Discussion Intel current and future Lakes & Rapids thread

Page 224 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,575
146
You might have noticed that the newer Renoir-U scores are quite a bit better than the leaks we got some weeks or months ago.

Fun fact. Renoir's iGPU scores roughly the same whether it has, say, a 35W cTDP and DDR4-3200, or a 25W cTDP and LPDDR4X-4266. The difference in scores is within margin of error.

I don't really think the configuration is limited by either of those, if anything the 8CUs@1850mhz is the real bottleneck for Renoir.

I'm fairly confident the Renoir numbers you see now are the best they'll ever get.

So then... what about Tiger Lake huh? How much of an uplift do you expect from the best score using LPDDR4X-4266 that comes in at just under 7% higher over these scores, and what do you think will cause such a drastic uplift?

DDR4 Renoir + 35W cTDP: https://www.3dmark.com/spy/11758228

LPDDR4X Renoir + 25W cTDP:
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
@uzzi38 I wonder how correct are the power settings?

On Picasso the monitoring applications do not show the full SoC power so its not exactly comparable to Intel numbers.

When you look at system load power numbers though, its comparable to the Intel systems. It's still too early to make a judgement. We need to see few systems reviewed by Notebookcheck. They are the only one thorough enough to show power consumption figures, and also screenshots load testing so we can see ourselves what the system is doing.

@naukkis I have some doubts on those numbers. The Cinebench for the Picasso is top notch. The lower scoring ones get in the 1100 range, not 1500.

The monitoring on most AMD systems are also pretty crappy. Some are ok but that's luck of draw on the system you get. You can approximate what power level the CPU is in by comparing the load average numbers from NBC.

The load average numbers are comparable between the 3580U Surface and the 1065G7 Surface. We know the 1065G7 is set at 25W for PL1 so that's likely the same for the 3580U.
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Was amazed with the results until i saw that according to videocardz was probably obtained with lpddr5.
When even small gpus like the nvidia 1030 took a huge hit in performance with ddr4 version vs gddr5.
Well rumours say that Intel removed LPDDR5 support from TGL so I don't see that as being likely
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
That's more like it.

The 48EU Xe iGPU is 40% faster than the 64EU Gen 11 iGPU. I bet the 32EU one in Rocketlake is fast as the 64EU Gen 11.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Looks like about 30-40% better than Icelake :tearsofjoy:

Dayman1225 is right, there is no LPDDR5 for Tigerlake, LPDDR5 has been removed earlier this year according to sharkbay. The fastest is LPDDR4-4266 they can use. Interesting to note that the i5 gets 80 EUs which is a surprisingly small difference, only 1 subslice disabled.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Look how "poor" the scaling is between EU counts. This is why increasing EU itself is nowhere near enough. It's also why low end configurations get a disproportionate gain.

Without its changes, a 96EU Gen 11 might perform like the 48EU Gen 12. 50% increase in EUs result in 20-30% perf boost.

In this particular test, the per EU gain is 2.1x. This is good considering I thought Gen 11 despite its advances were behind competitors in perf/area and perf/watt.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Look how "poor" the scaling is between EU counts. This is why increasing EU itself is nowhere near enough. It's also why low end configurations get a disproportionate gain.

Without its changes, a 96EU Gen 11 might perform like the 48EU Gen 12. 50% increase in EUs result in 20-30% perf boost.

In this particular test, the per EU gain is 2.1x. This is good considering I thought Gen 11 despite its advances were behind competitors in perf/area and perf/watt.
Well, bandwidth for the iGPUs aren't going up, so I'd expect the scaling to be pretty poor with more shaders.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
It's more than bandwidth. You have the fixed function units staying the same which will limit the improvement. More EUs use more power so it'll run into constraints. Faster graphics also run into CPU limitations, and will impact it a bit.

Lots of people expect 2x EUs, 2x TMUs, 2x FF, 2x bandwidth to be 2.5-3x the performance improvement, but in reality that just ends up being 2x as fast.

I expect a decent amount of gains coming from the much improved process as well. Icelake's 10nm must have been that bad. Typically low base clocks mean there's a fundamental problem either on the uarch or process level and the TDP targets are too low. This is why I keep saying yield issues are likely parametric as much as functional. You could have 100% chips working properly, but it'll be useless if it proves to be worse performing than the last generation.
 
Last edited:
  • Like
Reactions: Ajay

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
Notebookcheck is wrong in one important bit, there is no i3 with G4 graphics and 15W. There is a i3-1005G1 with 32 EUs which I think makes much more sense.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Interesting to note that the i5 gets 80 EUs which is a surprisingly small difference, only 1 subslice disabled.

i5 chips are quite a bit cheaper than i7 parts and also demanded much more by consumers. At 145mm2 its still a bit large for a mainstream part. 1 subslice disabled probably helps a long way helping production.

Oh, and its not NBC being wrong, but the source labelling their chart wrong.

The thing is though the gap approximates the difference between G4 and G7. G7 is way faster than 40% compared to G1. It's closer to 75%.
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136

RetroZombie

Senior member
Nov 5, 2019
464
386
96
Well rumours say that Intel removed LPDDR5 support from TGL so I don't see that as being likely
Ok. But you made me recheck the videocardz link, and AGAIN dual core and quad core cpus. o_O
Intel is really joking with all this, the 10nm is bad excuse no longer sticks.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Ok. But you made me recheck the videocardz link, and AGAIN dual core and quad core cpus. o_O
Intel is really joking with all this, the 10nm is bad excuse no longer sticks.

You have to remember that Tiger Lake U's design was probably completed like 2,3 years ago.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,575
146

5GHz boost would be quite an improvement if true. More than even I was predicting. Base clock seems a tad low compared to some of the other leaks though. Maybe 15W?
Likely a misreport. This chip failed several test runs, this is the only one that actually managed to complete.

Base clock is accurate. The chip in question is one of the early TGL-U samples.


This one to be specific. (It's 4GHz all-core boost btw).
 
  • Like
Reactions: pcp7 and lightmanek

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136

Seems the nominal TDP for Tiger-U is going to be 28 W. Also Gen12 has AV1 decode which is nice.