Tuesday, August 14th 2018
Intel X599 Chipset to Drive 28-core HEDT+ Platform
The introduction of 32-core AMD Ryzen Threadripper 2990WX at $1,799 has demolished the competitiveness of the similarly priced Core i9-7980XE, forcing Intel to "productize" its Skylake-X XCC (extreme core-count) silicon for the client-segment. We've already seen one or two motherboards for this platform at Computex, notably the ASUS ROG Dominus (pictured below). Intel's demo platform is reportedly powered by a GIGABYTE-made motherboard. Both these boards may have been prototypes based on Intel C629 "Lewisburg" chipset, as Intel was still mulling on whether to even launch the product.
With the 2990WX out, the fate of the client-segment cousin of the Xeon Platinum 8180 is sealed, and so is that of the C629. In its client-segment avatar, the chipset will be branded "Intel X599 Express." This chipset will support new SKUs derived from the "Skylake-X" XCC silicon (probably 24-core, 26-core, and 28-core), in the LGA3647 package. The platform features not just up to 28 cores, but also a 6-channel DDR4 memory interface, which will probably support up to 192 GB of memory on the client-platform. There's also a rumor that Intel could launch new 20-core and 22-core LGA2066 processors. Those, coupled with the 8-core LGA1151 processor, will be Intel's fig-leaf until late-2019.
Source:
HD Technologia
With the 2990WX out, the fate of the client-segment cousin of the Xeon Platinum 8180 is sealed, and so is that of the C629. In its client-segment avatar, the chipset will be branded "Intel X599 Express." This chipset will support new SKUs derived from the "Skylake-X" XCC silicon (probably 24-core, 26-core, and 28-core), in the LGA3647 package. The platform features not just up to 28 cores, but also a 6-channel DDR4 memory interface, which will probably support up to 192 GB of memory on the client-platform. There's also a rumor that Intel could launch new 20-core and 22-core LGA2066 processors. Those, coupled with the 8-core LGA1151 processor, will be Intel's fig-leaf until late-2019.
68 Comments on Intel X599 Chipset to Drive 28-core HEDT+ Platform
Annoying that AMD have Zen 2 as the third gen release, but thats life
edit: Haha.. nvm. They're long gone.
I'm not sure "duopoly" is the right word for the market either. Isn't AMD still beholden to Intel for some x86 license?
As for the performance scaling across higher core counts, TR wins the day. SMT works a lot better and core/threads on TR scale better than the higher clocks per core on Intel. At the same time, TR can achieve a higher baseclock on all-core loads.
So... no. You're wrong. Jumping to hypotheticals about # of memory channels in the near future won't change that either. Besides, this is not even about the 28/32 core halo products but about the entire product stack and having a cheaper but better performer at every price point. AMD checks that box, Intel does not. Performance is relative. Perf/dollar is what really counts.
so efficiency is out of window, and CPU will be given much more freedom in power limit
We've already seen this issue in both 8C Ryzen and 32C EPYC. High latency of core-core communication results in very bad performance in many typical scenarios.
In a consumer segment Intel's 6C compete with AMD's 8C thanks to higher clocks.
In servers Intel doesn't need frequency advantage. 24C Xeon beats a 32C EPYC with ease - just thanks to a better (albeit more expensive) mesh design.
And why would anyone disable a TR die to get higher clocks? Is this why people buy so many cores? To disable them? Bonkers argument. :-DIt doesn't! I'm not sure if you understand how IF works.
All IF issues you've read about - like when Ryzen gaming performance grows after disabling one CCX or switching SMT off - are multiplied in high core count systems.
And AMD can't fix this. The only thing they can do is give us more and more cores - that will give them an advantage in some tasks (like encoding) and attract some users.You're wrong here as well. If a CPU is faster, it is faster. End of story.
If Intel's CPU offers higher performance, people will keep buying it. It doesn't matter that AMD offers 80% for 60% of price. Yes, it wins in "perf/price", but that shrinks drastically when you look at the cost of the whole machine. And in the end you have a single CPU in your consumer PC.
People have exactly the same problem understanding how servers work. Yes, "perf/price" could work in huge datacenters. But smaller workstations/servers are limited to 1, 2 or 4 CPUs. It doesn't matter that you can buy 5 EPYCs instead of 4 XEONs, because you'll end up with a slower 4S server and a very expensive drink coaster. :-)
Or indeed "it can be faster because it cuts corners leading to security holes"?
I know this is a bit O/T, but I think you get the point.
Really what you see in reviews now is similar to many things we saw with new core count procs on both msdt and hedt. Lack of optimization and support. I contest that Intel does it faster, I rhink they are mostly still riding on clockspeed advantage but all the rest isnt up to what Zen is growing into. This doesnt hapen overnight but its clear AMD is pushing in that direction. This is why they take drastic measures to increase market share.
Its really hard to state Intel Core is better for the HEDT environment these days. Its the same thing you see in msdt: Intel leads by such a tiny margin it becomes irrelevant for most except some tiny niche.