HP's 2P DL385 Barcelona server frags HP DL580 4P Quad-Core Xeon server
A 2P DL385 G5 2.3GHZ Barcelona Opteron server (total 8 cores) can handle 174 Heavy Users in HP's benchmark.
A 4P HP 2.3GHZ Opteron blade server can handle 301 heavy users .
In comparison, a 4P 2.9 GHZ Intel Quad-Core Xeon Dl580 G5 server (total 16 cores) does 233 Heavy users.
The result: a 4P 2.3GHZ Opteron blade frags a 4P 2.9GHZ Quad-Core Xeon server by a whopping 29.2%.
But this is not the end of the story. If you look at the number of light users, the HP DL580 G5 4p Quad-Core Xeon (total 16 cores) can do 386 Light Users. The HP DL385 2P 2.3GHZ server (2.3GHZ) can handle 380 Light Users.
It is obvious that with a slightly faster 2.4GHZ Opteron, the 2P DL385 will frag the 4P DL580. In other words, one Opteron quad-core CPU frags two Xeon quad-core CPUs in an SMP configuration.
It is clear that Intel CPUs are only good for single threaded 1P operation. The fact that a 2P-8 core Opteron server frags a 4P-16 core Xeon server is an old story. With 16 Xeon cores hanging on a FSB, each Xeon core gets the bandwidth of a 80486.
posted by Sharikou, Ph. D. @ 4/27/2008 12:35:00 PM
10 Comments:
HP’s 4P 585 G5 16 core 2.3 Ghz system produces much the same kinds of results against the faster 73xx powered 580 G5 16 core set up seen here ftp://ftp.compaq.com/pub/products/servers/benchmarks/DL585G5_top4Ppriceperf_TPCC_033108.pdf and here ftp://ftp.compaq.com/pub/products/servers/benchmarks/dl580g5_tpcc_sept0507.pdf.
This is typical of the difference AMD has over the Intel cores where as real world heavy duty multi tasking work loads are concerned. A slower (both in clock rate and actual integer performance) AMD producing the same multi-tasking results is not uncommon. That’s also why the lower core count AMD models can approach the same performance level of the higher core count Intels when work loads increase. When Intel copies AMD’s on board memory manager that will probably change but the economic cost difference between the two will not. The TPC-C differences between the two nearly identical performing systems above is the result of the AMD having a lower system cost. It is not just the CPU cost differences that account for this. The specialized memory Intel continues to use pushes up Server cost a lot. Here again Intel is trying to off set the disadvantage it has with memory intensive tasking by avoiding putting the memory manager on die.
LostCircuits published here http://www.lostcircuits.com/ a run on the AMD Phenom X4 9850 B3 step in early April where they ran it up to 3.3-3.4 Ghz with passive cooling and upped the core voltage to 1.4 volts. It was stable at 3.2 Ghz and 1.4 volts with OEM supplied cooling. I ask them what they got without running up the core voltage and they said 3.0 at stock voltage and rock solid. That may not impress those that are willing to pay over $1000.00 for one (or two) of the Intel high end QX but I strongly suspect that for 99.9 % of the market a Phenom running at 3.0 Ghz safely without cooling modes and two or three 3870 Radeons for less money than just one high end QX will be more meaningful in the scheme of things for the overwhelming majority of computer users and gamers. The 2008 Corvette ZR1 with 620 hp is a great sports car but it’s pointless to buy one to drive to work in. Same principle here.
Louis, my name is Thom not Sharikou. I don’t use aliases. I don’t have anything to do with this Blog or the slightest knowledge of HTML or Webs. I’ve worked in the IT field since 1973. Wrote my first computer program on paper for a variable word length computer in high school in 1970. Know what a variable word length computer is? Know what that assembly language code looks like? I’ve got hundreds of Intel servers, dozens of Sun Sparc and half a dozen IBM mainframes to play with at work. My corporation has over 120,000 mostly Intel based desktops and laptops to maintain and the bulk of them are still Pent 4s with clock speeds less than 2.8 Ghz. The Corporation spent $100,000 just on extra AC to cool all the Pent 4 based Servers we have in the Data Center. The latest, greatest, fastest CPU doesn’t have much value in the real world where all this horsepower is actually put to use 24/7. Our first AMD 4P is going to be installed in June. It is 40% faster than the fastest Intel 71xx series Xeons we have and we have hundreds of those. The total system and operating cost over the lease life is less than the matching Intel model by a meaningful amount. The lights stay on at Intel and AMD based on what the masses buy and Corp America. The typical consumers aren’t really interested in who is on first any longer. If AMD disappears, total sales will decline. 30,000 less Intel employees times the average IT salaries where Intel is located is billions of profit each year just in that. I don’t see Intel going BK. I do see them being hit with a large fine for their business practices either in Europe, here or both. They are not your friend. No monopoly is.
Would you call anyone you didn’t know “douche bag” to their face?
Like Thom and Sharikou I too have had a history in computing going back to pre-computer era where I repaired mechanical teleprinters, then repaired gate level boards then Microprocessor single board design banking terminals.
A tremendous learning experience going from seeing how the things worked to running debuggers and analyzing machine code.
I designed a co-processor add-in for my atari 400 using a NatSemi32 bit FPU processor and wrote a maths interface routine , It worked but was too complicated for Byte Magazine to have as a DIY article..
anyway I look at processors for elegance of design and implementation.. core2 duo "P3 rob Caldwells" brilliant design with some good tweeks for IPC"
is A great chip, in dual configuration, anything beyond that one runs into FSB gridlock..
Nehalem Intel's recognition of AMD superior design for scalable gluless CPU is on the right track.
AMD's Barcelona is here now!
With Intels past history of failure to deliver promises pre-C2D era,
and that it took AMD to extend X86 to X86-64, Intel had 4 goes a it and then canned it as un-doable, Instead sank billions into Itanium.
Intel has it all to loose if it has delays for Nehalem ..
Good luck to Intel and AMD
thom you proved that people who use AMD are more intelligent than Intel users ;)). Off course its joke but this blog proves my words.
Regarding Phenom processors, they are very good and I think better than core2Quads, Amd just needs better cash managmant, ( lower cash latency e.t.c. ) i think it's just a question of time to return back P4 Athlon64 days when AMD fraged Intel analogues in every field.
Thom,
He wasn't talking to you, he was addressing the blog post and thus all his comments were directed toward Sharikou. You can tell this because the first word of his post is "Sharikou", not "Thom".
It would have been really rude of him to call you a douchebag without know much about you. However, he was calling Sharikou that, who has actually been scientifically proven to be a douche-bag.
Erm, the 4P Intel system ran out of memory (or rather, PTEs) before it saturated its processor. That's not telling us anything about the processor's capabilities; it's telling us that 32-bit Windows has some limitations when it comes to large memory systems (namely that it can run out of PTEs).
Which makes the entire benchmark useless.
By contrast, the 4P AMD system had 32 GB RAM and 64-bit Windows.
No wonder it scored higher--it could actually saturate the processors!
Louis,
If you have a MS in Computer Science and are employed in this field in accordance with that degree, your attitude and use of words that have nothing to do with the topic at hand would get you unemployed in about 30 minutes. Everyone has an opinion but an informed opinion worth what any employer typically pays a high end IT professional doesn’t sound anything like your post. Your response, along with others have side stepped the content of the original post completely and concentrated on juvenile name calling and personal attacks. I didn’t have that kind of time to develop those kinds of juvenile skill sets during all the nights I spend in school getting my degree. I’ve never met any employed IT professional that speaks the way you do. Trying to make a point with personal attacks and name calling is not the sign of a professional, degree or not.
Scott,
Trying to out do Louis?
In less than five years most large data centers will have some form of virtualized x86 platforms in place. All this heat and lightening about who’s on first is going to give way to what IT professionals have known for some time. There’s a glut of X86 CPU power that can’t be utilized. Glut equals wasted money which is always in short supply in the real world. Glut means less high end CPUs are going to be sold. I’ve spent five years trying to raise the CPU utilization of our X86 (and Sun Sparc) systems only to be blocked by things that most of you never consider when you buy a high end desktop. Two thirds of our X86 servers could be gone tomorrow in a virtualized environment. Guess which X86 brand handles high volume memory intensive loads better under such schemes? The Blog post that Sharikou posted and I appended to point to this simple fact. Is Intel aware of this, certainly. They’ve been aware of this for over a decade and done nothing about it. Our fastest Netburst Servers were 3.0 Ghz 7120 based. There was nothing to be gained from paying more to move up to the 3.5 Ghz level. There are a hundred cache based benchmarks I can Google that show the 3.5 Ghz 7150 will beat the crap out of a 7120 but none of that power showed up in server based benchmarks. The 51xx/53xx series didn’t even show up in 4P and above servers. Why? They are clearly faster than the 71xx or the AMD Opterons (per a hundred cache based benchmarks). The 73xx will clearly smoke the Opterons in all the important game based benchmarks. No question about it.
So in the larger scheme of things why is this important to any individual buying a desktop or even a laptop? As has been shown since the beginning of this profession, application use grows to fill the available memory capacity. As that grows more and more multitasking demands are made on the CPU and its memory structure. Just installing Vista 64 bit will more than double the OS’s need for memory over and above XP 32 Bit. It seems absurd that Vista Ultimate will actually gain performance going from 4 Gig of memory to 8 but that’s the way it works and that trend isn’t going to change in the near term. It won’t be long before your typical desktop has to have more memory than many of our servers at work now. Some laptops today have more memory than most of our 2P servers today. It won’t be that long before what once was called a server will simply be called a Home PC/Entertainment Console/Security Monitor/Telecommunication Central etc. When that day comes getting the most out of what your CPU(s), memory can provide at a cost you can live with will become more important than it generally is today. The bulk of the consumer market shops on price today all else equal. AMD systems usually cost 150-200 a system less for comparable performance. I haven’t seen anything that makes me just want to run out and buy the latest AMD Quad system even at what are bargain prices compared to Intels. The Extreme Edition Quads from Intel have no presence in the market place today outside of early adaptors and all those sites that get them for benchmark purposes for free. There are always some people with more money than brains around. AMD can’t play this game and Intel probably shouldn’t either. The 51/53/73xx Intel series weren’t cheap to design or produce. If they had been 30,000 Intel employees would not have lost their jobs to pay for their development.
Mean while my 2002 laptop with an Intel 1.6 Ghz Pent M installed is still doing what it has always done. Waiting on I/O or the Internet in this case. I’m sure Sharikou or who ever host his Blog is capable of defending his post but it would seem appropriate for professionals to address the content of the post rather than drop by to just demonstrate the juvenile nature of the person hiding behind the Blog name.
One last point Louis, in the wrong corner of this world, region, even street in some American cities, calling someone a name simply because you think you are entitled to do so will get you beat up at the very least and killed out right in some places even in this country. I seriously doubt you would follow through on your boast to a complete stranger in the flesh. If you would, you are beyond being a fool. Even Angels can’t protect all the fools all the time. Nothing in this life is worth learning that the hard way. Nothing in the Blog Sphere is that important to be that foolish about.
Only one post even came close. Folks, READ the pdf's. From pages 2, 3, and 2. I'd go into detail, but there's really no reason.
Important:
This document describes a performance characterization performed
utilizing the HP 64-bit test harness, which incorporates a Microsoft® Office
2003 workload.
Test results cannot be compared directly with the results of tests performed
using the 32-bit Office XP- or Office 2003-based harness.
Important:
This document describes a performance characterization performed
utilizing the traditional HP 32-bit test harness that incorporates a
Microsoft® Office XP workload.
Test results cannot be compared directly with the results of tests performed
using the 64-bit Office 2003-based harness.
My name is Matt and I work for Dell. There are a lot of great comments happening on this post.Well-threaded applications will benefit from a higher numbers of cores, a larger cache and the higher throughput and memory capacity of Quad-Core Xeon servers.
Post a Comment
<< Home