Discussion Intel current and future Lakes & Rapids thread

[DrMrLordX]

Only if Cezanne will be available in January but historically it will be available around April-May because 1 year is the usual minimum cadence.

People seem to think March is most likely. It really depends on whether OEMs are putting pressure on AMD to release it.

 

[LightningZ71]

I don’t know why they would be pushing so hard. Retailers are moving almost every laptop they get in stock in the sub $700 market within hours of it hitting the shelves. More expensive stuff hangs out a little bit longer. The market isn’t demanding higher performance, it’s demanding everything that can reach the store. And it’s doing this during a global recession!

that environment favors guarded change, which allows higher volume.

 

[itsmydamnation]

I am interested in worst case scenarios. Prime95, small FFTs, locked to 15W. Renoir ends up between 1-1.2ghz. Does Tiger Lake fair better? I imagine so.

my 4700u in 15w holds 2.5ghz just fine?

From what we're seeing on the CPU side, Tiger Lake is significantly better in single thread than any of the 4xxx series mobile chips,

At twice the power consumption. If AMD took the same path they could probably get within ~15 percent perf in single thead.

 

[DrMrLordX]

that environment favors guarded change, which allows higher volume.

Not necessarily. There's still a push for better perf/watt, and Intel hasn't exactly been excelling in that arena. TigerLake is a nice change of pace from IceLake-U and the 14nm chips, but a push for further advancements from AMD (after seeing what Renoir could do) would be a vote of no-confidence in Intel getting Alder Lake-P out in a timely fashion.

 

[tamz_msc]

At twice the power consumption. If AMD took the same path they could probably get within ~15 percent perf in single thead.

There is like a ~50pt difference between the 4800U and 4900HS is Geekbench 5 ST integer score, which means that there is no more headroom left in Renoir even if you double the power limits(in tests the U-series are limited to 30W and H/HS often reach 60W or more). Tigerlake on the other hand scores ~35% better in GB5 ST integer while pulling 50W. Granted that Intel is still not up to the mark in perf/watt, but it's ST perf is still very impressive.

 

[coercitiv]

Anandtech's TGL review is up.

 

[jpiniero]

Hey, it does beat Renoir in several IGP games. Probally due to the memory bandwidth.

 

[Hitman928]

Hmm, I thought Tigerlake was supposed to have AI powered turbo behavior for better power/turbo roll-off. In this graph it's showing Renoir doing what Intel said Tigerlake would do. Is something wrong here or does this AI powered turbo not work when set to 15W with a heavy load?

 

[Hitman928]

Nevermind, found it in the intel vs intel power charts. I guess you have to have Adaptix installed/configured and due to time constraints they only used Adaptix for this one chart:



Would be interesting to see the rest of the test redone with Adaptix on. Since it is a 'suite' of technologies I also wonder how many OEMs will actually use it?

 

[clemsyn]

Initial Video Reviews of Tigerlake. Very interesting

- YouTube

Auf YouTube findest du die angesagtesten Videos und Tracks. Außerdem kannst du eigene Inhalte hochladen und mit Freunden oder gleich der ganzen Welt teilen.

www.youtube.com

Looks like XE is getting good reviews but looks like they need to fix some driver issues.

 

[LightningZ71]

So, in perusing the AT review, and looking at the graphs, it appears that 10sf still has a bit of a power/performance curve disadvantage vs. TSMC N7. It's not huge, but it seems to be there. However, Intel really seems to be able to wring the most out of 10SF in single core. Looking at SPEC ST, at "15"w, TGL still seems to be keeping an edge on AMD. I think part of that is related to the larger L3, and some of it is related to the IPC advantage that Willow Cove has over ZEN2. Looking at the die sizes, Renoir and 4C+Xe TGL are similar in size, but AMD is shoving double the CPU cores in there and still managing to stay close in the iGPU game, though ,with driver improvements, I think that Intel will get better faster than AMD.

I don't think that Cezanne makes a huge impact on N7, but maybe take advantage of better headroom on N7"improved" and keeps in the race in ST, and manages even better in MT. I am interested in if AMD makes any sort of improvement in memory controller efficiency with Cezanne, as, it does appear that Intel's better controller is helping their iGPU scores.

I wonder how Xe scales going down market? This was a 96 EU part with a high power limit. How does the 80EU version perform? Is the 48EU config performing at half the score of the 96 EU one, or is the scaling very sublinear?

 

[eek2121]

Pushing it way past reasonable limits and using clocks to compensate everything else was the ideology of Netburst.



Is that using CPU only load?

Prime95 uses AVX/AVX2. The Anandtech deep dive already gave me the information I was seeking, however. Intel has a leg up on AMD when it comes to efficiency, but AMD has more cores so it wins at anything multithreaded. Too bas we won’t see an 8 core Tiger Lake U. At least we will see an 8 core H variant.

Those number speak to the situation for AMD that even the 80EU parts from Tiger Lake will produce gpu performance that is at least as fast as their top performing 8CU Vega mobile solution at similarly configured power limits.

Unless Lucien has remarkably better thermals, and Cezanne clocks those VEGA CUs at much higher levels, the next year or so is Intel's mobile market to loose with respect to performance in this sector. From what we're seeing on the CPU side, Tiger Lake is significantly better in single thread than any of the 4xxx series mobile chips, and it takes the full house 4800u to significantly beat it in multi-thread scenarios as the 4600u and 4700u aren't appreciably better or worse than the Tiger Lake parts in their MT scores.

I guess, at this point, what remains to be seen is volume numbers. How many chips can both manufacturers get out the door ad how many can the device manufacturers get into their products and shipped? Also, when shopping the remaining Ice Lake inventory against the new Tiger Lake parts, it's interesting to note that the bottom tier Tiger Lake I3-1125G4 is essentially as good or better than any of the Ice Lake I5s and even rivals the I7-1060/65 in many ways.

Price and availability are always an issue with Intel.

People seem to think March is most likely. It really depends on whether OEMs are putting pressure on AMD to release it.

I don’t see cezanne launching before March unless AMD is changing they way they launch.

That means OEMs release in April at the earliest, with most designs coming in late 2021.

AMD COULD decide to launch desktop and mobile versions of Zen 3 at the same time. However, renoir is still mid cycle.

 

[eek2121]

So, in perusing the AT review, and looking at the graphs, it appears that 10sf still has a bit of a power/performance curve disadvantage vs. TSMC N7. It's not huge, but it seems to be there. However, Intel really seems to be able to wring the most out of 10SF in single core. Looking at SPEC ST, at "15"w, TGL still seems to be keeping an edge on AMD. I think part of that is related to the larger L3, and some of it is related to the IPC advantage that Willow Cove has over ZEN2. Looking at the die sizes, Renoir and 4C+Xe TGL are similar in size, but AMD is shoving double the CPU cores in there and still managing to stay close in the iGPU game, though ,with driver improvements, I think that Intel will get better faster than AMD.

I don't think that Cezanne makes a huge impact on N7, but maybe take advantage of better headroom on N7"improved" and keeps in the race in ST, and manages even better in MT. I am interested in if AMD makes any sort of improvement in memory controller efficiency with Cezanne, as, it does appear that Intel's better controller is helping their iGPU scores.

I wonder how Xe scales going down market? This was a 96 EU part with a high power limit. How does the 80EU version perform? Is the 48EU config performing at half the score of the 96 EU one, or is the scaling very sublinear?

That isn’t necessarily true. My takeaway is that Intel has a slight lead in efficiency. Note that you can’t simply say that they have half the cores so they should use half the power. That isn’t the case. Intel could push out an 8 core tiger lake u chip and the power consumption would remain the same, however the boost profile would be completely different.

EDIT: I wish I could get my hands on this chip.

IMO The only real way you can measure efficiency would be to cap the power consumption to a fixed value (say, 8W) that both chips could top out at, and measure the frequency for a single core in a worst case scenario.

 

[A///]

I think most people would love to know what AMD is up to in private. We're just around the corner to finding out how many times a day Bob goes to the toilet with the way Intel leaks, pun intended.

 

[TheGiant]

There is like a ~50pt difference between the 4800U and 4900HS is Geekbench 5 ST integer score, which means that there is no more headroom left in Renoir even if you double the power limits(in tests the U-series are limited to 30W and H/HS often reach 60W or more). Tigerlake on the other hand scores ~35% better in GB5 ST integer while pulling 50W. Granted that Intel is still not up to the mark in perf/watt, but it's ST perf is still very impressive.

exactly
and looks like 4.8GHz is very high on the v/f curve
4,6GHz needs 20W for it
the anandtech review shows quite monster ST performance for a U CPU
and the MT isn't that bad
Microsoft go release S laptop 4

 

[eek2121]

So, some further thoughts now that I'm at my PC. It seems some people continue to be confused about power consumption. Tiger Lake uses 'up to' 50W because Intel allows it to. The chip could certainly be capped to a much lower value. Coincidentally, you won't find a single Zen 2 core using over 11W.That is what allows for a much higher turbo boost. If AMD allowed their own cores to use more power, we'd see much higher boost clocks. Note that there is nothing wrong with Tiger Lake briefly using 50W of power. As you see from the excellent testing that Ian has done, things settle down for long running tasks. AMD's own chips peak out at around 35W, but they don't ramp down as aggressively as the Intel chips. AMD could just as easily allow their chips to hit 50W, but then you'd see a more aggressive downramp.

 

[Hitman928]

That isn’t necessarily true. My takeaway is that Intel has a slight lead in efficiency. Note that you can’t simply say that they have half the cores so they should use half the power. That isn’t the case. Intel could push out an 8 core tiger lake u chip and the power consumption would remain the same, however the boost profile would be completely different.

EDIT: I wish I could get my hands on this chip.

IMO The only real way you can measure efficiency would be to cap the power consumption to a fixed value (say, 8W) that both chips could top out at, and measure the frequency for a single core in a worst case scenario.

Anandtech did two efficiency tests, Tigerlake had higher efficiency when using y-cruncher with AVX-512 and Renoir had higher efficiency when running Agisoft (does not support AVX-512). If you were to test additional programs I think you'd see Tigerlake winning efficiency when AVX-512/NN extensions are used or in short bursty applications but Renoir winning efficiency for non-bursty loads that don't support the additional instructions in Tigerlake.

Intel’s Tiger Lake 11th Gen Core i7-1185G7 Review and Deep Dive: Baskin’ for the Exotic

www.anandtech.com

 

[randomhero]

Just read review on anandtech. Meh.
That is all I can say about tigerlake. Try harder Intel.

 

[eek2121]

Anandtech did two efficiency tests, Tigerlake had higher efficiency when using y-cruncher with AVX-512 and Renoir had higher efficiency when running Agisoft (does not support AVX-512). If you were to test additional programs I think you'd see Tigerlake winning efficiency when AVX-512/NN extensions are used or in short bursty applications but Renoir winning efficiency for non-bursty loads that don't support the additional instructions in Tigerlake.

Intel’s Tiger Lake 11th Gen Core i7-1185G7 Review and Deep Dive: Baskin’ for the Exotic

www.anandtech.com

That's an efficiency test of the overall product, not of 10SF itself. A somewhat proper efficiency test would require disabling cores on Renoir, however, AMD artificially caps Zen 2 power consumption, so that doesn't give us a real idea of efficiency. Because of this you'd want to run a test on 1 core at capped at a low wattage both can hit and see where frequencies end up for each chip. You would want to do this for regular and AVX workloads. Even in that scenario we aren't just measuring TSMC vs. Intel (core efficiency still comes into play), but it is as good as it gets. Based on what AT provided, Intel looks to have the efficiency lead. They just boost to insane levels.

 

[mikk]

In almost all gaming tests Tigerlake beats Renoir and often with quite a big margin.

Intel Tiger Lake And Xe Graphics Benchmarks: Laptops Will Roar - Page 2

Intel has a lot riding on Tiger Lake, with its Willow Cove cores and Iris Xe Graphics. Here's what to expect. - Page 2

hothardware.com

Golem.de: IT-News für Profis

Tiger Lake Tested: We Benchmark Intel’s Latest With Iris Xe Graphics and 10nm SuperFin

SuperFin and Xe Graphics make their mark.

www.tomshardware.com

Intel Core i7-1185G7 in Review - First Tiger Lake Benchmarks

We test the Tiger Lake Core i7-1185G7 with different TDP configurations and check the performance in numerous benchmarks.

www.notebookcheck.net

GPU-Benchmarks - Seite 3 - Hardwareluxx

CPU- und Gaming-Benchmarks: Ein erster Blick auf Tiger Lake.

www.hardwareluxx.de

Intel Core i7-1185G Tiger Lake im Test: Testergebnisse und Benchmarks

Tiger Lake im Test: Testergebnisse und Benchmarks / Single-Core-Leistung: ein neuer Rekord / Multi-Core-Leistung: AMD bleibt vorne

www.computerbase.de

Intel 11th-gen Core i7-1185G7 Tiger Lake Preview: It’s (mostly) faster than Ryzen

11th-gen Core i7-1186G7 vs. Ryzen 7 4800U: Who wins?

www.pcworld.com

In the hothardware test it beats even the 4900HS in gaming and on golem it beats 4700U LPDDR4x @35W by 46 percent in the 5 games they have tested. And this is with early drivers.

 

[Hitman928]

That's an efficiency test of the overall product, not of 10SF itself. A somewhat proper efficiency test would require disabling cores on Renoir, however, AMD artificially caps Zen 2 power consumption, so that doesn't give us a real idea of efficiency. Because of this you'd want to run a test on 1 core at capped at a low wattage both can hit and see where frequencies end up for each chip. You would want to do this for regular and AVX workloads. Even in that scenario we aren't just measuring TSMC vs. Intel (core efficiency still comes into play), but it is as good as it gets. Based on what AT provided, Intel looks to have the efficiency lead. They just boost to insane levels.

I'm confused, are you wanting to compare efficiency of the product or the process they are built on?

 

[Hitman928]

Interesting that Tigerlake and Renoir when both limited to 15W perform about the same with Renoir seeming to have the edge, but at 25W or higher, Tigerlake clearly takes the lead outside of some games it seems to stumble on due to drivers. Maybe due to AMD only using 8 CUs so it can't really increase performance that much even given much more power headroom? Intel having better memory bandwidth but is power limited at 15 W? Combination of the two?

Tom's also mentioned that Intel is not allowing any battery life tests, only performance tests which seems a bit, uh, odd to me for a mobile processor.

 

[Shivansps]

At 15W Tigerlake is still slower in games, compared to a CPU with double amount of CPU cores, that is not looking very good for a mobile device.

BUT, it is good to see Intel finally becoming competitive on IGPs... AMD has to realise now they cant keep using the same Vega IGP after 4 years.

 

[naukkis]

AMD slim test system throttles at higher power levels, Intel reference system has enough cooling capacity to maintain performance at higher power levels. Of course they deny battery tests, with those sustained power levels Tigerlake battery life would be terrible.

-> very biased testing from Intel and all hardware sites take that bait without any doubts.

 

[JoeRambo]

Brutal ST performance as expected from GB5 leaks. One has to wonder how well it would score with desktop class memory setup.