Dayman1225
Golden Member
- Aug 14, 2017
- 1,152
- 974
- 146
Looks like Dell will also be offering the 28w i7-1068G7 on the XPS 2-1
Better IPC may not always be worth it due to diminishing return on that investment. Suppose you implement speculative execution on both sides of a branch instruction. Then you'll spend a bunch of transistors for exception handling and security to prevent malicious execution. So the chip will spend more energy per instruction on average and add a whole lot of complexity. Is it really worth it? What if you can spend the same logic to add a couple of extra DDR channels to expand memory BW or add more cache? There are plenty of other use cases for the logic/die area than merely squeezing everything out of IPC.
In most cases, observers/readers are willing to accept IPC tests at face value without worrying overmuch about how clock scaling will affect the measurement. This is especially true when dealing with stock configurations that won't vary much in clockspeed outside of testing.
IceLake in particular doesn't have much headroom. If someone manages to bench (for example) 3 GHz static clocked IceLake-U versus . . . I don't know, 3 GHz Picasso, I'm not going to pitch a fit since I know neither IceLake-Y nor Picasso will go much higher in clocks than that anyway.
Whiskey lake doesn't count for anyone interested in GPU because it doesn't have Iris Pro/Plus graphics.CPU speed seems to be roughly on par with Whiskey Lake, it isn't bad for a new process but definitively not groundbreaking. No wonder Intel is going for 14nm with Rocket Lake. GPU performance seems to be as expected a doubling for the 25W version.
Whiskey lake doesn't count for anyone interested in GPU because it doesn't have Iris Pro/Plus graphics.
An i7-8559u (as found in the NUC8i7BE) is a good comparison. It's 28W and compares to the i7-1065G7 in 25W mode.
CPU performance basically identical (link) 8559u ahead in some, behind in some, with very small margins.
GPU performance from the games measures there are variable, some ahead, some behind, with no more than a 10% advantage for the 1065G7 on average.
Note that the Ice Lake GPUs go up to 64EUs unlike previous which go up to 48EU, but they also miss the 128MB of EDRAM.
I'd estimate CPU performance the same, GPU performance 5% better based on the notebookcheck benchmarks, at a 10% lower TDP. Disappointing. They appear to have made very little progress and are covering it up - successfully as news outlets including AnandTech have taken the bait - by comparing the best new IceLakes to mediocre previous-gen processors, not to the best previous Intel processors.
This is an odd take. In games, the 25 W and 15 W 11th Gen Iris Pro perform essentially the same. The extra power budget goes to the CPU, not the GPU. As a result, you could have just as easily compared the 15 W Ice Lake to the 28W i7-8559u and concluded that at almost half the power it produces the same or better results (up to 30% better, for instance in Tomb Raider). And on the CPU side, the 25 W Ice Lake CPU is getting ~10% better performance in multi-threaded scenarios at 10% less TDP. That being said, the edram obviously gives the 8559u an advantage in select workloads.
Looks like to me the 15W and 25W configurations for Icelake only perform the same when the GPU is maxed out and the CPU is basically waiting most of the time for the GPU to catch up. In this case the SOC probably doesn't use more than 15W as the CPU is sipping power and the GPU cant' run any faster. If the GPU isn't an extreme bottleneck, the 25 W should perform better every time, just like the World of Tanks benchmark shows.
I feel like I said the same thing? The extra power goes to the CPU, not the GPU. The GPUs for the 15W and 25W configurations are exactly the same and appear to have the same power limits. Yes, if you are no longer GPU constrained, then the extra CPU oomph is going to show in benchmarks. My point was that using the comparison of the graphics performance of the 28W 8559u to the 25W Ice Lake to conclude that Intel hasn't made significant iGPU gains is extremely misleading when the 15W Ice Lake performs the same. It is very clear that Intel has managed to get a huge performance uplift in graphics. If Intel had made a new Iris Pro with edram, I assume it would have been quite the monster.
How would the 8559u perform in the same games at 20W (cTDP) compared to it's 28 W configuration though? My point is what you're saying about Icelake is most likely true about the 8559u as well though I haven't seen any benchmarks to confirm.
This is an odd take. In games, the 25 W and 15 W 11th Gen Iris Pro perform essentially the same. The extra power budget goes to the CPU, not the GPU. As a result, you could have just as easily compared the 15 W Ice Lake to the 28W i7-8559u and concluded that at almost half the power it produces the same or better results (up to 30% better, for instance in Tomb Raider). And on the CPU side, the 25 W Ice Lake CPU is getting ~10% better performance in multi-threaded scenarios at 10% less TDP. That being said, the edram obviously gives the 8559u an advantage in select workloads.
It depends on the game, in some games the difference can be big.
https://www.legitreviews.com/intel-10nm-ice-lake-benchmarks-on-the-core-i7-1065g7-processor_213374/2
Btw there are three games on Notebookcheck with the NUC and Iris Plus 655 against Icelake and in all three it's slower than Icelake 15W! Also a better comparison would be vs i7-1068G7, both have 28W.
But I do have high hopes for the Desktop/Server segment where power is less of an issue.
This is an odd take...
My point was that using the comparison of the graphics performance of the 28W 8559u to the 25W Ice Lake to conclude that Intel hasn't made significant iGPU gains is extremely misleading when the 15W Ice Lake performs the same.
Not exactly DT but it show a few related items :Keep your eye on IceLake-SP then. We won't see 10nm on the desktop for awhile.
Clearly the 10nm process for Icelake is not as efficient as the 14nm+++ process the latest gen lake's are on. Lower clock speeds handicapping the IPC gains from the sunny cove core.
But I do have high hopes for the Desktop/Server segment where power is less of an issue.
That makes no sense. It makes sense to compare the 28W 8559U to the 25W setting because the power is similar and the CPU performance is similar. If you compare the 28W 8559u to the Ice Lake at 15W you get similar GPU performance and much lower CPU performance. It's a much less clear comparison.
Not exactly DT but it show a few related items :
OK faster GPU but that's not relevant to the argument. You picked up on a red herring to extend this discussion but the appropriate comparison is obvious and fully explained in each of my posts. 25W to 28W is more similar than 15W to 28W, in power and in CPU performance and that's why the comparison is apt. I don't know why you are pushing this but if you want to reply on this point please see my previous posts for answers.Calling it "similar" GPU performance is exceptionally misleading.
what I am looking for is the ability to hold the clock
the problem with 8x/9x series of 15W U 14nmore+ mobile is the ability to not burst but hold the clocks, where it doesnt go above 3,1 GHz 4c8t
that icelake has all core boost of 3,5GHz and with that IPC uplift its like 4,1GHz skylake, the question is- can it run for more than a while at that frequency? without rocket sound fan ?
well I wanna see actual dell/hp/lenovo builds testedDoubtful. If you look at the Anandtech benches, when running the 3DPM benchmark it dropped to 1.0 - 1.1 GHz to stay at 15 W. This is an AVX512 load so it will use a significant more amount of power than a more standard load, but even when increasing the cTDP by 67% to 25 W, it only was able to get ~40% higher frequency (1.4 - 1.5 GHz). Obviously non AVX512 loads will use a lot less power / Hz but it seems doubtful that you will be able to get more than ~3x the all core boost rate and still stay under the actual 15 W TDP limit.
Their power test also reveals some tweaks intel made that (IMO) make Ice Lake look better than it probably should compared to current intel mobile CPUs. First, the test system was configured to be able to use up to 50 W of power (even in 15 W mode) for an extended PL2 window. As they mention, most laptops only allow up to 35 W or less. This lets the test system boost higher and longer than it will in most OEM configured laptops, especially any thin and light ones like you would see with a 15W CPU. Anandtech also noted that the test system's fan never stopped running which means it will take longer for it to thermally throttle compared to an OEM configured laptop.
These tweaks will have the biggest effect on single / lightly threaded workloads, especially ones that last only a short amount of time. Longer, well multi-threaded workloads will see the CPU spend most of its time thermally/power limited so you get a better comparison against the OEM laptops which will be in the same situation. In those tests (non AVX512 ones at least), the CPU perf/w comparison shows pretty minimal gains. GPU does show some nice gains in perf but I'll wait to see actual shipping systems with comparable configurations before trying to conclude how much of an increase it is.