Re: 36 cores at 4.2 GHz?
According to this table (from PCGamer) it can manage 3.4GHz with 18 cores, 42.GHz with two cores.
Intel's offered some more detail about the Core i9 range of desktop CPUs it announced in May. Here's what Chipzilla has planned for us all when these chips start to go on sale. The 12-core products will appear as of August 28th. 14-to-18-core kit will go on sale as of September 25th. Processor Name Core i9- 7980XE X- Core …
You could run WinNT quite nicely on that, although you'd need to tweak a BIOS setting to disable hyperthreading or else the number of cores on the top-end part would be too large.
I don't know if you could run Win95 on it. Do these things still have a mode where they can run 16-bit instructions?
Or is the 10 core budget version the highest performer?
Highest base clock, highest boosted clock, lower rated power useage.
For everything but the most obscure workloads individual core performance is likely to trump the number of cores once you get above, say, 8 (possibly 4 or less) especially with two threads per core.
The only answer that anybody could give you and be right is: It depends.
Heavy Multithreaded CPU workloads that aren't being palmed off to the GPU will definitely benefit from the extra cores at a lower frequency. Also there will probably be similar potential for overclocking across the chips so you'd probably be able to get any of these chips screaming along at 5GHz+ with watercooling.
If you're gaming, most games will bottleneck on the graphics first even down at the mid i5 range which reach similar frequencies anyway. Bringing streaming into the mix, more cores are handy as it means that any encoding and CPU manged network activity isn't using the same core(s) you're gaming on.
Video editing is dependent upon your setup but there will probably be some part of the workflow which is CPU intensive.
VR gaming could probably use more cores due to the number crunching required to prevent motion sickness but is also highly GPU dependent.
so yeah. It depends.
Xeons are slightly different beasts. This price range of ~$2k gives you 14 cores with a base speed of 2GHz, turboing up to 2.8GHz and Xeons don't overclock in the same way as the Core processors. You'll also get all sorts of datacentre gubbins and probably improved warranties etc.
For a Streamer, the i9 processors are a better deal.
The short version (and I'm trying to get this comment in before the Intel and AMD fanbois start fighting) is that AMD hasn't been making competitive CPUs for a few years, until now.
They've been making cheap CPUs, but Intel are still making the fastest. With their new Ryzen (silly name) architecture, it seems like AMD are finally at least in the same race as Intel, so I suspect you'll start to see them being used by more OEMs. AMD are cheap because otherwise nobody would buy them, they've been losing money to stay in the game.
Unless you're looking at the high end, the AMD chip will probably be better value for you.
AMD has better value CPUs if you don't need the absolutely fastest available. It has had this for a long time now. With Ryzen they may actually now compete, or beat, Intel in the top performance level too.
Cheap motherboards for AMD are easier to find, and AMD traditionally has had good upgrade paths for faster CPUs on older motherboards (i.e sockets). Meaning often RAM and Mobo investments can be kept for longer.
Sadly Ryzen isn't available for AM3+ sockets, so there is a definite break with the previous generation AMD CPUs. (AM3+ has had a good run though).
I have run AMD in all my PCs for the last 18 years, so someone may want to add Intel info and correct me on the value aspect..
P.S: There was a debacle about Intel's compilers fixing the binaries to run much faster on Intel CPUs, in effect making benchmark software (as well as actual applications) favour Intel. IRL AMDs are quite fast.
P.P.S: "Is there something about AMD i am missing - and why don't vendors use AMD more ?"
There is a lot of business decision making going on, with lock-ins, Intel leveraging it's size, sales trickery, and so on. Comparable to MS vs the rest.
P.P.P.S: The value of having at least one other player competing with Intel is immense. That's one reason II never abandoned AMD.
Because Intel gave out rebates to those that used Intel only. At one point Intel was giving Dell close to a billion dollars a year i rebates. When AMD offered to give HP 1 million free CPUs HP turned it down because the amount of money they would lose from intel was to great. Things have started to change in the last 5 years though.
Until recently AMD haven't been comparable except at the low end - their APU offerings are ok because the standard of bundled GPU is better (for desktops) than the Intel alternative. They also do some interesting embedded options.
With Ryzen, at the high end they're not quite as good as Intel, but a lot cheaper. If you don't need the absolute fastest single threaded performance they're a decent deal.
They haven't kept up with virtualisation enhancements like Intel, though, aside from the new encrypted memory options, which is a pity at the server end.
For a low end box, I'd have no issue using AMD. For a reasonably high end desktop that's mostly concerned with running lots of processes, but also needs to be quite fast, I'd also consider AMD. For an all out gaming box I'd go Intel, and for virtualisation I'd look at a Xeon.
For an embedded firewall I'm looking at an Alix APU2. AMD Jaguar core, fanless, decent encryption support on chip.
The "high end" of this HEDT thing is all about bragging rights: Intel scrambled to release an 18 core part because they weren't about to have AMD deliver "moar coars" .
So now they have problems getting the heat out of this thing and the AMD part has an advantage there because its "heat generating area" is larger and distributed (because it has 2 or 4 widely spaced die under a bigger slug of copper).
So I guess we're about to find out if Intel's "process advantage" is real ... right ??
Intel's i9 is a solution in search of a problem. As a gamer who builds his own boxes, unfortunately, I don't even remotely see how i9 helps gamers.
There are basically two big bottlenecks to gamers today: PCIe lanes (aka more graphics cards and m.2 SSDs please!) and memory bandwidth.
Throwing more cores at the problem is, at best, just making things worse. It lowers the top-end GHz. (To date that still matters a lot. It's why we OC. Duh!) Gamers need the opposite of the i9: less cores on a larger die with a better thermal interface and lower voltages so that they can push the need for speed with a chip that OCs well.
If Intel really wants to help gamers, they need to ditch the more cores = better concept and get back to basics: Faster is better, bandwidth is your bottleneck, and cooling is king. It's a recipe as easy as π.
18 cores. Miam.
Got a 4 core i7-6700 since 2015. I also slapped in 32GB of DDR4-3200.
Can I justify upgrading ? Not really. Doesn't matter. I want one of these babies. I'll get one in 2019 probably. With 64GB of DDR5 (by then).
I'll be able to push 7 Days to the full 3840 x 2160 of my widescreen. Finally.
Of course, by then another game will come out that will put my rig to its knees. As usual.
To be honest the bottleneck in most PC's is the GPU not the CPU, I've been running a Hasswell 5820K 6 cores clocked to 4.5 per core for a while now and I have tried 3 different GPU's GTX980 , TITAN X, GTX1080TI on a ASUS IPS Gysnc 144htz 27" .
Each card is happy to run everything at 1080p at 144htz , at QHD 2560 X 1440 the 980 wont hit 144htz in modern FPS's with all the goodies but the other two are happy , and at 4K UHD the TITAN X drops sub 144 htz every now and then but the 1080 seems flawless unless you start plugging in second monitors.
4K play back/encoding the 980 is utter shit , the TITAN X is ok but the 1080 wins hands down (and the cpu or how many cores makes no difference).
Spend your cash on a good monitor and GPU , and spend what is left on a mediocre CPU ...works a treat.
Simple rules for gamers
1. always spend more on your monitor and gpu than anything else.
2. you only need SLI if you have more than one monitor or you are trying to get 2 crappy cards working which is usually pointless.
3. you don't need a 60 core x 100 ghz to run Dota2/LOL/WOW/CSGO.
4. V.R ready does require big blue or quantum computing for a 10min on rails zombie train shooter
Paying a higher price for a CPU to get the highest possible performance - particularly when the cost of the rest of the system reduces the percentage extra one is paying for higher performance - is not irrational. Which is part of why Intel can get away with its current pricing.
My use case is cross platform and embedded development, my daily driver for this is a dual core Xeon E5-2697v2 (24C/48T) from the Ivy Bridge era. If you have thousands of source files to compile for multiple targets, or for the edit-compile-debug loop, it makes it a relative breeze. Going to the more mainstream i7-7700K or even Ryzen 7 1800x is really quite a disappointment for productivity (relatively speaking, of course!)
I strongly suggest you look at ccache. If it is an option (it should) you may cut compile time by a factor of 100.
In case you cannot use ccache and have some money to splash, Naples has 32 cores (64 threads) per socket. Dual socket systems should be available, that's 128 threads for you.
To put it simply, a thread is a single line of commands that are getting processed, each application has at least one thread, most have multiples. A core is the physical hardware that works on the thread. In general a processor can only work on one thread per core, CPUs with hyper threading can work on up to two threads per core. For processors with hyper threading, there are extra registers and execution units in the core so it can store the state of two threads and work on them both.
"For processors with hyper threading, there are extra registers and execution units in the core so it can store the state of two threads and work on them both."
Are there though? Pretty sure the horrible inefficiency of logical cores I see on most systems (especially craptops) comes down to that NOT being the case. It's just trying to execute two threads in the same compute resource and, frankly, there just ain't enough to go around. It's why in my software I tend to limit execution by physical cores. Logical cores are only good for background processes and services. ;)
(Which, in today's OSes at least, is a useful thing to have. But only helps in that it frees up the junk processes to run in their own hell of ineptitude so that everything else has real compute power to run on.)
Yes -- the OP is incorrect ; execution resources are shared and not particularly increased
for the second thread. The reg files are doubled up, but it is about utilising at higher fraction
of a fixed superscalar resource.
However, most cores on most code are idle most of the time. Waiting on memory. The sharing of execution resources isn't really why most people see "hype threads" as not working so well.
The more fundamental problem is the sharing of the L1 cache and the bus to L2. L1
thrashing in particular can be as painful to watch as two keystone cops vying for the same door
and neither making it through.
I use Macs & Final Cut Pro X for a living, & the loads appear to be distributed between GPU & CPU (couldn't tell you *exactly* what is doing what) from seeing the performance of a variety of machines that I use...so yes, a big fast new chip (I assume this is whats going in the forthcoming iMac 'Pro') would be very much welcome...also in the audio processing side of things its 100% CPU load
Most people seem to recognise that more cores/threads is not important above a certain number. The issue, as always, is getting enough data into and out of the CPU.
I would have thought that the deficit that Intel have regarding PCIe lanes is far more important to the high end people than actual core count. Ryzen will have 128 PCIe lanes and Intel are still stuck with 44, that amounts to a potentially huge data throughput deficit for Intel i9 processors and there is no indication from Intel that this will change.
The caveat is wait for real systems to turn up and crunch the numbers on, but on a system wide basis, it's looking far better for AMD than it has for years. Of course the system builders and MoBo manufacturers have to take advantage of those extra 84 PCIe lanes for Ryzen to really shine.
"If you are willing to pay that amount of money ... is it not better to go for a Xenon CPU ?"
If the year falls before the year < 2022, yes. Technically, these are the old/current gen Xeons rebadged for desktop, I see nothing different than the exact 14c I have besides you aren't guaranteed anything with these i9's (xan. they even run in parallel?). I've been running 2 OK'ish 14 core ES chips for nearly 2 years, cost $300usd total for both. $720 for 128GB Hynix and $450 for a supermicro mobo. Still less than 1 of these CPU's.
I think if I cared about gaming and streaming and all that, I clearly would buy AMD (actually I might stitch these CPU's and buy AMD anyway).
There are fundamental limits on what may be achieved with multiple CPU architecture.
It may not mean anything soon.
Less than a month since the above link we have this:
Biting the hand that feeds IT © 1998–2019