aristocratic3d Posted January 6, 2015 Share Posted January 6, 2015 HI, 1) xeon 2670 v3 = 12 core = 24 thread x 1 piece = 24 thread 2) xeon 2620 v2 = 6 core = 12 thread x 2 pieces = 24 thread 3) i7 4790 = 4 core = 8 thread x 3 pieces = 24 thread My questions, 1. are these 24 threads have the same power regardless of the processor? 2. If "no", what are the differences? Another question not related to the topic above- 1. Why should not I build a dual xeon 2670 v3 processor instead of six i7 4790 if I intend render nodes while both have 48 threads? waiting for experts, Thanks in advance! Abdullah Link to comment Share on other sites More sharing options...
numerobis Posted January 6, 2015 Share Posted January 6, 2015 (edited) 1. are these 24 threads have the same power regardless of the processor? 2. If "no", what are the differences? No. There are basically three factors besides the core count that have influence. 1. Processor architecture. Different CPU models or generations are normally based on different architectures. Different architectures can result totally different in single and multi core performances. The CPUs above are based on two different generations: Haswell/Haswell-E for the i7 4xxx and the Xeon E5 v3 and Ivy Bridge-E for the E5 v2. Haswell (-E) is a few percent faster than Ivy Bridge (-E) at the same frequency (depending on the app). Which leads to the second and most important point when compairing similar architectures... 2. Core frequency. You have to compare the operating frequency of the cores. There are huge differences with different processor models from roughly 2Ghz to 3.5 GHz today which will result in totally different performances at the same core count. And regarding frequency you have to differentiate between base clock, single core turbo and multi core turbo. Todays CPUs even apply different frequencys for the core counts inbetween. What counts for the combined multicore performance is the multicore turbo for all cores. 3. Hyper-Threading. Intel Hyper-Threading is based on two logical processors per core that share the same physical core. HT can improve the performance for multi threaded tasks. All bigger CPUs have it today , some low end or mid range CPUs do not have it. 1. Why should not I build a dual xeon 2670 v3 processor instead of six i7 4790 if I intend render nodes while both have 48 threads? Advantages of one big System: depending on the system count less RAM and other components are needed. Only one box that you have to administrate. Only one license needed (depending on the software). Direct access for a workstation, no network rendering needed. Disadvantages of one big System: Normally more expensive. But with todays high RAM prices the difference is not so huge and in some cases a dual CPU setup can be at the same level or even cheaper. BUT this all applies only if overclocking in no option for you. The current Xeons are not overclockable (only by a few percent) so this is an advantage of the i7. If it is an option, using i7's can give you more performance, especially in single threaded tasks (and hopefully the RAM prices will drop again in the not too distant future). Maybe less flexible if you have to render more than one job. For a rough comparison of similar architectures you can multiply the cores with the all core turbo clock. So for your systems this would mean: 1) Xeon E5-2670 v3: 12x 2.6= 31.2Ghz (single core turbo 3.1GHz) 2) Xeon 2620 v2: 6x 2.4GHz= 14.4GHz (single 2.6Ghz) 3) i7 4790: 4x 3.6GHz= 14.4GHz (single 3.9GHz) (all hyperthreaded - Haswell is maybe 3-5% faster than Ivy Bridge at the same frequency) http://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors#Haswell-based_Xeons A quadcore for rendering really makes no sense today with the availability of a cheap and overclockable hexacore like the i7 5820K. Edited January 7, 2015 by numerobis Link to comment Share on other sites More sharing options...
aristocratic3d Posted January 6, 2015 Author Share Posted January 6, 2015 HI Mr. Numerobis, Thanks for elaborate answer. It was a great help. I prepared some option based on your explanation and my budget. which one would you buy if you were me?- 1. Xeon 2670v3 2.6GH x 2 = 256000 MOBO = 50000 RAM (ecc) 32GB = 54000 psu = 20000 >>>>>>> 2.6GHx24 core = 62.4 HDD = 5000 casing = 20000 total = 405000 tk = $5200 2. i7 4790k = 25000 MOBO = 8000 RAM 32GB = 23000 psu = 8000 x 5 pieces = $4750 >>>>> 4GHx20 core = 80 HDD = 5000 casing = 3000 total = 72000 tk = $950 3. i7 5820k (6 core) = 33000 MOBO = 27000 RAM (ddr4) 32GB = 44000 psu = 12000 x 3 pieces = $4950 >>>>> 3.3GHx18 core = 59.4 HDD = 5000 casing = 5000 total = 126000 tk = $1650 4. i7 5960x (8core) = 93000 MOBO = 27000 RAM (ddr4) 32GB = 44000 psu = 12000 x 2 pieces = $4800 >>>>> 3GHx16 core = 48 HDD = 5000 casing = 5000 total = 186000 tk = $2400 5. Xeon 2620 v2 2.1GH x 2 = 60000 MOBO = 50000 RAM (ecc) 32GB = 54000 x 2 pieces = $5400 >>>>>>> 2.1GHx12 core = 25.2 psu = 20000 HDD = 5000 casing = 20000 total = 209000 tk = $2700 which one would you choose and why? Thanks again for your help. Abdullah Link to comment Share on other sites More sharing options...
numerobis Posted January 6, 2015 Share Posted January 6, 2015 (edited) The price of the processor is only one thing. To compare things on a realistic basis you have to look at the whole system. Configure some basic machines incl. software licenses and compare the cost per GHz. I have done this a few months ago but i think to get realistic numbers you should do it for yourself. But you should use the all core turbo values and not the baseclock like you have done in your calculation. You can find the values for the Xeons at the wikipedia page i posted above. Just as example here are the values of a dual 12-core system vs a 5820K stock and overclocked. One thing you have to consider is that you need 8 RAM modules to enable quad channel on a dual CPU system. So for 32GB this would mean 8x4GB. But i haven't found 8x4GB ECC so i have chosen 64GB for the 2P system vs 32GB for the 5820K which is a bit distorting competition but i think it is ok, because this is the only way it can be set up. All configurations include the same cheap video card because it was done as a render node comparison. 2x12-core E5-2690 v3; 24x3.1GHz=74.4GHz; 4400€ (excl.VAT); 59.1€/GHz 6-core i7 5280K; 6x3.4GHz=20.4GHz; 1150€; 56.4€/GHz 6-core i7 5280K @4.3GHz; 6x4.3GHz=25.8GHz; 1150€; 44.6€/GHz Edited January 7, 2015 by numerobis Link to comment Share on other sites More sharing options...
alexanderbong Posted January 7, 2015 Share Posted January 7, 2015 One thing you have to consider is that you need 8 RAM modules to enable quad channel on a dual CPU system. Ah, I am about to purchase a dual xeon system too. Good thing I ran into this thread. The system I had in mind has 4x16GB rams. Reasons are mainly room for upgrades at a later date when needed. Do you mind elaborating on the advantages you get from 8 ram modules, instead of 4? Link to comment Share on other sites More sharing options...
alexanderbong Posted January 7, 2015 Share Posted January 7, 2015 (edited) Read up some articles about this and arrived to some conclusions. Let me know if I'm right or wrong; In quad channel setting, you have 4 ram module to work in parallel to get a speed boost (quadruple the speed?). In a dual processor system, you mentioned that we need 8 ram module to activate this. Which begs the question, are ram modules not share-able in a dual processor system? ie, 64gb ram in dual xeon essentially only has 32gb? EDIT: Apparently the speed doesn't quadruple, Internets mention some 20-30% increase? In which case, the option to upgrade into 128 gb ram is more favorable than the speed increase. Edited January 7, 2015 by alexanderbong Link to comment Share on other sites More sharing options...
numerobis Posted January 7, 2015 Share Posted January 7, 2015 I think the difference should be more like 1-2%. I don't have numbers for a dual Xeon system - maybe someone could test this comparing the rendering performance with dual and quad channel - but here are some results for a 3960X: http://www.legitreviews.com/ddr3-memory-performance-analysis-on-intel-x79_1779/4 I think the video encoding could be most comparable to rendering, but i'm not sure about that. But you're right... it looks like 16GB DDR4 reg ECC modules are already cheaper than the same amount of RAM with 8GB modules... strange. Hopefully we'll also see standard DDR4 16GB modules soon. Link to comment Share on other sites More sharing options...
alexanderbong Posted January 8, 2015 Share Posted January 8, 2015 Yeah, once the scene are loaded into memory, i don't think the speed matters at all. I bought my system from Dell, at 64GB, the 16x4 cost $150 cheaper than 8x8. Link to comment Share on other sites More sharing options...
numerobis Posted January 8, 2015 Share Posted January 8, 2015 Ah, I am about to purchase a dual xeon system too. Good thing I ran into this thread. I bought my system from Dell, at 64GB, the 16x4 cost $150 cheaper than 8x8. Now that was a quick decision Link to comment Share on other sites More sharing options...
alexanderbong Posted January 8, 2015 Share Posted January 8, 2015 can't wait to get my hands on this beast. takes 3 weeks to ship out tho Q_Q. first thing im doing is opening task manager and counting the threads Link to comment Share on other sites More sharing options...
simonm Posted June 12, 2015 Share Posted June 12, 2015 hey guys - how did you go with this? Is it much better than your new i7s? Link to comment Share on other sites More sharing options...
alexanderbong Posted June 12, 2015 Share Posted June 12, 2015 Yes, they scale nicely when rendering. Single core usage is crap though. If you are thinking of buying multiple i7s for render slaves, then this is the better option. Link to comment Share on other sites More sharing options...
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now