Slashdot Mirror
Intel Releases Final Core i9 Specs and Release Dates -- And Threadripper Is Faster (Sometimes) (pcworld.com)
On Monday, Intel took the wraps of final details of its Core i9 microprocessors. From a report: Remember that Intel's Core X-series family (also called the Core i9) was announced with several key omissions: namely the clock speeds of the 12-core Core i9-7920X and above, as well as the thermal design power, or TDP. On Monday, Intel filled those in. The 12-core Core i9-7920X launches Aug. 28 while the 14-, 16-, and 18-core Core i9 chips ship on Sept. 25. Perhaps most important, though, is that we now know how fast Intel's Core i9s will run. When Intel inadvertently revealed that its 12-core Core i9-7920X was 2.9-GHz -- slower than the comparable AMD Threadripper -- a subset of the internet had a small freakout. We now know that that will be true for the remaining Core i9s as well, but with a big caveat. Here are the remaining speeds and feeds for the high-end Core i9 chips:
Core i9-7980XE (18 cores, 36 threads): 2.6GHz; Boost, 4.2GHz to 4.4 GHz.
Core i9-7960X (16 cores, 32 threads): 2.8GHz; Boost, 4.2GHz to 4.4 GHz.
Core i9-7940X 14 cores, 28 threads: 3.1GHz; Boost: 4.3GHz to 4.4GHz.
Core i9-7920X (12 cores, 24 threads): 2.9-GHz; Boost: 4.3-GHz to 4.4GHz.
Note that the boost speeds refer to both Intel's Turbo Boost Technology 2.0 and 3.0. [...] Essentially, both Intel and AMD can claim the title of fastest processor. Threadripper's base clock speeds are faster, but Intel's boost speeds climb higher than Threadripper can. It's also important to note that while Threadripper consumes 180 watts, even the fastest Core i9 chips Intel has announced have a lower TDP of 165 watts.
Core i9-7980XE (18 cores, 36 threads): 2.6GHz; Boost, 4.2GHz to 4.4 GHz.
Core i9-7960X (16 cores, 32 threads): 2.8GHz; Boost, 4.2GHz to 4.4 GHz.
Core i9-7940X 14 cores, 28 threads: 3.1GHz; Boost: 4.3GHz to 4.4GHz.
Core i9-7920X (12 cores, 24 threads): 2.9-GHz; Boost: 4.3-GHz to 4.4GHz.
Note that the boost speeds refer to both Intel's Turbo Boost Technology 2.0 and 3.0. [...] Essentially, both Intel and AMD can claim the title of fastest processor. Threadripper's base clock speeds are faster, but Intel's boost speeds climb higher than Threadripper can. It's also important to note that while Threadripper consumes 180 watts, even the fastest Core i9 chips Intel has announced have a lower TDP of 165 watts.
... on slashdot no less. How about some real workload comparisons?
That market segmentation is like a parasite to their thinking. They can't do anything without chopping the product line into as small as bites as they're able.
Intel compilers are used for a lot of backend libraries (eg. parts of Matlab), and tens to turn off some SSE extensions even on AMD processors that support them. Lack of vectorization leads to worse performance, 'nuff said.
"It's also important to note that while Threadripper consumes 180 watts, even the fastest Core i9 chips Intel has announced have a lower TDP of 165 watts."
The actual power draw of even the 10-core i9 is >200W. Intel are deceiving us yet again.
We're talking about CPUs with 24 or more threads of execution here. The bottleneck isn't hardware, it's the software. Most software just doesn't support that high degree of threading. That's why we need to start using programming languages built around concurrency, like Rust and Erlang. Rust is a great example. Look at the innovation we're seeing from Mozilla with their new Servo web browser engine. It is being designed from the ground up using Rust so that it can make full use of CPUs with many simultaneous threads of execution. We need to leave uni-threaded languages like C behind, even if they've tried to naively add half-baked threading support (like pthreads). We need to move to languages like Rust where the language itself is built with concurrency and thread safety as core features.
The 16-core i9-7960X is ~$1700, while the 1950X Threadripper is $999. And the Threadripper X models are the higher-clocked/premium version. If AMD has a non-X model in the works, the pricing difference could be even greater (AMD didn't reveal the X/non-X version of the R3 until way later, so it's possible).
Depends on longevity in the market.
Back in the Netburst days, you marvelled at how Intel did so poorly despite 'looking' like it should be faster.
From Conroe to about Bulldozer, things were about the same.
The bulldozer screwed up AMD in the same way that Netburst screwed up Intel for a while. Meanwhile Intel progressed well.
Now with Zen, at least on desktop it's back to mostly neck and neck. In high end server, it's a mixed bag, Epyc having more memory channels means better capacity, but individual memory performance is equal to their desktop product. This is fine for a lot of applications (e.g. VDI, similar virtualization) and gets more aggregate performance and capacity, though single thread/process memory throughput takes a hit.
XML is like violence. If it doesn't solve the problem, use more.
Threadripper Is half the cost with more pci-e lanes.
On Intel it's about $1000 min to get 48 pci-e lanes. AMD $550 for there 64 pci-e lane one.
How can you justify this many cores at that temperature? Sure if you are running a single gaming rig in your home you can water cool your computer but if you are running anything else I can imagine the cooling problems for a room of these things will be astounding.. I am just not sure of the applications in servers or regular business or home units at this wattage.
For the same reason that carriers don't upgrade infrastructure unless it literally crashes, why companies like HP got rid of expensive engineers, and why small oil companies don't dig lots of exploration bores. You don't need 'real' growth if quarterly profits look good on paper because you are slashing costs, and everyone at the top of the pile is going to be gone in 3 years when the old wells run dry.
Physics is nothing like religion. If it was, we'd have an easier time trying to raise money!
See subject - The OS process scheduler itself can use the extra cores for BIG overall gains ala e.g. in Windows:
Increase cpu core count @ hardware level
Windows Registry Editor Version 5.00
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Session Manager\Executive]
"AdditionalCriticalWorkerThreads"=dword:00000008
"AdditionalDelayedWorkerThreads"=dword:00000008
* I.E. - How much extra cores will help BEYOND today's CPUs for the OPERATING SYSTEM itself (in Critical Worker Threads) in juggling threads in itself & for other processes (in Delayed Worker Threads) per https://msdn.microsoft.com/en-us/library/cc615012(v=bts.10).aspx/
Here I use 8 for an Intel Core I7 as shown above (both in 1st a 920 & currently a 4790k, since they're quad core (& hyperthreaded) & it was lesser based on physical core count of earlier systems I had (this setting has been around since, iirc, Win2k (correct me IF I am off/wrong - it's been SO long since then)...
(Those are settings in WINDOWS you can adjust to take advantage of added cores as you upgrade to CPUs w/ more cores, for example).
ANYTHING/EVERYTHING, in theory, gains there alone (less "process scheduler thrashing" in other words) - I don't care so much about applications/programs (they are probably written to their practical limits anyhow as to what threadwork will gain them) but again, MORE about how the OS will utilize them (per the 2 TUNABLE PARAMETERS in the .reg file I note above as a way to REALLY use the extra cores, almost guaranteed - Windows allows it, not sure of other OS like *NIX based ones).
APK
P.S.=> The rest will be done @ compiler level (already good, only depends on HOW you can leverage it OR if internal-to-program itself datasets AND PROCESSES (imo, a Gannt chart illustrates this well) allow for it - not all do) & it's always that way, pretty much - hardware 1st, software catches up (& it does, mostly inefficiently @ 1st, sucking up the CPU cycles/efficiencies gained)... apk
I think I already could have spent 1000+ USD a couple of years ago on a xeon to get more than 4 cores. But at the normal 200-300 USD price point, there is nothing to upgrade to.
I will freely admit that sticking it to the man is on the list of considerations for me. Only as long as price/performance makes sense, though.
Threadripper officially launches on 8/10. Expect reviews to hit the internet first thing that morning as most hardware sites have had a week or two to play with their samples. Intel is trying desperately to make some last minute noise I suppose. Makes me suspect the official reviews of TR will be quite positive. Intel is going to find that having bigger numbers (Mhz, core count), is going to count for little when the other guy gets you 90% or more of the performance for half the price.
Cool! CNN's terrible advertising scripting can lock up Chrome even faster now!
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
Threadripper Is half the cost with more pci-e lanes.
On Intel it's about $1000 min to get 48 pci-e lanes. AMD $550 for there 64 pci-e lane one.
In addition to the base price, what matters here (for the datacenter market) is how much computing can be done per Watt with your average server application. We need a third-party review for that because AMD exaggerates while Intel lies it's ass off about everything.
Anons need not reply. Questions end with a question mark.
Epyc P series , the 1P processor models have 128 lanes. Tyan is releasing a board that has 124 lanes available. which would be great for san/nas or gpgpu build.
The Zen architecture IPC is pretty much on par with the Core one as long you're not using the intel AVX extensions.
It's pretty close depending on application and usage not really.
AMD also has usb 3.1 on die.
Intels 24 pci-e on chipset is also stacked with sata / usb / MB IO / sound pci-e / network link and do lower end cpus only having as low as 16 pci-e lanes. Some boards also wire storage M.2 slots to the chipset as well.
Point is, we will know when someone runs real world tests across a full suite of applications and benchmarks.