Intel Core i7-1185G7 "Tiger Lake" Ships with 4.70 GHz Turbo Boost Speeds

[btarunr]

Intel spoke of a "double digit percentage performance growth generation on generation" at its product reveal for "Tiger Lake" along the sidelines of its CES event. We now have a theory as to how they arrived at that. The company's 11th generation Core "Tiger Lake" processor, scheduled to launch sometime mid-2020, could bring about big gains in per-core performance for the ultraportable segment. PC enthusiast MebiuW, who has had a high hit-rate with Intel leaks, revealed that the flagship "Tiger Lake" part, the Core i7-1185G7, could ship with a CPU Turbo Boost speed of 4.70 GHz, a steep increase from the 3.90 GHz of the top current "Ice Lake" part, the i7-1065G7. The increased clock speeds, coupled with the more advanced "Willow Cove" CPU cores appear to be the 11th generation chip's value proposition.



View at TechPowerUp Main Site

 

[bonehead123]

WOW... a whole .80ghz increase....

w/o/w/////

I'm sooooo impressed I could just go poopoo in me doodoo.......

 

[xvi]

I'll take a 10% increase in clock speed. Curious to know if IPC is improved.

 

[Caring1]

Intel, confusing speed with performance still.

 

[tabascosauz]

I'll take a 10% increase in clock speed. Curious to know if IPC is improved.

Should be a slight IPC boost. Tiger Lake's Willow Cove has a "new cache subsystem" over Sunny Cove. One of Sunny Cove's defining features over Skylake is the L1 and L2 both getting bigger and more associative, but L3 remaining the same, so it sounds like Willow Cove is set up for L3 to get the same treatment this time around. Which is more or less what AMD did with Matisse, but Matisse's L3 is still segregated amongst CCX. It will be interesting to see a unified and doubled L3 in Willow if the rumors are true.

What concerns me is the reported jump in max Turbo speed. Has this lemon of a 10nm+ process really improved that much in just one generation (on the level of moving straight from the 14nm GloFlo to N7 TSMC)? Or is Intel just taking a page out of AMD's Matisse and their own Coffee Lake playbook by pushing both VID and clocks to the absolute breaking point, all in the name of trying to be competitive without actually having competitive hardware?

Also, just look at the current Ice Lake lineup. The current flagship 1068G7 tops out at 4.1GHz, and needs a bump to a 28W TDP to accommodate it. 28W is quite a bit outside of what a true ultraportable can really handle, and outside of ultraportable territory, Ice Lake makes no sense due to its core count limit. Consequently, the 15W 1065G7 has been the SKU of choice for implementing Ice Lake. From what it sounds like, 1185G7 is most likely not going to achieve 4.7GHz in an ultrabook TDP envelope, making it also a nonsensical proposition.

 

[btarunr]

Should be a slight IPC boost. Tiger Lake's Willow Cove has a "new cache subsystem" over Sunny Cove. One of Sunny Cove's defining features over Skylake is the L1 and L2 both getting bigger and more associative, but L3 remaining the same, so it sounds like Willow Cove is set up for L3 to get the same treatment this time around. Which is more or less what AMD did with Matisse, but Matisse's L3 is still segregated amongst CCX. It will be interesting to see a unified and doubled L3 in Willow if the rumors are true.

When AMD tinkered with caches (Zen+), it yielded a 3-5% IPC increase over the original Zen. I'm sure Willow Cove will come with an IPC increase.

 

[jeremyshaw]

Should be a slight IPC boost. Tiger Lake's Willow Cove has a "new cache subsystem" over Sunny Cove. One of Sunny Cove's defining features over Skylake is the L1 and L2 both getting bigger and more associative, but L3 remaining the same, so it sounds like Willow Cove is set up for L3 to get the same treatment this time around. Which is more or less what AMD did with Matisse, but Matisse's L3 is still segregated amongst CCX. It will be interesting to see a unified and doubled L3 in Willow if the rumors are true.

What concerns me is the reported jump in max Turbo speed. Has this lemon of a 10nm+ process really improved that much in just one generation (on the level of moving straight from the 14nm GloFlo to N7 TSMC)? Or is Intel just taking a page out of AMD's Matisse and their own Coffee Lake playbook by pushing both VID and clocks to the absolute breaking point, all in the name of trying to be competitive without actually having competitive hardware?

Also, just look at the current Ice Lake lineup. The current flagship 1068G7 tops out at 4.1GHz, and needs a bump to a 28W TDP to accommodate it. 28W is quite a bit outside of what a true ultraportable can really handle, and outside of ultraportable territory, Ice Lake makes no sense due to its core count limit. Consequently, the 15W 1065G7 has been the SKU of choice for implementing Ice Lake. From what it sounds like, 1185G7 is most likely not going to achieve 4.7GHz in an ultrabook TDP envelope, making it also a nonsensical proposition.

While Intel's current and immediate future L3 cache isn't really "unified" in that sense, either, it does have more uniform latency across all of it. AMD has really local L3 that is faster (for the 4 cores attached to it), futher that is slightly slower (the other 4 cores of the Zen2Chiplet/Zen1Die), and others that are varying degrees of even slower (off die, NUMA, etc). Since a lot of stuff is still lowly threaded (or largely still local), the layout doesn't suffer all that much in threaded tasks. AMD did increase the cache sizes to somewhat compensate for the extra latency.

That being said, I do wonder how the fully unified L3 would affect latency in Zen 3.

 

[watzupken]

Looking forward to see it. I am somewhat skeptical because I feel Intel's 10nm is does not feel as great as what you see on the 7nm TSMC chips from AMD. Currently while using an Ice Lake CPU for my laptop, I still feel that despite the slower clockspeed, it still runs somewhat hot as compared to the older 14nm. Now the Tiger Lake will have a new CPU architecture, the Xe GPU is the part that will likely take up most of the die space. So while Intel can slap a "4.7Ghz" boost claim, I am unsure if it can even get close to this boost rate under sustain load. AMD Renoir is beating them left right center because it seems it can sustain a higher clockspeed over the base clock better, and with more cores as compared to the 4c/8t config for current Ice Lake U, and likely also Tiger Lake U series.

Also, just look at the current Ice Lake lineup. The current flagship 1068G7 tops out at 4.1GHz, and needs a bump to a 28W TDP to accommodate it. 28W is quite a bit outside of what a true ultraportable can really handle, and outside of ultraportable territory, Ice Lake makes no sense due to its core count limit. Consequently, the 15W 1065G7 has been the SKU of choice for implementing Ice Lake. From what it sounds like, 1185G7 is most likely not going to achieve 4.7GHz in an ultrabook TDP envelope, making it also a nonsensical proposition.

I think this further adds on to my claim that Intel's 10nm is not doing well at all, even when compared to their existing 14nm. All Ice Lake U CPUs have a substantially lower base clock, starting from 1Ghz for the i5 1035G1 as compared to their 14nm counterpart. So I feel its already a tell tale sign. For the top end, we also see an increase in TDP from 25W to 28W, which I feel is another sign of a poor 10nm.

Even with an upgrade to 10nm+, it will not miraculously allow them to accomodate a new CPU and GPU architecture, yet allowing for such a high clockspeed. In reality, the TDP again is another cryptic metric to measure power requirement. I suspect at full tilt, the Tiger Lake U is going to need substantially more power.

 

[Vayra86]

Ah yes, "Up To".

I'll think I'll buy Up To zero

 

[efikkan]

What concerns me is the reported jump in max Turbo speed.

Rated clock speeds are increasingly becoming just a marketing thing. Existing U series CPUs can usually only hit their boost speeds for a few seconds, if at all.

Has this lemon of a 10nm+ process really improved that much in just one generation (on the level of moving straight from the 14nm GloFlo to N7 TSMC)? <snip>

Also, just look at the current Ice Lake lineup. The current flagship 1068G7 tops out at 4.1GHz, and needs a bump to a 28W TDP to accommodate it…

Marketing aside, the second generation 10nm process (10nm+) is significantly better in terms of yields. Intel have themselves stated that the yields are "better than expected".

The alleged benchmarks we've seen of Tiger Lake-U should be a good indicator: (even if we assume these are 27W TDP)

<unnamed ES>    2.7 GHz / 4.3 GHz
i7-1165G7       2.8 GHz / ?
i7-1185G7       3.0 GHz / 4.7 GHz

Compared to Ice Lake-U:

i7-1065G7 (15W) 1.3 GHz / 3.9 GHz
i7-1065G7 (25W) 1.5 GHz / 3.9 GHz(?)
i7-1068G7 (27W) 2.3 GHz / 4.1 GHz

Even if the rated speeds are a bit optimistic, this should be a good indicator, especially the base clock.

While the yields of 10nm were horrendous in the beginning, the yields of 10nm+ should be good. (Don't forget there were a huge difference in 14nm => 14nm+ too) The remaining issue is production volume, and since Intel under estimated the demand in laptops and servers when they planned these production lines many years ago, the 10nm family will never reach the required production volume to cover the entire lineup. 10nm+ will be used for at least Tiger Lake-Y/-U/-H and Ice Lake-SP/-X coming "this year", but for mainstream desktop nobody knows for sure.

 

[jpvalverde85]

Dude, U series CPUs have a reasonable amount of grunt even on constant crunching at 15w, i have a Skylake GT2 i5 (2.3GHz regular clockspeed) on my Thinkpad and usually on number crunching the CPU Turbo goes 2.5-2.7 GHz for large amounts of time (plus half hour). Also on Kaby Lake R i have observed similar behavior (Lenovo v330) (1.6GHz regular clockspeed but Turbo up to 2.5GHz on number crunching). Clocks may go down when IGP kicks in @15w power figure, gaming maybe? Anyway, Renoir will be a royal spank to Intel.

 

[watzupken]

Rated clock speeds are increasingly becoming just a marketing thing. Existing U series CPUs can usually only hit their boost speeds for a few seconds, if at all.


Marketing aside, the second generation 10nm process (10nm+) is significantly better in terms of yields. Intel have themselves stated that the yields are "better than expected".

The alleged benchmarks we've seen of Tiger Lake-U should be a good indicator: (even if we assume these are 27W TDP)

<unnamed ES>    2.7 GHz / 4.3 GHz
i7-1165G7       2.8 GHz / ?
i7-1185G7       3.0 GHz / 4.7 GHz

Compared to Ice Lake-U:

i7-1065G7 (15W) 1.3 GHz / 3.9 GHz
i7-1065G7 (25W) 1.5 GHz / 3.9 GHz(?)
i7-1068G7 (27W) 2.3 GHz / 4.1 GHz

Even if the rated speeds are a bit optimistic, this should be a good indicator, especially the base clock.

While the yields of 10nm were horrendous in the beginning, the yields of 10nm+ should be good. (Don't forget there were a huge difference in 14nm => 14nm+ too) The remaining issue is production volume, and since Intel under estimated the demand in laptops and servers when they planned these production lines many years ago, the 10nm family will never reach the required production volume to cover the entire lineup. 10nm+ will be used for at least Tiger Lake-Y/-U/-H and Ice Lake-SP/-X coming "this year", but for mainstream desktop nobody knows for sure.

I am skeptical that these Tiger Lake models run at 15W looking at the base clockspeed. While the yields improved, I doubt they can create a miracle just through minor refinements on the 10nm. And don't forget that Tiger Lake comes the greatly improved Xe graphics which I believe will take a big chunk of die space and likely also power, on top of the CPU architecture improvement.

In fact if you look at the current Ice Lake numbers you provided, the base clock don't add up to me. A 10W increase in TDP only resulted in 200Mhz boost, while a further 2W increase resulted in a 800Mhz boost. The higher the clockspeed goes, the more power it will require. So in this case, I can only ascertain that Intel kind of understated the base clock may be due to poor yields at the start. In fact if I look at the i5 Ice Lake U processor in my laptop, I don't observe it going below 1.3Ghz under heavy load.

 

[efikkan]

I am skeptical that these Tiger Lake models run at 15W looking at the base clockspeed. While the yields improved, I doubt they can create a miracle just through minor refinements on the 10nm. And don't forget that Tiger Lake comes the greatly improved Xe graphics which I believe will take a big chunk of die space and likely also power, on top of the CPU architecture improvement.

Considering how bad the yields were to begin with, this improvement shouldn't be unlikely. Also these clock speeds are fairly comparable to what AMD achieves on the "comparable" 7nm TSMC node.

In fact if you look at the current Ice Lake numbers you provided, the base clock don't add up to me. A 10W increase in TDP only resulted in 200Mhz boost, while a further 2W increase resulted in a 800Mhz boost. The higher the clockspeed goes, the more power it will require. So in this case, I can only ascertain that Intel kind of understated the base clock may be due to poor yields at the start. In fact if I look at the i5 Ice Lake U processor in my laptop, I don't observe it going below 1.3Ghz under heavy load.

I think i7-1068G7 deserves a big asterisk behind it. While it is listed in various tables, I don't think it has materialized yet.

 

[davideneco]

Rated clock speeds are increasingly becoming just a marketing thing. Existing U series CPUs can usually only hit their boost speeds for a few seconds, if at all.


Marketing aside, the second generation 10nm process (10nm+) is significantly better in terms of yields. Intel have themselves stated that the yields are "better than expected".

The alleged benchmarks we've seen of Tiger Lake-U should be a good indicator: (even if we assume these are 27W TDP)

<unnamed ES>    2.7 GHz / 4.3 GHz
i7-1165G7       2.8 GHz / ?
i7-1185G7       3.0 GHz / 4.7 GHz

Compared to Ice Lake-U:

i7-1065G7 (15W) 1.3 GHz / 3.9 GHz
i7-1065G7 (25W) 1.5 GHz / 3.9 GHz(?)
i7-1068G7 (27W) 2.3 GHz / 4.1 GHz

Even if the rated speeds are a bit optimistic, this should be a good indicator, especially the base clock.

While the yields of 10nm were horrendous in the beginning, the yields of 10nm+ should be good. (Don't forget there were a huge difference in 14nm => 14nm+ too) The remaining issue is production volume, and since Intel under estimated the demand in laptops and servers when they planned these production lines many years ago, the 10nm family will never reach the required production volume to cover the entire lineup. 10nm+ will be used for at least Tiger Lake-Y/-U/-H and Ice Lake-SP/-X coming "this year", but for mainstream desktop nobody knows for sure.

remember
ICL ES clock base is >2.7-3ghz

 

[remixedcat]

Is this gonna be as hype as tiger king..?

 

[Vayra86]

Considering how bad the yields were to begin with, this improvement shouldn't be unlikely. Also these clock speeds are fairly comparable to what AMD achieves on the "comparable" 7nm TSMC node.


I think i7-1068G7 deserves a big asterisk behind it. While it is listed in various tables, I don't think it has materialized yet.

I'm getting a strong impression that Tiger Lake is going to turn into 14nm+++ on steroids sooner than we might think, to be honest. Surely they didn't already apply their 14nm MO to this node too? It sure does look like it. Did the yields really improve, or is Intel already eating away at the optimal power curve? Meanwhile, no real volume behind it... and we don't know how many chips end up in the scrapper.

At the same time, if your node yield has indeed improved, wouldn't you be heavily focused on switching everything over AT LAST instead of weekly reminders of yet another 14nm Lake release...

I'm not convinced just yet... With this company now more than ever its about product on shelves, widely available. Until then things may 'materialize' but what are we really looking at... proof of concept.