Bozo
0
I was doing some reading on the newest computer hardware. The Intel Core i9 can have up to 18 cores and 36 threads. Doesn't the programs you are using have to be able to use all those cores? If so, what's the point of all those cores?
Core I9
- Thread starter Bozo
- Start date
Programs don't have to use all the cores. Some software doesn't need more than one core to do the work it needs. Other software can be written to schedule more threads and to perform work in parallel. Developing software to make use of multiple cores efficiently can be challenging and there can be a limit to the return of having so many cores.
In recent times there are more people making use of software containers (Docker for example) and virtualized environments (VMware, KVM, Xen, etc) to leverage the numerous cores to do more work in "isolated" environments which makes use of more cores. It is nice that there are more cores being offered but this really only became a thing once AMD lit a fire under Intel's ass with their Threadripper offering.
In recent times there are more people making use of software containers (Docker for example) and virtualized environments (VMware, KVM, Xen, etc) to leverage the numerous cores to do more work in "isolated" environments which makes use of more cores. It is nice that there are more cores being offered but this really only became a thing once AMD lit a fire under Intel's ass with their Threadripper offering.
sechs
0
The programs only have to be able to use any of those cores. You could run 36 single-threaded programs, and each could run on its own virtual processor core.
This is not to mention the number of programs that can spawn multiple threads that run in parallel.
timwhit
0
Every desktop user that uses a modern browser is using multiple cores. The OS also uses multiple cores.
Maybe so, but they are not exactly pegged at 100%.
Gödel
Storage is especially nice if it doesn't rotate
But wouldn't these processes each be "residing" on one core, and avoiding being swapped out and back in to memory, as they would be on a processor with fewer cores?
They may not be residing on any one specific core because it will be up to the operating system to schedule CPU time which can fluctuate base on demand. There is an exception if the user has set a process affinity to a core/CPU that can make the process stick but this is typically a manual configuration.
Paging in and out of memory is not necessarily a function of switching cores within a single socket that I know of. My understanding is that the core sits in front of some cache (L1, L2, L3) and connects to a memory bus/controller for the system. Cores can perform read/write to regions of memory but I don't believe the act of scheduling from core 1 to core 2 specifically requires a call to page memory out and back in. If you're talking about a multi-socket Intel system then you could run into performance concerns with scheduling time for a process on cores between different sockets. This may spend time using the QuickPath Interconnect (QPI) links to communicate between sockets.
time
0
Hi Bozo
Short answer is YES and NONE, for most desktops.
Although Timwhit's point that browsers use multiple threads is technically true (and the OS can run multiple processes across different cores), it doesn't lead to better performance once you have enough cores to avoid saturation of a single core. Queuing theory shows that by far the biggest improvement is gained when comparing a single core with a dual core. There are raw processing tasks (eg 7z archiving) that scale with more cores, but this soon peters out due to network or storage performance constraints. Considering a browser is limited by the speed of internet requests (and the web server), 18 cores seems a trifle excessive ...
Short answer is YES and NONE, for most desktops.
Although Timwhit's point that browsers use multiple threads is technically true (and the OS can run multiple processes across different cores), it doesn't lead to better performance once you have enough cores to avoid saturation of a single core. Queuing theory shows that by far the biggest improvement is gained when comparing a single core with a dual core. There are raw processing tasks (eg 7z archiving) that scale with more cores, but this soon peters out due to network or storage performance constraints. Considering a browser is limited by the speed of internet requests (and the web server), 18 cores seems a trifle excessive ...
Stereodude
Not really a
With current software for the typical user there's very little benefit beyond 4 cores. My main desktop has 8 cores / 16 logical threads and it's only well utilized with video compression and when mass converting audio files.
Hi Bozo
Short answer is YES and NONE, for most desktops.
Although Timwhit's point that browsers use multiple threads is technically true (and the OS can run multiple processes across different cores), it doesn't lead to better performance once you have enough cores to avoid saturation of a single core. Queuing theory shows that by far the biggest improvement is gained when comparing a single core with a dual core. There are raw processing tasks (eg 7z archiving) that scale with more cores, but this soon peters out due to network or storage performance constraints. Considering a browser is limited by the speed of internet requests (and the web server), 18 cores seems a trifle excessive ...
I would argue that it's No, and Marketing. Programs don't need to use all the cores however they can be designed to scale with the opportunity when more are available.
You haven't seen the number of tabs I keep open in my browser on a regular basis. :-D
timwhit
0
Modern websites use way more CPU than you probably realize. If you keep multiple tabs open or load a lot of tabs at the same time you could easily peg 4 cores. Run the Javascript profiler on a complex website sometime.
I have nothing but a browser and some background services running right now. All 12 cores are 5-10% used.
Sure 6C/12T will do that, but is it really faster than 4C/8CT at 15% utilization? I'm thinking that maybe there is some benefit for an instant when a new page loads or something.
Lately I'm seeing web pages with a message that there is a slowly moving script running, but the CPU cores are little used.
Bozo
0
Seems like multiple cores would be great for virtual machines. If you can assign the cores to each virtual machine. In VMWare, you can tell the virtual machine how many cores it can use, but does this lock those cores from being used by something else?
Stereodude
Not really a
If you mean the host OS, no.
Chewy509
Wotty wot wot.
I'm just trying to work out the target market for the i9?
Most people who need the cores (eg CAD, etc) will also want ECC RAM (which the i9 lacks support for) hence would be looking at Xeon's and most likely nVidia Quadros or AMD Raedon Pro? Additionally, the X299 chipset only supports 24 PCIe lanes, so without looking at workstation chipsets, any system with the X299 chipset will be gimped in some manner? (most i9s support 44 PCIe lane setups)
Game engines are having troubling scaling past 4 cores, so it's not gaming. Software developers who need the cores/threads would be looking at the most stable platforms (eg Xeons w/ECC RAM)... And as mentioned, CAD users will already be on Xeon's with ECC RAM... And if it's for scientific use, then a lot of users will demand ECC RAM to ensure a single bit flip doesn't taint any data in their research...
Or is the i9 simply a "who has the biggest ... " contest for marketing?
Most people who need the cores (eg CAD, etc) will also want ECC RAM (which the i9 lacks support for) hence would be looking at Xeon's and most likely nVidia Quadros or AMD Raedon Pro? Additionally, the X299 chipset only supports 24 PCIe lanes, so without looking at workstation chipsets, any system with the X299 chipset will be gimped in some manner? (most i9s support 44 PCIe lane setups)
Game engines are having troubling scaling past 4 cores, so it's not gaming. Software developers who need the cores/threads would be looking at the most stable platforms (eg Xeons w/ECC RAM)... And as mentioned, CAD users will already be on Xeon's with ECC RAM... And if it's for scientific use, then a lot of users will demand ECC RAM to ensure a single bit flip doesn't taint any data in their research...
Or is the i9 simply a "who has the biggest ... " contest for marketing?
Tea
0
Usage of multiple cores:
#1: Microsoft's spyware sending data to Microsoft
#2: Google's spyware sending data to Google
#3: Facebook's spyware sending data to Facebook
#4: Your anti-virus spyware sending data to Russia.
#5: The Fed's spyware sending data to Uncle Sam
#6: Your application software.
Clearly, 6 cores is all you need.
#1: Microsoft's spyware sending data to Microsoft
#2: Google's spyware sending data to Google
#3: Facebook's spyware sending data to Facebook
#4: Your anti-virus spyware sending data to Russia.
#5: The Fed's spyware sending data to Uncle Sam
#6: Your application software.
Clearly, 6 cores is all you need.
Stereodude
Not really a
It's simply Intel's response to AMD's high core count CPUs. Even the typical power user can't take advantage of either, but Intel already had the CPUs being sold as Xeons, so they disable ECC make an i9 SKU and sell them for big money to the high end enthusiast market. More money for Intel with virtually no additional effort and they don't leave that niche for AMD to exploit alone.
Stereodude
Not really a
For what? I have a feeling your "need" doesn't actually exist. You're not running a 3-way or 4-way SLI setup, so...I just need more PCIe lanes not all those cores.
No, but the current computer has a video card, RAID card, M.2 SSD, and 10GbE card. Does that work with a measly 16 PCIe lanes?
Stereodude
Not really a
Of course it will work. However, I don't think you understand what you're talking about. The CPU has PCIe lanes and so does the chipset. For example, the Z170 chipset has 20 lanes and the CPU has 16 lanes. So the graphics card connects to the CPU's 16 lanes. That leaves 20 from the chipset for everything else. How does that not work, even with the peripherals you listed?
Further, are you going to notice if your graphics card only gets 4 lanes, your M.2 SSD only gets 2 lanes?
I was looking at the Coffee Lake.
Expansion Options
Scalability 1S Only
PCI Express Revision 3.0
PCI Express Configurations ‡ Up to 1x16, 2x8, 1x8+2x4
Max # of PCI Express Lanes 16
Stereodude
Not really a
And the Z370 chipset for Coffee Lake has 24 more lanes. So there are 40 total to work with in a Coffee Lake system. The 16 from the CPU will go to the graphics card. The 24 from the chipset will support the rest of your crap.
Again, you're creating a problem where there isn't one.
Interesting I wasn't aware that's how this was implemented. The additional 24 lanes on the X370 interface with the Coffee Lake CPU using DMI 3.0. So, there may still be bandwidth issues given that the DMI 3.0 interface (if I'm reading it correctly) allows for 3.93 GB/s for the CPU–PCH link. A PCIe 3.0 x4 connection has the bandwidth of 3.94GB/s means that the chipset is over-subscribed quite a bit if one were to make use of all the 24 lanes for IO cards and NVMe storage not to mention whatever additional latency is introduced.
Stereodude
Not really a
Yeah, but what actual use case are you going to push more data than that simultaneously? Even a 10GbE copy to or from a NVMe x4 drive isn't going to hit that (presuming it actually all has to go through the CPU and can't be handled DMA style directly by the chipset).
My experience with older chipsets like the 4790K board is that certain combinations just don't work and one slot is non-functional when the board is overloaded.
For example SSD on M.2 and SSD in PCIe, GbE card and video card will not work.
sechs
0
Fools with money.
sechs
0
Threadripper 1900x. "Just" eight cores/16 threads, but boatloads of PCIe lanes.I just need more PCIe lanes not all those cores.
I thought the per thread performance was not so good and do they OC to 5.1 or 5.5MHz like the Café Lake Dave was using?
sechs
0
I think that you mean IPC, and, no, it's not as good as Coffee Lake. My feeling is that most people wouldn't notice the difference.
On the other hand, you might notice 60 PCIe lanes from the processor, and the money still in your pocket.
Stereodude
Not really a
He's not going to notice that either.
I would not notice a small difference in price, but more lanes are better. I don't understand why Intel has only 16.
It's like comparing one car with 160 PS and another with 600. I don't need 600, but 160 can be limiting in a traffic jam merger.
Chewy509
Wotty wot wot.
I was under the impression that IPC for Zen based cores was equivalent to Haswell on the Intel side of things? And from what I've read there isn't a lot of difference between Haswell and current Intel cores...
https://www.sweclockers.com/test/24701-intel-core-i9-7980xe/19#content
Basically, all CPU's clocked at 2.8GHz and run the Cinebench R15 single core test... Ryzen gets' the same score as the i7-4790, meaning it's IPC is very similar to that CPU...
https://www.sweclockers.com/test/24701-intel-core-i9-7980xe/19#content
Basically, all CPU's clocked at 2.8GHz and run the Cinebench R15 single core test... Ryzen gets' the same score as the i7-4790, meaning it's IPC is very similar to that CPU...
Stereodude
Not really a
Because that's all they put on the CPUs. You still can't seem to put together any sort of specific hypothetical use case where it's going to actually pose a noticeable limit to yourself though. Or even, how about some benchmarks showing where it's a real limit vs. the Intel CPUs with more lanes?
Stereodude
Not really a
I guess the moral to the story is that if you basically tread water in your product development in terms of IPC performance for almost a decade even AMD can eventually catch up.