Discussion Intel current and future Lakes & Rapids thread

[Dayman1225]

Intel Posts Initial Open-Source Graphics Driver Patches For Icelake "Gen 11" Hardware - Phoronix

 

[mikk]

I'm curious about the changes from Gen9 to Gen 11 because there is a very long time between the first Gen9 product (2015) and Gen11 (2019), time will tell.

ICL 11 has a greater number of maximum subslices. This patch updates the
subslice max define to reflect this.

#define I915_MAX_SLICES 3
-#define I915_MAX_SUBSLICES 3
+#define I915_MAX_SUBSLICES 8

8 subslices for ICL would be nice. Gen9 has only 3 subslices (3x8=24 EUs @GT2). A fully working Cannonlake GT2 would come with 40 EUs (5x8), if there are still 8 EUs in one subslice ICL GT2 would come with 64 EUs, but this is just a guessing at this point without further infos.

 

[IntelUser2000]

I'm curious about the changes from Gen9 to Gen 11 because there is a very long time between the first Gen9 product (2015) and Gen11 (2019), time will tell.

The differences may be much smaller than you think. Without the delay, Gen 10 may have been 2016, and Gen 11, 2017. That may be why Tigerlake is planning on another update with Gen 11, to catch up with the loss at least a bit.

I don't know anything about Gen 11, but patches indicated few things about Gen 10.

Gen 10 seems to move to a setup where rather than having 7 threads per slice that applies to everything, it goes to a system where number of threads differ depending on what resources are being used. Probably a balance of area/power.

Update:

GT2 Gen 9 GPU has 8 ROPs and 12 TMUs. That compares with the 8 and 32 for Bristol Ridge. Raven Ridge has 32/40 for the Vega 8 and Vega 10, with unknown ROP counts.

GT3 is at 16 ROPs and 24 TMUs, and GT4 at 24 ROPs and 36 TMUs.

Prior to Gen 8(Broadwell) graphics, it had 2 subslices per slice. It made GT2 have 4 ROPs and 8 TMUs.

 

[mikk]

The differences may be much smaller than you think. Without the delay, Gen 10 may have been 2016, and Gen 11, 2017. That may be why Tigerlake is planning on another update with Gen 11, to catch up with the loss at least a bit.

Icelake taped in mid 2017 according to Intel. Not sure about Skylake but there were working engineering samples in the wild in H2 2014, so I would guess the design was finalized sometime in H2 2013, it's a very long gap.

 

[IntelUser2000]

Icelake taped in mid 2017 according to Intel. Not sure about Skylake but there were working engineering samples in the wild in H2 2014, so I would guess the design was finalized sometime in H2 2013, it's a very long gap.

You are talking about the plans after Intel realized they had to change their 10nm plans, and that Cannonlake may be DoA. Tape-Ins also come before Tape outs. If everything goes as planned, its generally said to take 12 months from Tape outs to see a product. Nowadays, it may be even longer because devices are so complex.

Originally, end 2016 would have been Cannonlake which makes Icelake end 2017. I highly doubt Intel was able to drastically change the architecture of Icelake. Delays happen because unexpected events happen. It's undesirable. Its when your plans get thrown off. When your plans get thrown off, you don't get to advance, you "waste" time trying to keep to the original plans.

 

[mikk]

i3-8121U is a 2+0 Cannonlake part, may be i3-8130U as well.

https://twitter.com/InstLatX64/status/957551712490934272


Geekbench results of the i3-8121U are available but it's not clear if there is a Turbo and how fast it clocks, so it's difficult for IPC comparions with Skylake.

https://browser.geekbench.com/v4/cpu/compare/5556402?baseline=6246666

 

[jpiniero]

I'd say it's possible that it is Whiskey Lake instead but it does look legit. It has the same model number as the very old Cannonlake result. What gives me pause is whether Lenovo would really sell a U laptop without the IGP.

Turbo clock is roughly 3.3 or 3.4 based upon the score.

 

[mikk]

Whiskey Lake is a 4+2 part and comes much later this year. Lenovo could sell this SKU with something cheap like MX130. Would be better for gaming than KBL-R with UHD 620.

Based on the scores the Turbo must be lower than 3.3-3.4 Ghz

https://browser.geekbench.com/v4/cpu/compare/6246666?baseline=5456924

The singlethread scores are constantly slower than KBL-R @3.4 Ghz apart from 3 outliner where the Cannonlake CPU µarch looks much better than Skylake.

 

[jpiniero]

In most of the tests it's only like 5% tops though, and that could be partially due to the extra L3 and faster memory. So maybe it's more like 3.2.

Besides the AES, the other two scores where the 8121U wins might be AVX-512 aware, which is why it doesn't help on MT at all since there's only one AVX-512 unit basically.

 

[mikk]

Besides the AES, the other two scores where the 8121U wins might be AVX-512 aware

There is no AVX 512 support in Geekbench 4 according to their workload pdf.

which is why it doesn't help on MT at all since there's only one AVX-512 unit basically.

Doesn't help at all? The score scaling is almost linear from one to two cores, there is a big advantage for AES and SFFT in MT as well. You may not be aware that in MT it is 2 vs 4 cores. SFFT are using AVX and AVX2 instructions sets, it seems to work much faster on Cannonlake.

 

[jpiniero]

There is no AVX 512 support in Geekbench 4 according to their workload pdf.

It was added in 4.1.

https://www.primatelabs.com/release/geekbench4/

 

[LTC8K6]

https://wccftech.com/intel-i3-8310u-cpu-specs-sheet-leak/

i3-8310U 2C4T 2.2/3.4 4mb

 

[John Carmack]

Are a lotta people really all excited about improvements to the iGPU?

 

[LTC8K6]

Are a lotta people really all excited about improvements to the iGPU?

Sure, gotta' catch up theoretically to those new chips with built in graphics from the other guys.

 

[mikk]

https://wccftech.com/intel-i3-8310u-cpu-specs-sheet-leak/

i3-8310U 2C4T 2.2/3.4 4mb

What i3-8310U? Laptopmedia were talking about the i3-8130U. The problem is that it is specified with UHD 620 graphics, in this case it isn't Cannonlake based.

 

[jpiniero]

What i3-8310U? Laptopmedia were talking about the i3-8130U. The problem is that it is specified with UHD 620 graphics, in this case it isn't Cannonlake based.

WTFTech is basically fake news, so I would take their specs as a grain of salt.

 

[mikk]

WTFTech is basically fake news, so I would take their specs as a grain of salt.

It's coming from laptopmedia not wccftech.

 

[IntelUser2000]

Based on the scores the Turbo must be lower than 3.3-3.4 Ghz

https://browser.geekbench.com/v4/cpu/compare/6246666?baseline=5456924

It may be, it may not be. The scores shown for the 8250U may be on the high end of the scale, and benchmarks scores can easily vary 20-30%.

There should be some perf/clock improvements because of the AVX-512 support. The widening of vectors come with expanded L/S units which help with general purpose code, although there will be diminishing returns with the size they are at now. And we don't know if there's any other changes.

The point is, even 10% general improvement can be hidden because of variation in scores.

Lenovo could sell this SKU with something cheap like MX130. Would be better for gaming than KBL-R with UHD 620.

I think Cannonlake would have been pretty nice, if they could have got it released the way it was planned.

Dual cores are still fine, especially with SMT support, and if it had an iGPU, Cannonlake may have resulted in laptops with record battery life for Intel Core platforms.

 

[Bouowmx]

I believe the i3-8121U results were obtained using 3.0-3.1 GHz. Compare with (best results using Microsoft Windows) i7-6500U (3.0 GHz, 4 MB L3) and i5-7200U (3.1 GHz, 3 MB L3).

Some food for thought: Why is the mystery processor named "8121", instead of just "8120"? "1" at the end is used to represent Cannon Lake? If so, that can explain why "8130" has UHD Graphics 620 (or that part about the "8130" could be wrong).

 

[John Carmack]

Sure, gotta' catch up theoretically to those new chips with built in graphics from the other guys.

I was told iGPU is good enough already and any better is pointless when you could/should get a dGPU MX130 or GT 1030 so it's surprising to find so many people in here salivating over more Intel EU's!

 

[Bouowmx]

How many layers of irony was that?

 

[mikk]

I believe the i3-8121U results were obtained using 3.0-3.1 GHz. Compare with (best results using Microsoft Windows) i7-6500U (3.0 GHz, 4 MB L3) and i5-7200U (3.1 GHz, 3 MB L3).

Some food for thought: Why is the mystery processor named "8121", instead of just "8120"? "1" at the end is used to represent Cannon Lake? If so, that can explain why "8130" has UHD Graphics 620 (or that part about the "8130" could be wrong).

The last 1 digit might be Intels new label for SKUs without iGPU, Cannonlake is the only option for such SKU. So it would make sense that the i3-8130U comes with iGPU. However if it comes with UHD 620 it is Skylake/Kabylake/Coffeelake based. Several sources say it comes with UHD 620.

https://ok2.de/media/ThinkPad/Datenblatt/ThinkPad X280 Datasheet.pdf
https://www.x-kom.pl/p/402173-noteb...e-5-i3-8130u-4gb-2561000-win10-mx130-fhd.html

 

[24601]

I was told iGPU is good enough already and any better is pointless when you could/should get a dGPU MX130 or GT 1030 so it's surprising to find so many people in here salivating over more Intel EU's!

Intel iGPU performance determines how good games look (talking about things other than console ports or AAA games) as it is the lowest common denominator.

Any significant increase in Intel iGPU performance means we will have a higher baseline graphics in PC games in general which will allow more room for developers to make their games better (obviously a good thing).

 

[LTC8K6]

I was told iGPU is good enough already and any better is pointless when you could/should get a dGPU MX130 or GT 1030 so it's surprising to find so many people in here salivating over more Intel EU's!

We still fondly remember IP6200 and long for the return of something similar.

 

[IntelUser2000]

We still fondly remember IP6200 and long for the return of something similar.

Iris Pro parts have mostly the same problem of graphics still sucking compared to discrete. Not to mention it had all the disadvantages of discrete graphics without the advantages of integrated. Expensive, poor battery life, used a lot of power.

That's why KBL-G is a far better idea. If its going to be expensive anyway, might as make it perform well. Really, I think KBL-G is 2018's Iris Pro.