Intel Core i9 Skylake - E clocks no higher than Broadwell - E

[Mr_Troll]

 

Quote

Intel Core i9 Skylake - E clock no higher than Broadwell -E

Doing my regular round of interneting this morning looking for good news to pass along, cageymaru passed along this tidbit from Guru3D that I found very interesting. Guru is passing along some of its findings out of Computex about the upcoming Intel Core i9 CPUs. First and foremost, it reported that the new 10-Core 7900X has been spotted pulling a Cinebench benchmark at 4.3GHz on a Liquid Cooling System (LCS), scoring 2,364. Now, while the 10-C Broadwell 6950X CPU was not that popular at $1400 in retail (7900X is reported to cost $1000), it was still easily pushed into the 4.3GHz range with a simple Corsair H80 cooling system with a push/pull fan setup. I would know because that is exactly what my system is set up with right now. I built this system recently to render 4K video, which it does quite well. Also this system will easily score close to 2,300 in Cinebench with the 2800MHz memory it is running. So all in all, the Skylake-E does not look much different than Broadwell-E, as many of you probably expected.

 

 

 

Also a line that caught my eye is this one in regards to LNG overclocking....

Quote

Its getting old and it is not at all important or relative towards you guys, the end user.

No, LNG overclocking has not been important or relative literally since its inception to the end user. Fun? Yes! Entertaining? Yes! But it not "getting old," it has been old for quite a while and Intel's lack of clocking scale in any significant form in its retail products is the reason LNG is shoved down our throats like it is actually meaningful.

Also Guru went on to say this about CPUs available at launch.....

Quote

None, and I do repeat this none of the partners have had their hands on, or even seen the 12, 14, 16 or that 18-core part. So that does raise some questions as earlier on we have already stated that Intel is rushing things as an answer towards AMD’s upcoming Threadripper processor series. So on that note, we can confirm that X299 and the respective processors will launch with up-to 10-core processors in the Skylake-X processor series. So that means that there will be three (Skylake-X) SKUs at launch, the Core i7 7800K six core, the 7820X with eight cores and the Core i9 7900K with 10 cores.

 

Expected little bit higher clock tbh.

 

Source:https://www.hardocp.com/news/2017/06/05/intel_core_i9_skylake_e_clocks_no_higher_than_broadwell/

https://www.guru3d.com/news-story/computex-2017-intel-core-i9-launches-with-up-to-10-cores-first-does-4-3-ghz-on-lcs.html

[Guest]

AMD seems to ripping intel apart with their threadripper

[samcool55]

L3 cache is 11-way, why is that such a weird number?

I won't be surprised broadwell-E is a better option because it has better thermals due to it being soldered.

The IPC gain can easily be offset by an extra 100-200Mhz, i guess.

[Bananasplit_00]

so x299 seems like even less of a thing i would want! great

[Dabombinable]

1 hour ago, samcool55 said:

L3 cache is 11-way, why is that such a weird number?

I won't be surprised broadwell-E is a better option because it has better thermals due to it being soldered.

The IPC gain can easily be offset by an extra 100-200Mhz, i guess.

Broadwell-E is also better suited to mITX as well because of its FIVR, which means that you should in theory get higher+more stable overclocks than Skylake-E due to the superior power delivery (considering Haswell/Devil's canyon and Broadwell can reach their overclocking limits on 4 phase motherboards).

[Taf the Ghost]

That i7-7900X score seems low. We have videos of 2400 scores with ghetto-level cooling.  I expect 2500-2700 under a decent water cooling + OC.

 

Now, will it take deliding and a new TIM? Probably. But that seems to be a "thing" for now.  (Yeah, let's test a delided, OC'd 7700k and call that a comparison test!)

 

Though the important detail is about the 12c part. I hadn't seen anyone mention that being unavailable for testing.  Very interesting.

[sazrocks]

The 12 core not being available is interesting. Was intel not planning for it until ryzen was announced?

[79wjd]

7 minutes ago, sazrocks said:

The 12 core not being available is interesting. Was intel not planning for it until ryzen was announced?

The original slides for Skylake-X showed a max of 10 cores.

[Taf the Ghost]

The slides from the end of 2016 had 12c listed, but maybe that was always supposed to come a month or two later. 

 

http://www.pcadvisor.co.uk/new-product/pc-upgrades/intel-core-i9-release-date-features-specs-3590751/

 

They've got a mention of August for the 12c part, so about 2 months after X299 launches.  If it launches at the end of June that's been hinted a few times.  Though that's "mainstream" parts of the platform launching on time is expected. I wonder if AMD is going to release more TR information a few days before the Intel launch. 

 

In related news for both platforms, it's going to be interesting to see what programs can actually reach saturation of all cores & threads. Though, me, I'm thinking if you could toss in 3 high-end graphics cards, tossed 5c/10t into a VM and run 3 separate 4K gaming machines at the same time.  (You can saturation 4K on a GPU with a 4c8t Ryzen, as is.)

[captain_to_fire]

10 hours ago, deXxterlab97 said:

AMD seems to ripping intel apart with their threadripper

AMD is very happy it seems

 

[haloharry97]

if they can get there gaming speed =< then skylake then i buy it.  

[rn8686]

1 minute ago, haloharry97 said:

if they can get there gaming speed =< then skylake then i buy it.  

Ryzen wont be quite as fast for gaming, but neither really are gaming CPUs. Only worth buy if you are doing some rendering alongside in which case Ryzen (Especially thread ripper) will be much better value). 

[Simon771]

1000$ for 2300 cinebench score with 10 core CPU

 

or 300$ for R7 1700 that will give you 1700 cinebench score with 8 cores.

 

7900X = 115 score per thread

R7 1700 = 106 score per thread

 

7900X = 2,3 cb score for 1$

R7 1700 = 5,6cb score for 1$

 

Wondering what will threadriper bring.

[Taf the Ghost]

1 hour ago, Simon771 said:

1000$ for 2300 cinebench score with 10 core CPU

 

or 300$ for R7 1700 that will give you 1700 cinebench score with 8 cores.

 

7900X = 115 score per thread

R7 1700 = 106 score per thread

 

7900X = 2,3 cb score for 1$

R7 1700 = 5,6cb score for 1$

 

Wondering what will threadriper bring.

Top SKU is somewhere between 3000 and 3200, apparently, depending on who's been running it.  But that's also pretty much 2 1800X working together at stock clocks. I fully expect to see the top SKU Threadripper getting 3500 to 3600 Cinebench R15 under a high all-core OC. (Obviously on a very large Rad.) 

 

Skylake-X scores are coming in almost exactly at Kabylake scores. So you can pretty much predict, the scores by taking a 7700k, at stock, and multiplying its Multi-core score by the Core Count / 4.  That should top out the i7-7900X (10c/20t part) at around 2700 in Cinebench under a high all-core OC.  This is going to roughly similar to the 12c Threadripper model.

 

That's actually what Intel is most concerned about. The Clock difference (much more than the IPC difference) is what they have going for them. Once you start talking 14 cores compared to 10 cores, Intel is going to be rather far down.  It gets weirder once you hit 16c & 18c on SK-X though, as those are going to be much lower clocked than the 10c and below parts.  Ryzen 7 can't clock over 4.1 Ghz because of Process issues; not the architecture. That means the performance we see, currently, is capped by a non-thermal limit.  Thus you can slap two of them together and they're going to perform pretty dang close to simply putting two of them together. 

 

There's so many angles to all of this.  I find it utterly fascinating. AMD's performance deficit compared to Intel in all but gaming goes away when you can simply throw more cores at the problem, and it adds a few other benefits when you do that.  If you're a heavy VM users, more cores = more VMs. You need enough power for those VMs, but you don't need super high clocks for them.  If you can rock a 16c32t at Consumer-level Gaming Speeds for 1/2 the price of the Intel offering, that's going to be valuable for a lot of use cases.

 

This explains why Intel has responded they way they have. They added the 12c, originally, because they knew details of Ryzen in late 2016. They wanted to push their HEDT stack far enough away from Ryzen 7 that it doesn't make much sense to use it as a cheap HEDT setup.  (Even if it'll work for almost all of those work flows.) Once Threadripper became more than just a few engineering samples, Intel set about to addressing the problem.  Which should be around Jan/Feb 2017.  That explains the time-line, finally.

 

- 2015 Cannon Lake was "canceled" and replaced with Icelake for 10 nm. Cannon Lake is, still, actually happening but not in the normal desktop space.

- In 2016, X299 got the 4 core Kaby Lake because there was issues with 10 nm, plus some odd uses cases for 4c parts on the platform. (Cheap to do?)

- Mid 2016, they added Coffee Lake for Z370 as another mid-cycle refresh because the die-shrink wasn't happening on schedule.

- Late 2016, Intel seems to have looked to move Coffee Lake up to late 2017/early 2018, depending on model range. (Probably timed for 10-12 months before Icelake in 2018.)

- Late 2016, detailed Ryzen information gets to Intel. HEDT could be a problem even with IPC/Clock advantage with just a 10c part against Ryzen 7 with 8c.  Intel adds 12c by pillaging from Xeons.

- March 2017, Snowy Owl becomes Threadripper on HEDT. Intel really panics to find an 18c part by digging even deeper into the Xeon stack that's coming on Skylake-EX / EP platforms.

- May 2017, Computex leaves everyone confused about what Intel was doing.

 

Aligning the timeline explains what happened.  Intel got hit with a 4 separate issues. 1) 10 nm on Mainstream will be 2 full years late. 2) Yearly Product Refreshes were inserted because timing became a mess and Intel will keep grinding uplift from small process & architecture tweaks.  3) Ryzen came out 20% better than even AMD was expecting. 4) Questionable 4c on X299 became more questionable when 12c was added; problem compounded with adding 14c, 16c and 18c parts.

 

Yes, X299 is a real mess. And, just to make the situation worse, Intel brought back Motherboard dongles because they'd end up cannibalizing their own product segmentation.

[Simon771]

27 minutes ago, Taf the Ghost said:

Top SKU is somewhere between 3000 and 3200, apparently, depending on who's been running it.  But that's also pretty much 2 1800X working together at stock clocks. I fully expect to see the top SKU Threadripper getting 3500 to 3600 Cinebench R15 under a high all-core OC. (Obviously on a very large Rad.) 

 

Skylake-X scores are coming in almost exactly at Kabylake scores. So you can pretty much predict, the scores by taking a 7700k, at stock, and multiplying its Multi-core score by the Core Count / 4.  That should top out the i7-7900X (10c/20t part) at around 2700 in Cinebench under a high all-core OC.  This is going to roughly similar to the 12c Threadripper model.

 

That's actually what Intel is most concerned about. The Clock difference (much more than the IPC difference) is what they have going for them. Once you start talking 14 cores compared to 10 cores, Intel is going to be rather far down.  It gets weirder once you hit 16c & 18c on SK-X though, as those are going to be much lower clocked than the 10c and below parts.  Ryzen 7 can't clock over 4.1 Ghz because of Process issues; not the architecture. That means the performance we see, currently, is capped by a non-thermal limit.  Thus you can slap two of them together and they're going to perform pretty dang close to simply putting two of them together. 

 

There's so many angles to all of this.  I find it utterly fascinating. AMD's performance deficit compared to Intel in all but gaming goes away when you can simply throw more cores at the problem, and it adds a few other benefits when you do that.  If you're a heavy VM users, more cores = more VMs. You need enough power for those VMs, but you don't need super high clocks for them.  If you can rock a 16c32t at Consumer-level Gaming Speeds for 1/2 the price of the Intel offering, that's going to be valuable for a lot of use cases.

 

This explains why Intel has responded they way they have. They added the 12c, originally, because they knew details of Ryzen in late 2016. They wanted to push their HEDT stack far enough away from Ryzen 7 that it doesn't make much sense to use it as a cheap HEDT setup.  (Even if it'll work for almost all of those work flows.) Once Threadripper became more than just a few engineering samples, Intel set about to addressing the problem.  Which should be around Jan/Feb 2017.  That explains the time-line, finally.

 

- 2015 Cannon Lake was "canceled" and replaced with Icelake for 10 nm. Cannon Lake is, still, actually happening but not in the normal desktop space.

- In 2016, X299 got the 4 core Kaby Lake because there was issues with 10 nm, plus some odd uses cases for 4c parts on the platform. (Cheap to do?)

- Mid 2016, they added Coffee Lake for Z370 as another mid-cycle refresh because the die-shrink wasn't happening on schedule.

- Late 2016, Intel seems to have looked to move Coffee Lake up to late 2017/early 2018, depending on model range. (Probably timed for 10-12 months before Icelake in 2018.)

- Late 2016, detailed Ryzen information gets to Intel. HEDT could be a problem even with IPC/Clock advantage with just a 10c part against Ryzen 7 with 8c.  Intel adds 12c by pillaging from Xeons.

- March 2017, Snowy Owl becomes Threadripper on HEDT. Intel really panics to find an 18c part by digging even deeper into the Xeon stack that's coming on Skylake-EX / EP platforms.

- May 2017, Computex leaves everyone confused about what Intel was doing.

 

Aligning the timeline explains what happened.  Intel got hit with a 4 separate issues. 1) 10 nm on Mainstream will be 2 full years late. 2) Yearly Product Refreshes were inserted because timing became a mess and Intel will keep grinding uplift from small process & architecture tweaks.  3) Ryzen came out 20% better than even AMD was expecting. 4) Questionable 4c on X299 became more questionable when 12c was added; problem compounded with adding 14c, 16c and 18c parts.

 

Yes, X299 is a real mess. And, just to make the situation worse, Intel brought back Motherboard dongles because they'd end up cannibalizing their own product segmentation.

Thanks for your detailed explanation. It makes more sense now, why this is happening.

On side note ... you said that Ryzen is capped because of the process. Is that why we can expet Ryzen + (or Ryzen 2 ... whatever they will name it=, to work on 4,7GHz on all cores?

[Taf the Ghost]

24 minutes ago, Simon771 said:

Thanks for your detailed explanation. It makes more sense now, why this is happening.

On side note ... you said that Ryzen is capped because of the process. Is that why we can expet Ryzen + (or Ryzen 2 ... whatever they will name it=, to work on 4,7GHz on all cores?

 

I'd be surprised if the 14nm+ process gets more than 4.3 Ghz, honestly. I'd expect some % of IPC uplift just because they know much more about the architecture in practice. 

 

Zen 2 will be on 7 nm. Apparently the first Tapeout is later in the year, so it should be around for 2019.

[tom_w141]

X299 sinks deeper and deeper in to despair. Hopefully this is enough of a wakeup call for Intel to pull their god damn finger out and get back to seriously competing, rather than spoon feeding incremental gains.

 

I hope TR takes a nice % of HEDT sales

[Valentyn]

9 minutes ago, Taf the Ghost said:

 

I'd be surprised if the 14nm+ process gets more than 4.3 Ghz, honestly. I'd expect some % of IPC uplift just because they know much more about the architecture in practice. 

 

Zen 2 will be on 7 nm. Apparently the first Tapeout is later in the year, so it should be around for 2019.

AMD mentioned in the old Ryzen AMA that they already know where they can make easy gains for IPC.

 

Zen+ might seen a 5-7% IPC bump, which would put it around Kaby Lake, which is nice.

One would be sorting out the bandwidth of Infinity Fabric, and other "simple" improvements. Add in 7nm, that'll give a much appreciated clock speed bump and we could see some decent gains.

Add in AMD's better SMT implementation and they're on to a fantastic future architecture.

One thing that I'm happy about Intel responding so aggressively with, is that they're market leaders. If they start making 6 cores mainstream to replace the 7600 and 7700 it might finally give games developers the kick up the rear to use more cores.

 

Although with that, it makes the 4core x299 processors look even sillier; as those and 6 core one will be very short lived.

[Space Reptile]

 

my reaction

[haloharry97]

3 hours ago, rn8686 said:

Ryzen wont be quite as fast for gaming, but neither really are gaming CPUs. Only worth buy if you are doing some rendering alongside in which case Ryzen (Especially thread ripper) will be much better value). 

;( if that is true then i just have to wait for intel to do 8core at a cheap price.

[rn8686]

4 minutes ago, haloharry97 said:

;( if that is true then i just have to wait for intel to do 8core at a cheap price.

Not going to happen, at all. Ryzen is still very good for gaming, and with fast ram can basically reduce to the gap to almost nothing. 

[porina]

On the original topic of the new Skylakes not OC-ing, that is interesting if it is confirmed once they are more widely available. On 115x, Skylake easily clocks past Broadwell. But the potential spanner in the works is that Skylake-X is a more radical change, at least with respect to the attached cache structure. Maybe that is the limiting factor?

 

6 minutes ago, Valentyn said:

AMD mentioned in the old Ryzen AMA that they already know where they can make easy gains for IPC.

 

Zen+ might seen a 5-7% IPC bump, which would put it around Kaby Lake, which is nice.

Add in AMD's better SMT implementation and they're on to a fantastic future architecture.

This has been the tricky part when doing comparisons. A single thread on a single core still favours Intel right now, but two threads on a single core (with HT, SMT) already can swing towards Ryzen. While I don't doubt there will be improvements, the question is under which scenarios? I'd also hope they do some changes to allow somewhat higher clocks to be more easily attained, in a similar way to Kaby Lake compared to Skylake. Get turbo clocks well up into the 4+ GHz range and that would be sweet.

[Valentyn]

5 minutes ago, haloharry97 said:

;( if that is true then i just have to wait for intel to do 8core at a cheap price.

By the time that happens AMD will be out of debt, and will have the cash for R&D to match or beat intel most likely.
Ryzen is around 7% behind Kaby Lake for IPC, despite AMD's company worth being less than the R&D Budget for Intel.  (Bit Wit found a 4Ghz 1700 was 7% behind a 5Ghz 7700K in this gaming suite)

The improved 14nm+ process alone could potentially close that IPC gap; and help bump up the clock speed a little to close the gap even more.

[MageTank]

Just now, rn8686 said:

Not going to happen, at all. Ryzen is still very good for gaming, and with fast ram can basically reduce to the gap to almost nothing. 

I wouldn't go as far to say "almost nothing". I'd say it's enough, especially if you are pushing a realistic resolution with some kind of variable refresh rate monitor (Freesync, G-Sync, etc). There is still a real IPC disadvantage of about 5-7%, and a significant clock speed disadvantage of about 20-25% when compared to the 7700k at it's average overclock. That being said, the addition of more cores for a cheaper consumer product means more stable minimum framerates in applications that can take advantage of them (minimum framerates being the most important) and any sort of multi-tasking shows a pretty serious boon when using Ryzen.

 

If all you do is strictly gaming, and nothing else, a highly clocked 7700k with highly clocked ram, is untouchable at the moment. Not trying to bring Ryzen down, only being honest here.

[rn8686]

Just now, MageTank said:

I wouldn't go as far to say "almost nothing". I'd say it's enough, especially if you are pushing a realistic resolution with some kind of variable refresh rate monitor (Freesync, G-Sync, etc). There is still a real IPC disadvantage of about 5-7%, and a significant clock speed disadvantage of about 20-25% when compared to the 7700k at it's average overclock. That being said, the addition of more cores for a cheaper consumer product means more stable minimum framerates in applications that can take advantage of them (minimum framerates being the most important) and any sort of multi-tasking shows a pretty serious boon when using Ryzen.

 

If all you do is strictly gaming, and nothing else, a highly clocked 7700k with highly clocked ram, is untouchable at the moment. Not trying to bring Ryzen down, only being honest here.

I was referring to the Skylake i9s.