Haswell E

I think the most interesting CPU in the lineup is the 5930K. The 8 cores of the 5960X are great but 3Ghz is a bit pathetic given there are quad cores with the same architecture running at 4Ghz and I'm pretty sure there are Ivy based Xeons with 8 cores running at more than 3Ghz.

At 3.5Ghz though the 6 core 5930 looks pretty nice.

For those interested in the raw specs it can use over 68GB/s of memory bandwidth (the entire DDR3 bandwidth available to the XBO) and has a SIMD capacity 672 GFLOPS.
 
I think the most interesting CPU in the lineup is the 5930K. The 8 cores of the 5960X are great but 3Ghz is a bit pathetic given there are quad cores with the same architecture running at 4Ghz and I'm pretty sure there are Ivy based Xeons with 8 cores running at more than 3Ghz.

At 3.5Ghz though the 6 core 5930 looks pretty nice.

For those interested in the raw specs it can use over 68GB/s of memory bandwidth (the entire DDR3 bandwidth available to the XBO) and has a SIMD capacity 672 GFLOPS.

They have need to cut TDP at 140W .. Im pretty sure that you will not have any problem for get 4GHZ ( 4.5ghz with proper cooling )
 
Quadcore haswell does close to 4GHz for half the 140W of Haswell-E. It's a strangely low clock, considering the process they make the chip on will be quite mature by now, and it's not a new architecture either...
 
Can't say I'm too impressed by Intels decision to leave the 8 core only to the Extreme model and gimping the PCI-lanes on the entry model.
 
The pricing of the extreme model is also...ahem...fitting. Although not in a good way.
 
WTF? Intel is cutting down on the PCI-Express lanes and asking an additional $200 for them?

That's quite the blow on the people who were thinking of using the bridge-less Crossfire.
 
You can always count on Intel to be cutting-edge on both tech and evil.

IE, no surprise, really... They cut features from haswell-K while charging more for it for example, for absolutely no other reason than that they can.
 
You can always count on Intel to be cutting-edge on both tech and evil.

IE, no surprise, really... They cut features from haswell-K while charging more for it for example, for absolutely no other reason than that they can.

The annoying part is that its predecessors had 40 lanes for two generations now:

Core i7 3820: 40 lanes PCI-Express 2.0 at release price $294
Core i7 4820K: 40 lanes PCI-Express 3.0 at release price $310
Core i7 5820K: 28 lanes PCI-Express 3.0 at release price $390


The lower end of the LGA2011 platform used to be a good alternative to the higher-end of the LGA115x line.
Prices were similar, CPUs clocked somewhat better, much more PCI-Express lanes for upgrades and the socket lasted a lot longer.
I guess that wasn't interesting to Intel.
 
The annoying part is that its predecessors had 40 lanes for two generations now:

Core i7 3820: 40 lanes PCI-Express 2.0 at release price $294
Core i7 4820K: 40 lanes PCI-Express 3.0 at release price $310
Core i7 5820K: 28 lanes PCI-Express 3.0 at release price $390


The lower end of the LGA2011 platform used to be a good alternative to the higher-end of the LGA115x line.
Prices were similar, CPUs clocked somewhat better, much more PCI-Express lanes for upgrades and the socket lasted a lot longer.
I guess that wasn't interesting to Intel.

I actually think 5820k isn't that bad. It still has quite a bit more PCI-E lanes than the 1150 socket and I think 8x PCI 3.0 is enough for multi GPUs. So there is a fair chance that you won't have to give up anything meaningful and it's still six cores at lower price than what was available previously. The initial high prices for DDR4 push the platform costs quite high though. Well I'm glad I have no urge to upgrade at this point in time.
 
Actually, the 5820K is cheap, its a good way for make enter peoples into the 6 cores ( and entusiast series ). For peoples coming from 4 cores / 8threads Haswell or Ivy. ( well, little things anyway, motherboard are in general not cheap in this series ).

Intel have do an excellent choice here, as the 4820K was not really do any sense over a 4770K on the Ivybridge series. ( outside offcourse quadchannel vs dual channel who double the memory bandwith, more Sata port, more advanced features on the motherboard southbridge ).

Strangely the 5930K-5820K offers nothing against a 4930K, the performance gain is negligble at best ( 2 to 5% performance gain maximum in Cinebench, encoding etc ( even slower on some like H264 4K encoding or winrar ).

Then ofc, here come the 8cores, but you really want to need it for professional software, and even at 1000$ it is still cheaper than any 8 cores Xeon based system.

Personally i have an I7 4930K, i dont see any upgrade here possible, outside the 8cores, but for now, i pass. ( new motherboard + cpu + ram = 1600$ upgrade, i can live with my 6 cores for a bit more of time ).
 
Last edited by a moderator:
Ooh good, i have not think to go see the TR review, and i see many benchmarks interesting there. Too bad they have not include 5930K, but i expect this is given a better light on the 5960X performance and they will follow with a second article.
 
5820k could be the new 920, I just can't find any review where they test the OC on that. Anyone come across that?
 
Bizarrely, h264 doesn't seem to scale to 16 threads, at least according to Tech Report: http://techreport.com/review/26977/intel-core-i7-5960x-processor-reviewed/10

That's because, for whatever reason, they chose to add scaling to x264's parameters. (Test one thing at a time people!).

x264's scaling filters are based on libswscale and is not multithreaded. As the TR folks are only testing `preset medium`, the multiple cores doing the actual encoding are chilling around waiting for the single core doing to the scaling. They might as well rename that specific benchmark to "single-threaded Bilinear scaling speed".

Handbrake uses x264 as its main encoder, so that's the more relevant benchmark.
 
That's because, for whatever reason, they chose to add scaling to x264's parameters. (Test one thing at a time people!).

x264's scaling filters are based on libswscale and is not multithreaded. As the TR folks are only testing `preset medium`, the multiple cores doing the actual encoding are chilling around waiting for the single core doing to the scaling. They might as well rename that specific benchmark to "single-threaded Bilinear scaling speed".

Handbrake uses x264 as its main encoder, so that's the more relevant benchmark.
They are talking about utilizing 16 threads, not frame scaling and filtering.
 
They are talking about utilizing 16 threads, not frame scaling and filtering.

And the reason they are not utilizing 16 threads is that they have added

--video-filter resize:1280,720

to their benchmark command line, which runs single-threaded frame scaling as part of the test. Should they remove that, the 16 cores would be well utilized.
 
And the reason they are not utilizing 16 threads is that they have added



to their benchmark command line, which runs single-threaded frame scaling as part of the test. Should they remove that, the 16 cores would be well utilized.
I see, thanks. I guess I misunderstood. :)
 
Back
Top