Another Look At Intel's Lynnfield Linux Performance

[phoronix]

Phoronix: Another Look At Intel's Lynnfield Linux Performance

Earlier this month we provided a launch-day preview of the P55 Chipset on Linux along with benchmarks from the Core i5 750 and Core i7 870, which are the new quad-core Lynnfield processors. We noticed some odd performance issues under Linux when testing out these new processors, but Intel has since chimed in and we are in the process of running an updated set of tests.

http://www.phoronix.com/vr.php?view=14203

[0e8h]

Good to see someone doing reviews on these processors on gnu/linux.

Personally I'm a little confused by the i7/i5 series on our platform. Are these cpu's optimised for the kernel? Why si that major distros force people on AMD64 opcodes when both AMD and Intel have unique abilities that surely can make a difference in cpu operations.

Maybe someday I'll have to learn how to build my own kernel, like I did in early days when I tried linux.

[loll]

Hi Phoronix men,

It's my first post and it's a pleasure to visit you almost every day.

I have just a remark about the test suite articles : they miss a summary graphic at the end to recall all tests in a quick view look...

I hope you will provide this one day.

Thanks for your work !
Good luck !

[nanonyme]

It probably doesn't mean what you think it does. AMD64 pretty much means "the 64bit CPU architecture that AMD designed and which Intel's desktop 64bit CPU's are based on too" afaik. It has nothing at all to do with CPU optimizations. It just says that the system is 64bit x86 computer (and not Itanium or Sparc64 or whatever), nothing more.

[smitty3268]

Yep, AMD64 is just AMD's brand name for the x64 architecture, which Intel calls IA64. Both have the exact same intsruction set, just like they both implement the same x86 instruction set to remain compatible with each other. AMD64 is sometimes used instead of x64 simply because they were the ones who designed it and were the first to implement it. IIRC, Intel was thinking of creating their own competing 64bit architecture, but MS had already started working on an AMD64 port of windows and said they wouldn't support a 2nd 64bit architecture.

[nanonyme]

IIRC, Intel was thinking of creating their own competing 64bit architecture, but MS had already started working on an AMD64 port of windows and said they wouldn't support a 2nd 64bit architecture.Not exactly accurate, I think. Intel had 64bit Itanium processors long before AMD thought of doing 64bit. The problem was that Intel's processors couldn't run 32bit code as far as I've heard.

[jpiszcz]

Does anyone who has this board know if you can put a raid controller in the adjacent x8 slot?

[jpiszcz]

In the manual:

• Intel® Core™ i7 and Intel® Core™ i5 processors in an LGA1156 socket:
― 1 x16 PCIe 2.0 Graphics interface (operates in x8 mode when second slot
is populated)
― 1 x8 PCIe 2.0 Graphics interface

So one would presume(?) that the x8 slot cannot be used for anything else but a graphics card?

[lem79]

Yep, AMD64 is just AMD's brand name for the x64 architecture, which Intel calls IA64. Both have the exact same intsruction set, just like they both implement the same x86 instruction set to remain compatible with each other. AMD64 is sometimes used instead of x64 simply because they were the ones who designed it and were the first to implement it. IIRC, Intel was thinking of creating their own competing 64bit architecture, but MS had already started working on an AMD64 port of windows and said they wouldn't support a 2nd 64bit architecture.
No that is wrong. IA64 is Itanium. Intel's AMD64 implementation is called EM64T.

[Kano]

Those vendor provided results look resonable, the others have been really to slow to show the real speed of the cpu.

[dashcloud]

So, I'm confused looking at the charts there- is the i5 actually the successor to the i7 or a cheaper counterpart? If it's the cheaper counterpart, it's doing a fine job, because it's beating the i7 in a number of tests.

[movieman]

So, I'm confused looking at the charts there- is the i5 actually the successor to the i7 or a cheaper counterpart?

Cheaper counterpart: for example, the i7 has hyperthreading, so it can run 8 threads on 4 cores, the i5 doesn't.

In the benchmarks I've seen it's generally behind the i7 870, ahead of the i7 920 on single-threaded tasks and behind the i7 920 on multithreaded.

Note that you can't directly compare the Intel i5 results to the i7 results in this article because they're using a different BIOS. I'll be interested to see how they do compare once Phoronix has the same configuration.

[Kano]

i5 and i7-8xx is for dual channel boards (currently only P55) and i7-9xx is for triple channel boards (X58). i5 has got no HT while i7 has it.

[justapost]

Hmm their apache benchmark results are much lower than mine, can it be that ext4 is slower than ext3 here? Or is it the Intels SSD beeing lower than the OCZ vertex?
Also their stream results are lower, i used DDR3 1333 CL7 dunno what intel used.
Their EIST and Turbo off results are all much lower than mine, as if they ran at sub 2.66GHz.
Beside that turbo on results look valid.
http://global.phoronix-test-suite.com/?k=profile&u=justapost-24176-4674-30777

[lem79]

justapost, what sort of cooling are you using? Curious about this turbo thing with the stock cooler in a closed case. Intel sending out monster coolers in the review kits is kind of questionable too, says to me "hey our stock cooling is crap, you'll have to buy something better if you want results like you see here".

[sabby]

Does anybody else find it extremely strange that a few benchmark shows about double performance improvement from enabling turbo mode which is at most ~30% increase in clock speed (nas, lame, john the ripper md5)?

[Kano]

Thats easy, only with good cooling and 2 cores in sleep mode the max turbo is activated. Apps which are accellerated with turbo mode use 1 core only in most cases. I don't know if there are apps with 2 threads hardcoded.

[sabby]

Thats easy, only with good cooling and 2 cores in sleep mode the max turbo is activated. Apps which are accellerated with turbo mode use 1 core only in most cases. I don't know if there are apps with 2 threads hardcoded.

That's not my understanding of what I have read from the reviews. Turbo mode can be enabled anywhere from 1-4 core active, table being below for the i750 (stock/4/3/2/1)

2.66GHz 2.80GHz 2.80GHz 3.20GHz 3.20GHz

So it should never happen to have benchmark that produce double performance improvement from just having turbo enabled since from stock to the maximum turbo scaling possible is about 30% increase in clock speed.

[lem79]

Kano that's not right for Lynnfield. There's this table on Anandtech (http://anandtech.com/cpuchipsets/showdoc.aspx?i=3634&p=5) detailing Lynnfield's turbo modes.

[Ant P.]

Why does it have to be BIOS-controlled? Give us back the Turbo button or at least bring back the tacky LCD MHz readouts :D

[sabby]

Why does it have to be BIOS-controlled? Give us back the Turbo button or at least bring back the tacky LCD MHz readouts :D

Had to reply since I got such a chuckle thinking about it... Funny that we went back from a nice push button/lcd combo to bios control. They should remember since it was with their cpu, perhaps the engineer weren't around in the push button days;)

[justapost]

justapost, what sort of cooling are you using? Curious about this turbo thing with the stock cooler in a closed case. Intel sending out monster coolers in the review kits is kind of questionable too, says to me "hey our stock cooling is crap, you'll have to buy something better if you want results like you see here".
I use the stock cooler but the board is not yet in a case. I just received an corsair h50 water cooling kit and plan to run those benches whom benefit from turbo again for comparison.

[Pepazdepa]

Why si that major distros force people on AMD64 opcodes when both AMD and Intel have unique abilities that surely can make a difference in cpu operations.

Thats not true. Only Debian (and its clones) use term AMD64 for 64bit x86 environment. Most of distributions use general term "x86-64" insted of "AMD64". No matter what term is used, it means 64bit x86 processor. No vendor optimizations.

[Apopas]

Not exactly accurate, I think. Intel had 64bit Itanium processors long before AMD thought of doing 64bit. The problem was that Intel's processors couldn't run 32bit code as far as I've heard.
Well Itanium 64 shipped during 2001 for enterprises, while AMD introduced Athlon 64 and Opteron during 2003 for both desktops and enterprises. So it wasn't that long before. Also, before Intel came IBM and even before SUN and even before others that made 64 bit processors. So Intel has nothing to note in history books while AMD is the one who brought 64 bit in desktops.

[numasan]

Thats not true. Only Debian (and its clones) use term AMD64 for 64bit x86 environment. Most of distributions use general term "x86-64" insted of "AMD64". No matter what term is used, it means 64bit x86 processor. No vendor optimizations.

Totally useless, but Gentoo also uses the term AMD64. :p

[Apopas]

Totally useless, but Gentoo also uses the term AMD64. :p
Actually, most of the distros use both expressions, gentoo including.

[jntesteves]

64bit and even 128bit microprocessors exist for more than 20 years now, you know. It only wasn't needed in a desktop until recently. All actual x86 processors with 64bit instructions processing capabilities are actually running with 48bit really. No need for full 64bit yet, so it's cheaper to produce those 48bit chips. Itanium came first than AMD64, and Microsoft had actually finished porting Windows to run on top of it. But since AMDs x86_64 architecture is backward compatible with x86 and the old operating systems and all the needed software that wasn't ported to the IA64 yet, it just fall into disuse.

[jntesteves]

About the article. To me, it still looks like the AMD Phenom II is the best bang for the buck. Some of the most prominent differences in those graphs seems to be related to other hardware, like the SSD Intel used for their tests. Looking forward to see Michael's results with the same hardware and the updated BIOS. For now I remain agnostic to these results.

And if you just had a Phenom II X4 in there it would be really nice to have in this comparison.

[nanonyme]

Itanium came first than AMD64, and Microsoft had actually finished porting Windows to run on top of it. But since AMDs x86_64 architecture is backward compatible with x86 and the old operating systems and all the needed software that wasn't ported to the IA64 yet, it just fall into disuse.From what I've read about it sounded like yet another technologically superior design that got hit hard by practicalities. (who the heck wants backwards-compatiblity anyway :o)

[Apopas]

From what I've read about it sounded like yet another technologically superior design that got hit hard by practicalities. (who the heck wants backwards-compatiblity anyway :o)
Well, that's not true. In many cases Opteron was superior to Itanium.
Here are few of them from muhdzamri.blogspot.com :

1. Allows end users to run their existing installed base of 32-bit applications and operating systems at peak performance, while providing a migration path that is 64-bit capable.

2. HyperTransport technology - provides a scalable bandwidth interconnect between processors, I/O subsystems, and other chipsets. This feature is not available in Itanium.

3. Integrated DDR DRAM Memory Controller - this memory is integrated in CPU itself. For Itanium, the memory is outside of CPU. According to Mr Mohan,Itanium introduced FSB (Front Side Bus) to connect CPU to external RAM. This increases latency.

4. Low-Power Processors - the AMD Opteron processor offers industry-leading performance per watt making it an ideal solution for rack-dense 1U servers or blades in datacenter environments as well as cooler, quieter workstation designs. This is a critical factor for HPC environment.

The bottomline is Opteron is more scalable than Itanium in terms of speed. If we add more CPUs, the Opteron speed will increase as opposed to Itanium. To make things worse, the bandwidth between CPUs will be divided evenly.

These are distinctive features on AMD Opteron that made it suitable for SUN to bundle it for their server products. Although Intel is popular, for technical people, popularity is nothing. The technical side of it is more important and pricewise it is cheaper too.
And again, 64bit proccessor even Nintendo64 had. But AMD's design was the big leap, because it extended x86 processors to gain 64 bit capability rather than design a new stricly 64 bit processor, just like dozens of others had made in the past.

[movieman]

But AMD's design was the big leap, because it extended x86 processors to gain 64 bit capability rather than design a new stricly 64 bit processor, just like dozens of others had made in the past.

It's an odd world when adding a bodge on the side of an existing design is considered a 'bigger leap' than building a new CPU from scratch; by that argument MMX was a bigger leap than Cray's vector processing CPUs :).

Incidentally, Intel claimed that the i860 was a 64-bit CPU in the 90s, but I think that only applied to certain components while most remained 32-bit... and it really sucked as a desktop Unix CPU.

I also suspect that the push to make Itanium 'the next big thing' was what made Intel try to avoid going to 64-bit on x86 until AMD left them with no choice; a 64-bit P4 would probably have beaten Itanium just as AMD's 64-bit CPUs often did... if not on raw performance, then at least on price/performance.

[Apopas]

It's an odd world when adding a bodge on the side of an existing design is considered a 'bigger leap' than building a new CPU from scratch; by that argument MMX was a bigger leap than Cray's vector processing CPUs :).
Since 64 bit processors was a reality for the last 20+ years, it wouldn't be something different to make just a new one ;)

[jntesteves]

From what I've read about it sounded like yet another technologically superior design that got hit hard by practicalities. (who the heck wants backwards-compatiblity anyway :o)

If what people want was really superior technology we all would be using PPC64 Cells on our PCs right now, I think (And Apple would never have moved away from it).
(I'm planning on trying to build something like that with some PS3 spare parts some day xD)

But really, after a decade of development and billions of dollars in investments from HP, Intel and their partners, Intel totally failed with Itanium. It doesn't offer any real improvements over what is AMD64.

[Immo]

Does anyone own this i5 lynnfield? Is it safe buy for linux?

[wizard69]

Phoronix: Another Look At Intel's Lynnfield Linux Performance

Earlier this month we provided a launch-day preview of the P55 Chipset on Linux along with benchmarks from the Core i5 750 and Core i7 870, which are the new quad-core Lynnfield processors. We noticed some odd performance issues under Linux when testing out these new processors, but Intel has since chimed in and we are in the process of running an updated set of tests.

http://www.phoronix.com/vr.php?view=14203

This is all well and good but I'd really like to see how the improved BIOS impacts the i7 processors. Also how the kernel and the Linux distro impacts the results across all processors. Of course that means rerunning the tests again and to be perfectly honest I'd rather see more testing on Apples new OS., Snow Leopard.

In fact now that you have a Lynnfield based board how about exercising your computer skills and mocking up a Hackentosh. The only reason I want to see such a platform is to study the performance of Apples new kernel and Grand Central Dispatch feature. I'm very curious about this advance and it would be interesting to see performance figures vs lets say a two core platform. Right at the moment it is my position that the Mac community is underplaying the importance of lots of hardware threads and evidence supporting the value of the new multi core architectures is in order.

Of course even Apple is playing catch up here with software releases that truly leverage GCD but it is an interesting tech that needs the focus of the user community. Especially now that Apple has released the source to libdispatch. If the tech is as good as we hope I wouldn't be surprised to see the Linux community adopt GCD wholesale.


Dave

[wizard69]

justapost, what sort of cooling are you using? Curious about this turbo thing with the stock cooler in a closed case. Intel sending out monster coolers in the review kits is kind of questionable too, says to me "hey our stock cooling is crap, you'll have to buy something better if you want results like you see here".

It is always useful to see what a chip can do running flat out with the best heat-sinking that can be had. But optimal conditions are hardly ever seen in real life, even the hobbiest in his cellar lab can seldom reach optimal conditions.

So while this reporting, from Phoronix, is very interesting it doesn't mean a lot if it doesn't reflect what will happen under average or reasonable conditions. Especially when it comes to multithreading and Turbo Boosted numbers. What we need here is a little honesty, like what happens to each platform in a 80 deg room with a stock cooler on the CPU. It is something the community needs to be aware of because if you only benefit from Turbo Boost or multithreading part of the time then you might as well run an AMD chip. Actually I have to think that Turbo Boost was a sneaky way for Intel to deliver good benchmarks on margianl processors.

Extend these thoughts a bit to the newly introduced Clarksfield processors. A laptop is an even more restricted thermal environment. Are we going to see benchmarks where Clarksfield gets outstanding numbers on things like video encoding of short clips and then have these numbers fall apart when somebody tries to encode a real movie? That is what happens to Clarksfield when put in a position where it has to perform for more than a few minutes at a time. Like it or not we need to see things in a different light and it becomes important that we factor in other elements into the benchmark numbers like video clip lengths. That is video benchmarks should show times to encode clips of lengths like 10, 30 60 minutes. This to give us an idea of what any possible thermal throttling looks like, which would seem to be highly likely in a notebook.

That is video encoding but this needs to be extended to any sorts of operations that a normal user may expect to take a lot of CPU time and thus heat up the processor possibly triggering throttling. This means three D renderings, and many of the other tests in the bench mark suite. The only point here is to discover just how sensitive the processors are to throttling based on stock and tricked out configurations.

Dave

[lem79]

Well, one of the first Anandtech Core i7 articles tested a Core i7 with stock HSF with a Radeon HD 4870 in a closed case. Turbo never engaged.

Intel has made honest benchmarking that much more complicated now. Because we all know that we run our processors with premium air cooling on an open test bench in an air conditioned room. :P

[wizard69]

Well, one of the first Anandtech Core i7 articles tested a Core i7 with stock HSF with a Radeon HD 4870 in a closed case. Turbo never engaged.

Intel has made honest benchmarking that much more complicated now. Because we all know that we run our processors with premium air cooling on an open test bench in an air conditioned room. :P

Yeah I know! I just want some hard evidence that this is a lot of BS on Intels part. It is not like I'm going to buy a desktop with Lynnfiled right away, I'm actually more interested in the results one will get from the Clarksfield in the coming "laptops". If it is truly difficult to realize a Turbo Boost advantage in a laptop I think many people will be very disappointed as these processors don't look all that good without that feature kicking in.

One of the big promises of this generation of processor was or is, that it can run older single threaded code really well and at the same time offer up impressive results for the more heavily threaded code of the future. It is beginning to look like this is not the case for the stock processor on an average PC board and heat sink combo.

I'm just hoping the benchmarking community clears this up. It certainly looks like skipping this series of i7 & i5 processors might be a good idea. Or at the very least pay real close attention to how a manufacture cools that processor. In a notebook I'm really wondering if the processor would ever break out of the base speed configuration, or even possibly throttle back from there.

Frankly it is a good thing the economy is so bad, it just means I won't rush out and buy something that is half assed.


Dave

[justapost]

Well, one of the first Anandtech Core i7 articles tested a Core i7 with stock HSF with a Radeon HD 4870 in a closed case. Turbo never engaged.
Interesting, however powerconsuumption is alot better with p55 and i5 so things might have changed.

Intel has made honest benchmarking that much more complicated now. Because we all know that we run our processors with premium air cooling on an open test bench in an air conditioned room. :P
One thing makes honest benching really difficult, that is neighter /proc/cpuinfo nor cpufreq-info show multis above 21x even if benchmark results clearly indicate that the cpu ran at 3-3.2GHz. Here I can disable temperature based downclocking in the bios but the tdp limit is fixed, so temps will still be a limit in not so suitable environments. It would be nice to monitor the cpu's current power consumption via lm_sensors and to get the right multis displayed in /proc/cpuinfo.

[lem79]

Interesting, however powerconsuumption is alot better with p55 and i5 so things might have changed.
I was thinking that.

One thing makes honest benching really difficult, that is neighter /proc/cpuinfo nor cpufreq-info show multis above 21x even if benchmark results clearly indicate that the cpu ran at 3-3.2GHz. Here I can disable temperature based downclocking in the bios but the tdp limit is fixed, so temps will still be a limit in not so suitable environments. It would be nice to monitor the cpu's current power consumption via lm_sensors and to get the right multis displayed in /proc/cpuinfo.
It isn't clear what a "not so suitable environment" actually is. I'm guessing Summer time will see a lot less (if any) Turbo action in a lot of places.

I do like the idea of Turbo, it makes logical sense. Though Intel's current implementation and software support for it still seems to be a bit immature (like you said incomplete monitoring support). Supposedly AMD's next arch will have something akin to Turbo, hopefully by then both companies' implementations will support complete monitoring .. then we could see Phoronix Test Suite monitor and provide results which take turbo modes into account, relative to temperatures etc.

[bnolsen]

I put together an i7 860 system with a gigabyte p55m-ud4 board and 4x2GB dimms.

Ran it in parallel with a dual 1.6 clovertown setup with 8GB.

Programs are multi threaded, scaling linearly. Threads are multiple instances of the same instructions run on data packets.

Automatch: 860: 7x 1.6x faster
Least Squares Sparse System: 860: 2.6x faster
OrthoRectification: 860: 1.6x faster

Other tests resulted in the 860 system being IO limited.
I have a dual hapertown 2.33 system I'd also like to run tests against.

I would hopefully expect a dual core i7 system to really rock.