<html><head></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><div>There are some comparisons on Phoronix from a while back... like these...</div><div><br></div><div><a href="http://www.phoronix.com/scan.php?page=article&item=amd_bulldozer_gcc47">http://www.phoronix.com/scan.php?page=article&item=amd_bulldozer_gcc47</a></div><div><a href="http://www.phoronix.com/scan.php?page=news_item&px=MTE5Mzk">http://www.phoronix.com/scan.php?page=news_item&px=MTE5Mzk</a></div><div><br></div>there are so many classifications now what with rankings like Green Graph500 and HPEC Challenge... you can probably find a system of benchmarks that fit your programming model or architecture best. General Purpose computing seems to be disappearing as a priority... probably because most folks never used their architecture for more than one primary applications...<div><br><div><div>On Mar 13, 2013, at 3:29 AM, Tom Metro wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div>Federico Lucifredi wrote:<br><blockquote type="cite">Any nice comparison of per-core performance between AMD and Intel?<br></blockquote><br>Do you literally want a comparison that looks at the performance of an<br>individual core, in isolation?<br><br>The comparisons I've seen uses the usual benchmarks, which attempt to<br>emulate real-world usage, and thus they're impacted by the imperfect way<br>tasks are distributed to multiple cores, and memory bandwidth, etc.<br><br>What's your application?<br><br><br><blockquote type="cite">Because, looking at the FX processors six and eight core prices, they<br></blockquote><blockquote type="cite">are nowhere near what a 6 core Intel chip costs… so i'd like to educate<br></blockquote><blockquote type="cite">myself on the why.<br></blockquote><br>The short answer is they're cheaper because they perform considerably<br>worse. At least that's what I've read.<br><br>I don't recall a specific comparison, but at some point I read one at<br>Anandtech or Tom's Hardware.<br><br>Here's one:<br><a href="http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested">http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested</a><br><br> Last year's launch of AMD's FX processors was honestly disappointing.<br> The Bulldozer CPU cores that were bundled into each Zambezi chip were<br> hardly power efficient and in many areas couldn't significantly<br> outperform AMD's previous generation platform. Look beyond the direct<br> AMD comparison and the situation looked even worse. In our conclusion<br> to last year's FX-8150 review I wrote the following:<br><br> "Single threaded performance is my biggest concern, and compared to<br> Sandy Bridge there's a good 40-50% advantage the i5 2500K enjoys<br> over the FX-8150. ..."<br><br>Here's a thread linking to reviews comparing the FX-6300 to the i3:<br>http://www.tomshardware.com/forum/id-1623327/6300-gaming.html<br><br><br>The problem with these sites is that they review with desktop or gaming<br>performance in mind. What I wanted to know about was VM performance.<br>Can, for example, you attain better performance per dollar running 8 VMs<br>on an AMD 8-core CPU than you can on a 4-core i7?<br><br><br>In other CPU news, I see...<br><br>Calxeda's ARM server tested<br>http://www.anandtech.com/show/6757/calxedas-arm-server-tested<br><br> At first sight, the relatively low performance per core of ARM CPUs<br> seems like a bad match for servers. The dominant CPU in the server<br> market is without doubt Intel's Xeon. The success of the Xeon family<br> is largely rooted in its excellent single-threaded (or per core)<br> performance at moderate power levels (70-95W). Combine this<br> exceptional single-threaded performance with a decent core count and<br> you get good performance in almost any kind of application.<br> [...]<br> Having four 32-bit Cortex A9 cores, each with 32 KB instruction and 32<br> KB data L1 per-core caches, the processor block is somewhat similar to<br> what we find inside modern smarphones. One difference is that this SoC<br> contains a 4MB ECC enabled L2 cache, while most smartphone SoCs have a<br> 1MB L2 cache.<br> [...]<br> A dual Xeon E5 or Opteron 6300 server has much more processing power<br> than most of us need to run one server application. That is the reason<br> why it is not uncommon to see 10, 20 or even more virtual machines<br> running on top of them.<br> [...]<br> Each server node has one quad-core Cortex A9 with 4MB of L2 cache and<br> 4GB of RAM. With that being the case, the question "what can this<br> server node cope with?" is a lot more relevant.<br> [...]<br> The ARM based server is a pretty bad choice right now for memory<br> intensive workloads. Even with four cores and DDR3-1333, the useable<br> bandwidth is less than one sixth of what one Xeon core can sustain.<br><br> In a similar vein, the ECX-1000 is not capable of providing more<br> bandwidth than an Atom system equipped with DDR2-667. However, both<br> the Atom and ARM cores are pretty bad when it comes to bandwidth.<br> [...]<br> Clock for clock, the out-of-order Cortex-A9 inside the Calxeda<br> EXC-1000 beats the in-order Atom core. A single A9 has no trouble<br> beating the older Atoms while likewise coming close to the much higher<br> clocked N2800. The N2800 and ECX-1000 perform similarly.<br> [...]<br> Looking at both decompression and compression, it looks like a quad<br> ARM A9 is about as fast as one Xeon core (without Hyper-Threading) at<br> the same clock. We need about six A9 cores to match the Xeon core with<br> Hyper-Threading enabled. The quad-core ECX-1000 1.4GHz is also close<br> to the dual-core, four-threaded Atom at 1.86GHz. This bodes well for<br> Calxeda as the 6.1W S1240 only runs at 1.6GHz.<br> [...]<br> We created 24 virtual machines on top of the Xeon server. ...<br> [Calxeda's] server gets the same workload, but instead of using<br> virtual machines, we used the 24 physical server nodes.<br> [...]<br> At the low concurrencies, the Intel machine leverages turboboost and<br> its exceptionally high per core performance. At the higher web loads,<br> the total throughput of the 96 (24x quad-core SoCs) ARM A9 cores is up<br> to 50% higher than the low power 32 thread/16 core (2x Octal core)<br> Xeons. Even the mighty 2660 cannot beat the herd of ARM SoCs.<br> [...]<br> ...each server needs about 8.3W (200W/24), measured at the wall. That<br> is exactly what Calxeda promised: about 6W (at 1.4GHz) per server node<br> (measured internally), up to 8.5W measured at the wall (again at<br> 1.4GHz). That is nothing short of amazing if you consider the<br> performance numbers.<br> [...]<br> Let's be clear: most applications still run better on the Xeon E5. Our<br> CPU benchmarks clearly indicate that any application that accesses the<br> memory frequently or that needs high per thread integer processing<br> power will run better on the Xeon E5. Compiling and installing<br> software simply feels so much faster on the Xeon E5, there is no need<br> to benchmark.<br><br>(See the article for lots more details.)<br><br>The 8-core AMD parts may be useful in a similar way to the ARM parts,<br>except the AMD parts tend to get dinged for power consumption, even<br>compared to the higher performing Intel parts. So even if they do make<br>sense in this context, it won't be long before the ARM servers take over<br>that space.<br><br> -Tom<br>_______________________________________________<br>Hardwarehacking mailing list<br>Hardwarehacking@blu.org<br>http://lists.blu.org/mailman/listinfo/hardwarehacking<br></div></blockquote></div><br></div></body></html>