Intel Haswell-EX Gets up-to 18-cores and 36 threads

Published by

Click here to post a comment for Intel Haswell-EX Gets up-to 18-cores and 36 threads on our message forum
https://forums.guru3d.com/data/avatars/m/68/68055.jpg
And it will cost gazillion bucks. But looking forward to Cinebench numbers.
https://forums.guru3d.com/data/avatars/m/258/258664.jpg
Still workstation CPUs, but hey, so many cores... zomg!
https://forums.guru3d.com/data/avatars/m/217/217375.jpg
What a nice chip to replace my Xeon 6core/12thread baby... was just thinking how it would rip through Handbrake encodes! though there is still that minor gazillion bucks issue :s Where is AMD with something Awesome, Shiny, Fast and Cheap when I need it !? Guess I will find my Zen eventually, time will tell in what form it will come I guess 🙂
https://forums.guru3d.com/data/avatars/m/253/253059.jpg
Is that last row in the chart correct? 4 cores with 45MB cache? Is that for people with really cache-dependent workloads or is the cache value a typo?
https://forums.guru3d.com/data/avatars/m/156/156133.jpg
Moderator
Edit- That was v2, my bad.
https://forums.guru3d.com/data/avatars/m/186/186805.jpg
2 of these in a dual socket board for 36 cores and 72 threads!!!! This thing will dominate multithreaded apps!!! But.... 165w TDP!?? WTF???
https://forums.guru3d.com/data/avatars/m/198/198862.jpg
2 of these in a dual socket board for 36 cores and 72 threads!!!! This thing will dominate multithreaded apps!!! But.... 165w TDP!?? WTF???
Yeah but in games it will loose to a 4 core at 4Ghz. 😀
https://forums.guru3d.com/data/avatars/m/186/186805.jpg
Yeah but in games it will loose to a 4 core at 4Ghz. 😀
:D Yea more than likely LMAO
https://forums.guru3d.com/data/avatars/m/258/258664.jpg
Those CPUs are not meant for gamers. 🙄
data/avatar/default/avatar39.webp
What a nice chip to replace my Xeon 6core/12thread baby... was just thinking how it would rip through Handbrake encodes! though there is still that minor gazillion bucks issue :s
Dont tell me you use Handbrake by CPU when it support OpenCL .. ( this said i dont use handbrake, dunno if there's a quality difference ).
https://forums.guru3d.com/data/avatars/m/246/246171.jpg
Intel reports 37.5 mb of cache. Great for workstations.
Not that I'm fluent in a lot of professional workstation applications, but I'm not aware of a single program that can really take advantage of even half of these threads, though I'm open to being corrected about this. This is a server CPU, and it is excessively powerful to almost a stupid level. I think the best use application for a system like this is running dozens of virtual machines, since you could swap those out for a quad core system if you needed to (and therefore are easier to make redundant). The problem, though, is HT is known to perform poorly in some VMs. Otherwise, this CPU is really impractical since it is probably way too costly to be redundant, and redundancy is crucial for servers of this magnitude. I don't know - I'm just thinking out loud. If I wanted a CPU like this, I would prefer a model without HT at a lower price. That's just my opinion though.
data/avatar/default/avatar12.webp
Not that I'm fluent in a lot of professional workstation applications, but I'm not aware of a single program that can really take advantage of even half of these threads, though I'm open to being corrected about this. This is a server CPU, and it is excessively powerful to almost a stupid level. I think the best use application for a system like this is running dozens of virtual machines, since you could swap those out for a quad core system if you needed to (and therefore are easier to make redundant). The problem, though, is HT is known to perform poorly in some VMs. Otherwise, this CPU is really impractical since it is probably way too costly to be redundant, and redundancy is crucial for servers of this magnitude. I don't know - I'm just thinking out loud. If I wanted a CPU like this, I would prefer a model without HT at a lower price. That's just my opinion though.
Hehe, the Primes generator I made can use up to 2^32 -1 threads. And it scales almost linierly per CPU core as well. HT gives about a 20% increase in speed. http://sourceforge.net/projects/fastestprimes/?source=directory
https://forums.guru3d.com/data/avatars/m/156/156133.jpg
Moderator
Not that I'm fluent in a lot of professional workstation applications, but I'm not aware of a single program that can really take advantage of even half of these threads, though I'm open to being corrected about this. This is a server CPU, and it is excessively powerful to almost a stupid level. I think the best use application for a system like this is running dozens of virtual machines, since you could swap those out for a quad core system if you needed to (and therefore are easier to make redundant). The problem, though, is HT is known to perform poorly in some VMs. Otherwise, this CPU is really impractical since it is probably way too costly to be redundant, and redundancy is crucial for servers of this magnitude. I don't know - I'm just thinking out loud. If I wanted a CPU like this, I would prefer a model without HT at a lower price. That's just my opinion though.
Workstations for virtualization, easy swap through different instances in Windows Server. We have one physical server that runs 10 instances at once, we have at least 6 of them turn off at night since they won't be needed. But we need them to also be able to start up at a moments notice. High cache helps out a lot. What's funny is the first instance we have in Windows server is created just as a caching site.
https://forums.guru3d.com/data/avatars/m/228/228458.jpg
2 of these in a dual socket board for 36 cores and 72 threads!!!! This thing will dominate multithreaded apps!!! But.... 165w TDP!?? WTF???
I have 4 folding rigs that I would love to consolidate into 1 rig. But after looking at my options, I have come to find out that it's just too damn expensive lol.
https://forums.guru3d.com/data/avatars/m/217/217375.jpg
Dont tell me you use Handbrake by CPU when it support OpenCL .. ( this said i dont use handbrake, dunno if there's a quality difference ).
Hi Lane, Handbrake's use of OpenCL option to help with scaling only accounts for perhaps 5% of CPU time. I would love if it could make better use of my 290X, but sadly that has relatively little impact on the high quality conversions I do. What do you use instead that can make better use of my GFx card? ----- Sadly for these new chips, I doubt I will be able to OC them like we could in the past. I've been Really enjoying my 2.66GHz Xeon X5650 running at 4GHz these last few years or so. I would Definitely be interested in one of those CPUs if we could add some over the top cooling and see about getting some nice GHz out of them! ^_^
https://forums.guru3d.com/data/avatars/m/246/246171.jpg
Hi Lane, Handbrake's use of OpenCL option to help with scaling only accounts for perhaps 5% of CPU time. I would love if it could make better use of my 290X, but sadly that has relatively little impact on the high quality conversions I do. What do you use instead that can make better use of my GFx card?
It largely depends on the application. Based on my observation, AMD tends to have more streaming processors but shorter pipelines than nvidia. That's why AMD is better for mining and nvidia is better for complex equations. That's also why it's stupid to buy a GPU for gaming purposes strictly because you like the brand - if money isn't a factor, you should buy a GPU for whichever games it plays best with. If you don't have a particular set of games in mind, nvidia is usually the safer bet. If money is a factor, AMD is usually the safer bet. Anyway, I'm guessing handbrake just happens to work better on nvidia.
https://forums.guru3d.com/data/avatars/m/217/217375.jpg
In the case of Handbrake, it simply does not yet make any extensive use of any GPUs capabilities, so it is more or less CPU bound for now. For me as a rule, AMD GFx cards suit my various needs best, pretty much every app I use prefers AMD. - that and helping support the underdog to help keep that Healthy competition alive Always seems a Good move 🙂
data/avatar/default/avatar08.webp
Not that I'm fluent in a lot of professional workstation applications, but I'm not aware of a single program that can really take advantage of even half of these threads, though I'm open to being corrected about this. This is a server CPU, and it is excessively powerful to almost a stupid level. I think the best use application for a system like this is running dozens of virtual machines, since you could swap those out for a quad core system if you needed to (and therefore are easier to make redundant). The problem, though, is HT is known to perform poorly in some VMs. Otherwise, this CPU is really impractical since it is probably way too costly to be redundant, and redundancy is crucial for servers of this magnitude. I don't know - I'm just thinking out loud. If I wanted a CPU like this, I would prefer a model without HT at a lower price. That's just my opinion though.
I've worked with quite a few programs which can take as many threads as those, or even more. Maybe not the ones we use in desktops. Scientific software I mean. In fact I was already using 36 threads, or 244 sometimes :=) Plus virtualization and servers usually can use as much parallelism as memory bandwitch allows.
data/avatar/default/avatar39.webp
These are brilliant for visualization especially, Couple of these in a hypervisor with 1/4 of a TB of ram, you could technically have 72 VM's running off it. Certainly for business level customers though as I'm sure these won't be cheap. PHP-FPM would be another great candidate to consume those cores at full load. Seen it hog 32 threads at full load (along with another 8 via MySQL)), sure it could handle more!
https://forums.guru3d.com/data/avatars/m/242/242471.jpg
2 of these in a dual socket board for 36 cores and 72 threads!!!! This thing will dominate multithreaded apps!!! But.... 165w TDP!?? WTF???
Yeah, imagine it @ 3-4ghz, 300-400w+ 😯c: