Intel Skylake / Kaby Lake

[csbin]

Wrong again.



Here's what a regular (stock) 2P Xeon E5-2699 v4 system scores @ CB R15, tipically ranges from ~5200-5500 pts:

- 5459 pts
- 5459 pts
- 5225 pts
- 5214 pts

lol

103.9*36=3740MHZ（Single-core）

 

[TheF34RChannel]

Intel(R) Core(TM) i9-7900X CPU @ 3.30GHz
Score 1378.40Mpix/s
Multi-Media Integer 1443.56Mpix/s
Multi-Media Long-int 515.50Mpix/s
Multi-Media Quad-int 5120kpix/s
Multi-Media Single-float 1693.28Mpix/s
Multi-Media Double-float 1071.43Mpix/s
Multi-Media Quad-float 23848kpix/s

Intel(R) Core(TM) i7-6950X CPU @ 3.00GHz
Score 858.62Mpix/s
Multi-Media Integer 1Gpix/s
Multi-Media Long-int 339.38Mpix/s
Multi-Media Quad-int 4.87kpix/s
Multi-Media Single-float 1Gpix/s
Multi-Media Double-float 585.5Mpix/s
Multi-Media Quad-float 21Mpix/s

As I said, AVX-512 is fully enabled on SKL-X. It just scales back frequency in AVX-512 mode.

@coolaler.com an user claims 2x Xeon Platinum 8180 deliver a CB R15 MT score of 8301, as long as you cool the beasts properly (<80 °C). He also suggests an all-core Turbo of 3.2 GHz, really impressive for a 28C/56T part. In comparison, Xeon E5-2699 v4's all-core Turbo = 2.8 GHz for 22C/44T.

Yummie! Nice find!

 

[csbin]

2x Xeon Platinum 8180：3.2 GHz（ all-core）*56=179（ 8301 Score？）Screenshots？
2x Xeon E5-2699 v4：2.912 GHz（ all-core）*44=128（5922 Score）
2x Xeon Platinum 8168：3.4 GHz（ all-core）*48=163（ 7212 Score）

 

[Sweepr]

You can repost the world record score for a 2x Xeon E5-2699 v4 system at CB R15 using custom water @ 3.740 GHz as much as you want csbin. It still won't make it a valid comparison to Skylake-SP (till someone pushes it the same way). Once again, the typical scores from 2P Broadwell-EP here are in the ~5200-5500 range, which means >50% improvement in a single generation and 4-5% perf/clock uplift in this benchmark (2P Xeon Platinum 8180 = 8301 pts).

- 5459 pts
- 5459 pts
- 5225 pts
- 5214 pts

 

[csbin]

2x Xeon E5-2699 v4：2.912 GHz（ all-core）*44=128（5922 Score）
2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）

https://item.taobao.com/item.htm?spm=a1z10.3-c.w4002-16141298066.38.2RlVrn&id=42655744496

 

[csbin]

You can repost the world record score for a 2x Xeon E5-2699 v4 system at CB R15 using custom water @ 3.740 GHz as much as you want csbin. It still won't make it a valid comparison to Skylake-SP (till someone pushes it the same way). Once again, the typical scores from 2P Broadwell-EP here are in the ~5200-5500 range, which means >50% improvement in a single generation.

- 5459 pts
- 5459 pts
- 5225 pts
- 5214 pts

OK

2x Xeon E5-2699 v4：2.8 GHz（ all-core）*44=123（5200-5500 Score）
2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）

 

[LTC8K6]

OK

2x Xeon E5-2699 v4：2.8 GHz（ all-core）*44=123（5200-5500 Score）
2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）

Oh No! Intel is doomed! Run away!

 

[Sweepr]

It's happening.

First X299 motherboards listed in the US (ASUS)

http://www.shopblt.com/search/order_id=!ORDERID!&s_max=25&t_all=1&s_all=ASUS+X299&search=Search

Estimated arrival date = June 27. Pre-order, expect inflated pricing.

 

[Atari2600]

expect inflated pricing.

Naturally...

 

[TahoeDust]

It's happening.

First X299 motherboards listed in the US (ASUS)

http://www.shopblt.com/search/order_id=!ORDERID!&s_max=25&t_all=1&s_all=ASUS+X299&search=Search

Estimated arrival date = June 27. Pre-order, expect inflated pricing.

Hmmm...I wonder how inflated those prices are. I am interested in preordering my board/CPU from somewhere, but I'm not looking to get gouged.

 

[SAAA]

2x Xeon E5-2699 v4：2.912 GHz（ all-core）*44=128（5922 Score）46pt/GHz
2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）41pt/GHz

2x Xeon E5-2699 v4：2.912 GHz（ all-core）*44=128（5922 Score）46pt/GHz
2x Xeon Platinum 8168：3.4 GHz（ all-core）*48=163（ 7212 Score）44pt/GHz

Both give lower IPC for Skylake, that we know isn't a thing since all existing Skylake cpus have higher IPC than Broadwell in cinebench, only answer is that the record score for the E5-2699 v4 is indeed done running all core at 3.74GHz (well it's an ES with unlocked multiplier so single thread and all core speed can be the same. Also in all the posted screens the CPU-Z reported speed = all core speed...)

If that's so then:

2x Xeon E5-2699 v4：3.74 GHz（ all-core）*44=164（5922 Score）36pt/GHz
2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）41pt/GHz

2x Xeon E5-2699 v4：3.74 GHz（ all-core）*44=164（5922 Score）36pt/GHz
2x Xeon Platinum 8168：3.4 GHz（ all-core）*48=163（ 7212 Score）44pt/GHz

Now on average Skylake gives 18% better IPC. It's on the optimistic side given Intel itself stated some 10-15% better and that was counting higher clocks too.

At least this paints a better picture than other results like Geekbench scores: maybe at over 5.5GHz results aren't reliable after all.

 

[Ferzerp]

Again, as far as I am aware, cinebench does not support processor groups. I'm happy to be shown incorrect, but unless they've changed their program, it is meaningless when core count > 64.

 

[Ajay]

UGH! Multiplying cores times frequency doesn't equal anything! These are total bogus numbers and have no use. That's simply not how multi-core processors and multithreaded programs work

 

[crashtech]

UGH! Multiplying cores times frequency doesn't equal anything! These are total bogus numbers and have no use. That's simply not how multi-core processors and multithreaded programs work

Well, it's slightly useful as a hypothetical where everything scales linearly. There are some workloads that approach this. As a way of comparing relative execution throughput of very similar CPUs, it's marginally helpful but not probative by any means.

 

[wildhorse2k]

ROG STRIX X299-E GAMING specifications to study https://www.asus.com/ROG-Republic-Of-Gamers/ROG-STRIX-X299-E-GAMING/specifications/

44-Lane CPU-
Intel® Core™ X-Series Processors Family
3 x PCIe 3.0/2.0 x16 (x16, x16/x16, x16/x16/x8)
2 x PCIe 3.0 x4 (max at x4 mode) *1
28-Lane CPU-
Intel® Core™ X-Series Processors Family
3 x PCIe 3.0/2.0 x16 (x16, x16/x8, x16/x8/x1)
2 x PCIe 3.0 x4 (max at x4 mode) *1
16-Lane CPU-
Intel® Core™ X-Series Processors Family
3 x PCIe 3.0/2.0 x16 (x16, x8/x8, x8/x8/x1)
2 x PCIe 3.0 x4 (max at x4 mode) *1

It doesn't look too messy to me. You look at your CPU and know what to expect. 1 more row to look at than in X99. It appears X299 could be usable with 16 lane CPUs as well.

 

[SAAA]

UGH! Multiplying cores times frequency doesn't equal anything! These are total bogus numbers and have no use. That's simply not how multi-core processors and multithreaded programs work

That's just a way to approximate IPC, so it's definitely useful. Especially with cinebench that scales wonders with clocks, threads and IPC (and honestly that's how any cpu benchmark should behave).

Actually it's impressive how the score/GHz rises when the core count increases because it should be the opposite due to interconnect/scaling limitations:

2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）41pt/GHz

2x Xeon Platinum 8168：3.4 GHz（ all-core）*48=163（ 7212 Score）44pt/GHz

2x Xeon Platinum 8180：3.2 GHz（ all-core）*56=179（ 8301 Score, leaked so not 100% reliable) 46pt/GHz

Hey maybe platinum is just better than gold... or the greater L3 affects scores this way.

 

[TheF34RChannel]

What do you think if Intel would have introduced the KBL-X parts as purely parts for extreme overclockers to play with, and SKL-X for the rest of us? At least then it would have made some sort of sense. Heard the idea on a stream and it sounds sensible.

 

[Eug]

What Core m3 is this? The new Kaby Lake MacBook 12" inch has a Core m3 at 1.2 GHz base and 3.0 GHz turbo. I have not seen this chip anywhere.

Is this a Core m3-7Y32 with a higher TDP or something? If so, it makes the upgrade to a Core i5-7Y54 kind of pointless. I believe that's what is in the i5 MacBook, although their base frequency is 1.3 GHz (not 1.2 GHz), with Turbo up to 3.2 GHz.

That's just a 6-8% performance difference.

Their Core i7-7Y75 is 1.4 GHz (instead of 1.3 GHz), with the same 3.6 GHz Turbo. That should be an 8-13% improvement over i5 and 17-20% improvement over m3.

I was all set to get the i5 but now I'm considering the m3.

 

[jpiniero]

What Core m3 is this? The new Kaby Lake MacBook 12" inch has a Core m3 at 1.2 GHz base and 3.0 GHz turbo. I have not seen this chip anywhere.

Could be an offlabel model that is either binned a bit better or slightly higher TDP.

 

[Eug]

Could be an offlabel model that is either binned a bit better or slightly higher TDP.

That would make all the 12" MacBooks - Core m3, Core i5, and Core i7 - off label if the m3 is is off label, because I don't see those i5 and i7 chips listed anywhere either. So perhaps your latter statement is correct that they're just using standard m3-7Y32, i5-7Y54, and i7-7Y75 chips at a higher TDP.

In any case, that 7Y32 (or whatever it is) is a pretty nice jump over the 7Y30. I was not aware of this chip because it was only launched this year. The 7Y30 was out last year.

 

[Ajay]

What do you think if Intel would have introduced the KBL-X parts as purely parts for extreme overclockers to play with, and SKL-X for the rest of us? At least then it would have made some sort of sense. Heard the idea on a stream and it sounds sensible.

Well, it may turn out to be true that some Xtreme Overclockers will hit new highs on CPU clocks and Dram clocks. I think someone else in this thread mentioned that it may have been part of a strategy to move all unlocked processors onto one platform. I think that strategy is probably no longer a valid one with renewed competition from AMD.

 

[csbin]

2x Xeon E5-2699 v4：2.912 GHz（ all-core）*44=128（5922 Score）46pt/GHz
2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）41pt/GHz

2x Xeon E5-2699 v4：2.912 GHz（ all-core）*44=128（5922 Score）46pt/GHz
2x Xeon Platinum 8168：3.4 GHz（ all-core）*48=163（ 7212 Score）44pt/GHz

Both give lower IPC for Skylake, that we know isn't a thing since all existing Skylake cpus have higher IPC than Broadwell in cinebench, only answer is that the record score for the E5-2699 v4 is indeed done running all core at 3.74GHz (well it's an ES with unlocked multiplier so single thread and all core speed can be the same. Also in all the posted screens the CPU-Z reported speed = all core speed...)

If that's so then:

2x Xeon E5-2699 v4：3.74 GHz（ all-core）*44=164（5922 Score）36pt/GHz
2x Xeon Gold 6152：2.8 GHz（ all-core）*44=123（ 5069 Score）41pt/GHz

2x Xeon E5-2699 v4：3.74 GHz（ all-core）*44=164（5922 Score）36pt/GHz
2x Xeon Platinum 8168：3.4 GHz（ all-core）*48=163（ 7212 Score）44pt/GHz

Now on average Skylake gives 18% better IPC. It's on the optimistic side given Intel itself stated some 10-15% better and that was counting higher clocks too.

At least this paints a better picture than other results like Geekbench scores: maybe at over 5.5GHz results aren't reliable after all.

only answer is that the record score for the E5-2679 v4 is indeed done running all core at 1.2GHz (well it's an ES with unlocked multiplier so single thread and all core speed can be the same. Also in all the posted screens the CPU-Z reported speed = all core speed...)

https://item.taobao.com/item.htm?spm=a1z10.3-c-s.w4002-16087277618.30.sPYcrj&id=19752971013

 

[aigomorla]

You see those cpu that says Genuine Intel CPU 0000 <--- this is a stage 1 circulation cpu .
They tend to be on the first stepping, also most are intended for inhouse only.

They have no microcode for cpuid, because they are not intended for circulation at all...

Also they are the bane of my testing days because when something didnt work, it would take many emails to shimano on a custom bios for my classified to get the cpu to work right...

But it is safe to say, that is the first generation batch of skylake-e with A0 steppings....
Sort of like what i had to mess with on gulftown when they first came out and i was shrowded with a ton of NDA's.

BTW i had that gufly almost 6 months before launch. So its safe to say Skylake-E was in the hands of these guys at least 6 months prior to that guy leaking that post.

 

[IntelUser2000]

- 5459 pts
- 5459 pts
- 5225 pts
- 5214 pts

This is precisely the reason I pretty much disregard screenshot-based benchmark results and random leaderboard scores. The variance in the results easily overshadow any difference in perf/clock(unless its massive). It just spurs pointless posts in forums.

The only result that matters for the chips are on review sites when the NDA is lifted on the release day. Server guys wouldn't operate it out of spec anyway. So the core and clock comparisons are meaningless, even if we can completely separate them so it can be compared in the right way, which we can't.

 

[formulav8]

Argg a bug