about the future is never the best value. Years before you replace a system, I'll be on something faster and my total spend will be less (not even counting the fact that high end products -tend- (Macs excepted) to lose their value much more than more mid-range purchases (percentagewise).
I spend $500-700 on each system and replace every 3-5 years. My newest system is an Ivy Bridge i5 I got off ebay for $130 (CPU+mobo) when Haswell came out. Haswell is about 5-10% faster than Ivy Bridge straight up, but an equivalent Haswell CPU+mobo at that time would have cost about $300-400.
Other option is thermal paste gone bad. Might be worth removing the CPU and heatsink, scraping off the paste (it will be hard now) and reapplying.
SJW n. One who posts facts.
What's wrong with Phoronix?
They seem pretty decent compared to a lot of the review sites. Thi benchmark suite is opena and the results seem decently reproducable. And also, they produce benchmarks on Linux which is what I care about.
SJW n. One who posts facts.
I just replaced the CPU cooler last year, though, so I doubt the paste has gone bad.
Capacitors -- unlikely. They're the square block type mil-spec capacitors on this mobo, not the tube-type with electrolytic fluid.
I do not fail; I succeed at finding out what does not work.
I think the more likely issue is a failing PSU or the CPU itself.
I do not fail; I succeed at finding out what does not work.
Pshaw, I still have a 350MHz Pentium II desktop, and a 133 MHz Pentium MMX (Oooo!) laptop with the maximum 96M RAM it can have. Okay, I don't run them as my main boxes, but I do still use them occasionally. The laptop needs 30 seconds to bring up Firefox 3.5, Stellarium is unusably slow, taking 5 minutes to come up. You might think a 133 MHz processor should be able to do better than that, but actually MHz is much less important than capabilities and, at that level, RAM. Below 256M, every megabyte counts. For instance, the Pentium II has a Riva TNT, the very oldest Nvidia graphics card that the Nouveau driver supports, and it smokes a 1GHz Pentium III with Intel integrated graphics (845G if I remember right).
Wait... I still have a working Apple ][+ and Commodore 64 in the closet! At least, they worked the last time I booted them up. Which might have been a decade ago?
Intellectual Property is a monopolistic, selfish, and defective concept. It is "tyranny over the mind of man"
The same thing thats wrong with all other "reviews" - they put together an arbitrary test system in each case, which all by itself leaves tons of room for gaming the results.
Even when these "review" sites are trying hard to be honest, they often drop a $50 cpu into a box with a high end motherboard, high end ram, very high end video card, etc... such a "review" is complete garbage.
Passmark has it right. Test real world end user systems. Do these tests on hundreds or even thousands of systems and report the averages. The fact that most of the passmark benchmarks are performed on systems that were absolutely not in any way built specifically to be benchmarked is the key. The systems had other typical motivations with regard to their component selections. Often times in these discussions someone will complain that some of those passmark samples were on completely borked/mis-configured/virus-ridden systems, not wanting to admit that such things are washed out by the average.
The "downside" to passmark is that the benchmark is completely synthetic, but thats only a minor downside when you consider that the non-synthetic benchmarks are still benchmarking software that few people are running anyways. For instance, in this case things like "timed apache compilation" that phoronix uses is just as much a synthetic proxy as passmark if you do things other than compile apache all day long every day. Its just as much a proxy as a synthetic test.
Now maybe the phoronix guys are trying hard to be fair and honest, but that doesnt end the discussion because honesty just isnt the whole picture.
"His name was James Damore."
What about the other caps? Those of the power supply, those of ancillary equipment (video card, sound card, whatever else you have plugged in)?
A bad cap anywhere can cause enough of a glitch in the works that something unexpected happens, and things freeze up.
Kid-proof tablet..
Agreed. His description pretty much cries out that its a heat problem, and thats going to be the cpu, motherboard (one of the i/o chips), or ram.
Its not the PSU because lockups would be growing in frequency at a quite noticeable rate. if his issue is still only once or twice a month then that cannot be the case.
"His name was James Damore."
Here's one specific case:
http://www.phoronix.com/scan.php?page=news_item&px=LGPL-GPGPU-NyuziProcessor
Multiple people have contacted Michael Larabel and told him that it is in fact Jeff Bush, not Theo Markettos, who developed Nyuzi. I emailed Larabel directly. Multiple people pointed the error out on the forums. People mentioned it in response to Larabel's tweet. Larabel has been contacted in enough different ways that he simply could not have missed this, so he's intentionally refusing to correct the error. BTW, this is not his only error in the linked article.
Just to be clear, this is not an isolated incident, otherwise it wouldn't bother me so much. Innocent mistakes happen. But I've caught him on numerous occasions posting factual errors on his site. It is clear that he has zero interest in correcting his errors.
Here's another one:
http://www.phoronix.com/scan.php?page=news_item&px=MTEwNTQ
http://www.phoronix.com/scan.php?page=news_item&px=MTEwODY
And some others on this topic.
The OGP is more or less dead, although Nyuzi has the potential to bring is new life. However, Larabel got his information about the demise of the OGP from this weasel named Michael Meeuwisse who for whatever reason got a bug up his ass to undermine the OGP. I think he wanted to take attention away from us to put it onto his ProjectVGA. Now, disagreements happen. But Larabel is a sensationalist, not a journalist, so it was more in his favor to get his "facts" from some malcontent. Instead of me, the guy who founded the OGP who would therefore be able to present a different perspective and correct some of the factual errors. But that wasn't interesting to Larabel.
Does any of this matter in the long run? Not really. Nobody who matters really believed any of what Meeuwisse was saying, and it hasn't had any impact on what's going on with Nyuzi or any other work in this area. Meeuwisse and his ProjectVGA have disappeared into obscurity even moreso than the OGP.
The fact still remains that Larabel has zero interest in correcting errors or presenting rebuttals or any kind of balanced viewpoint.
Are you satisfied now?
The fastest CPUs today generally aren't more than 30% faster than a baseline (let's say a $150 Core i5). If all you care about is single threaded performance, then a Pentium G 3258 (overclockable) will get you anywhere you want to go for $70.
This.
The way to be thinking about this is that you will be spending $X on a computer over a period of time T. Figure out how much that is per year, and then formulate a buying strategy that keeps you ahead of the curve for the longest proportion of time.
In his case I suspect that he is probably looking at an average 6 year cycle on his $1200, which is $200 per year. Clearly and obviously he would be much better off spending $400 every 2 years instead of $1200 every 6 with regards to how much computation he could do over those 6 years. Not sure where the sweet spot is, but clearly its far south of $1200.
"His name was James Damore."
Blow the dust off the motherboard.
love is just extroverted narcissism
My router is still a 600MHz Coppermine P3. Runs 24/7, and never gives me a problem. Those are (were) good chips, the early Coppermine CPUs were only 10-15W, much better than the later P4's.
You can get a FX9590 (is that the part? the silly/awesome 220W one) with the stock water cooler, a decent mobo, 16G ECC 1866 RAM, a 500W PSU, a decent enough case, an unimpressive NVidia graphics card and some extra 120mm fans for about 450 GBP or so. Maybe $700.
Yes, but that's a false economy.
You can buy cheap crappy computers like that, but you end up replacing them twice as frequently, and since they cost 70% as much, you spend 140% overall.
I buy dual socket workstations with 1 CPU to begin with. With a dual-socket machine, I can upgrade the CPU to a larger one; then I can install a second CPU; then I can install more RAM; the motherboards have more IO and so I can install more drives and NICs as needed. My current machine has a 6 core CPU, can be upgraded to 2x 18 core CPUs, and 1TB of RAM (64GB currently). The price of the motherboards pales into insignificance compared to the RAM, and in my long experience with computers, RAM is always the first thing to be short supply. This machine cost $3k, but will last me 3 years with it's current configuration, and another 3 years after an upgrade (which will be cheap due to secondhand parts). I expect to spend $4.5k over 6 years. It's also a hell of a lot faster NOW, than the system you describe, and in 6 years will be about the same performance or better (Moore's law tapered off already) as the system I could buy for $700 in 6 years (even following over-optimistic projections).
So while I could have bought 4 systems for $700 each in that same period, I would have only seen 1/8 to 1/4 the performance under the curve. The only thing that can be said, is my workstation uses more electricty than your cheap system, but electricity is cheap, very cheap, so it really doesn't matter how much power my workstation uses.
Also setting up a new workstation costs somewhere between a day and a week depending on how problematic the new hardware is; at $1k/day in lost earnings, the cost of 4 new systems is at least $4k in addition to the purchase price, and could be as high as $20k, especially as these cheap systems are notorious for buggy hardware with half-written drivers.
Nope. High-quality Dual/multi-socket workstations are the way to go, no question about it.
The AMD price advantage is an illusion. You pay the difference in higher power consumption.
Contribute to civilization: ari.aynrand.org/donate