Thursday, September 5th 2019
Intel to Increase Cores-to-the-Dollar Across the Board with Cascade Lake-X?
Intel is preparing to increase core-counts across the board with its upcoming Core X "Cascade Lake-X" HEDT processor family, launching next month. The first indication of this comes from an Intel slide that claims a 1.74 to 2.09x increase in performance-per-Dollar over "Skylake-X." The "Cascade Lake" microarchitecture already made its debut in the enterprise market as Intel's 2nd generation Xeon Scalable processors, and its IPC is similar clock-to-clock, to its predecessor (Skylake).
If Intel is claiming such performance-per-Dollar increases, it only points to a significant increase in core counts to the Dollar (think 16-core at $999, 28-core at $1999, etc.). Adding value to these chips are certain new AI accelerating instruction sets, such as DLBoost, support for Optane DC Persistent Memory, increased memory clock-speeds, and higher CPU clocks across the board compared to the Core X 9000-series. The Core X "Cascade Lake-X" processor family debuts this October.
Source:
Tom's Hardware
If Intel is claiming such performance-per-Dollar increases, it only points to a significant increase in core counts to the Dollar (think 16-core at $999, 28-core at $1999, etc.). Adding value to these chips are certain new AI accelerating instruction sets, such as DLBoost, support for Optane DC Persistent Memory, increased memory clock-speeds, and higher CPU clocks across the board compared to the Core X 9000-series. The Core X "Cascade Lake-X" processor family debuts this October.
50 Comments on Intel to Increase Cores-to-the-Dollar Across the Board with Cascade Lake-X?
Let me fix that for you: "intel still holds the 1% gaming market and AVX 512 workloads for people with more money the brains". For the majority of production software and productivity applications, AMD has surpassed intel, in gaming they are nipping at the heels of the 9700k, and in the server market they are obliterating intel's performance per $ and absolute performance results; For HDET usres threadripper 2 is going to be a godsend, much as threadripper 1 was.
Intel is on the defensive, and is refusing to admit AMD is hammering them at every turn. It's Pentium 4 VS athlon 64 all over again.
- AVX512 is generally used by people with both brains and money. In most cases where money is not that big of a problem.
- Intel still holds enough of the gaming market. 9900K is a stupid desktop flagship thing. 9700K is above anything AMD can put out for gaming and it is now priced between 3700X and 3800X. 9600K is priced between 3600 and 3600X and does OK. 9400F is probably the biggest spoiler of the gaming market party for AMD.
- AMD still does not have good presence in mobile and Intel is now working with a limited set of 10nm CPUs there.
From this news bit it looks like Intel will bring Xeon prices down considerably. This may change some of these performance/$ calculations. It will not bring Intel back to markets where cores matter like cloud where compute power is sold per core. There are places where Intel was pretty competitive. SQL Servers and some compute aspects come to mind. There seems to be little public testing on how good Rome does in these areas.
Edit:
I mean, Intel is not stupid. They are playing to their strengths. Mobile is still Intel's by a large margin. Their problem on the desktop is primarily lack of HT that kills productivity performance but perhaps more importantly looks bad in reviews. I am willing to bet that their next generation will bring HT back. HDET is lost to Threadrippers. Xeons are so-so and Intel can play with prices until they get something new and worthwhile.
-9700k is a whopping 5-6 FPS faster then a 3700x in average FPS. Congratulations I guess? Both are well north of 60 FPS, hell both are ususally north of 90-100 FPS, and when they are not the two are usually quite close. intel can hardly justify their prices, hence why they have been cratering down to AMD pricing. 3600 is comparable to 9600kf in gaming, 9400f isnt spoiling a lot there, and the r5 3500 will fix that.
-Yes, AMD has little mobile presence, and likely wont until next gen ryzen mobile 7nm chips exist. That doesnt mean AMD is not a threat here.
If this price change will not breing intel back to the markets it lost, that seems like a pretty short sighted move, no? They are sacrificing profit margins for 0 gain, only to prevent AMD's further encroachmet into their position. That is the same strategy AMD tried with bulldozer, and it backfired spectacularly. SQL is sold per core (socket) for enterprice, and here AMD's rome is vastly superior to intel's XEON, until intel offers 64 core xeons for rome pricing. The cost saving rome represents is jaw dropping, and a price cut across the board from intel isnt going to come close to closing that gap.
Also, AVX 512 really is amazing. Or at least it has amazingly poor scaling most of the time :
lemire.me/blog/2018/08/13/the-dangers-of-avx-512-throttling-myth-or-reality/
blog.cloudflare.com/on-the-dangers-of-intels-frequency-scaling/
Oh and Rome is equal or better most of the time with just AVX2 vs dual socket 28 core Xeons that use AVX 512 in the only market where this is somewhat more relevant. Though this is questionable, a lot of people recognize the limitations of AVX 512 and at this point in time, with the ever increasing number of cores, you'd be shooting yourself in the foot driving these things even more into a power wall with very wide SIMD.
Let's stick to what the original 7nm ryzen mobile comment meant: Ryzen mobile, as of right now, isnt competitive in pure power efficiency. It does offer better GPU performance. When the 7nm shrink comes (along with the move to zen 2), much of the remaining gap with intel will be closed, at which point ryzen mobile will likely begin to take some marketshare. This is another market Intel is confident they will not lose, despite that same argument applied to desktop CPUs 3 years ago. Intel's 10nm rollout is still not in full swing, and we have yet to see their full pwoer 15watt parts on 10nm, suggesting they are still having yeild issues. If Ryzen 7nm comes out before 10nm is available across the board for intel, intel will be at a serious disadvantage.
9400F is 30% cheaper than 3600. Uses less power as well. Did not mean that. The point was, that time is running out on this. Intel will not be stuck on 14nm forever and they have a pretty good environment built up and they continue working on that. Even more so right now because the possible performance progress on heavily power-limited laptops is quite limited, 7/10nm or not.
But 28 cores om 14 nm. You know what that means, you need a chiller to cool that thing:p.
I will stick to My choise with ryzen 9 3950X. I dont need more than 16 cores.
Currently we see Intel struggle somewhat with multicore scaling, so when 10nm is refined they might actually achieve higher sustained multicore clocks. When it comes to ~5 GHz sustained turbo I'm more doubtful. But even if they achieve boost clocks at "just" ~4.5-4.6 GHz, the IPC gains of Sunny Cove should still outperform Coffee Lake boosting at ~5 GHz. I assume that Intel at some point will do core boosting more independently like AMD already does.
Additionally, all the delays of Ice Lake(Sunny Cove) means that Intel will launch two architectures with IPC gains in rapid succession. Ice Lake-SP will launch early 2020 and Sapphire Rapids-SP early 2021, wich hopefully a Ice Lake-X derived from that between there. In parallel with this, AMD plans Zen 3, 4…, so we should expect steady gains from both companies for the next ~3 years or so.
But going forward, IPC gains and SIMD will be they way to gain performance. More cores is good, but at some point there will be diminishing returns either way for synchronized tasks, which all will scale badly beyond ~16 cores or so anyway.
In previous years AMD manufactured PowerPoint presentations at frantic pace to show they still exist...
Now it's Intel turn!
Blues have nothing to respond, so they spew nonsensical propaganda.