Intel Broadwell (Gen8)

It's a little sparse on detail, but the GPU L3 and its system interface leverage virtual memory support and IO virtualization to provide shared virtual memory, which seems similar at a high level to GCN's introduction of x86-aligned page tables and its IOMMU to keep the GPU from stomping on things.

The global memory type seems to allow hardware-supported coherence between the GPU last-level cache and interface and CPU cores, in both directions, which on the face of it is significantly better than what AMD currently offers, or what seems to be in the pipeline anytime soon.
 
The global memory type seems to allow hardware-supported coherence between the GPU last-level cache and interface and CPU cores, in both directions, which on the face of it is significantly better than what AMD currently offers, or what seems to be in the pipeline anytime soon.
Even if the EU L3 Cache cannot cache lines from the coherent memory, it is certainly a more advanced implementation due to a system-level, shared LLC. But speaking of features, both Kaveri (Sea Islands?) and Gen8 are at the same level, it seems.
 
Even if the EU L3 Cache cannot cache lines from the coherent memory, it is certainly a more advanced implementation due to a system-level, shared LLC. But speaking of features, both Kaveri (Sea Islands?) and Gen8 are at the same level, it seems.

According to the presentation it can, which is why it appears to be better.
 
4.5W is what, 450% more than Apple A7?

Yeah, Arm's sooooooo dead. ;)
Really do you have a link for that . You kinda took me by surprize I not use to seeing this sort of thing on this forum . I thought this forum was above that sort of thing/:oops:
 
Keep in mind that 3DMark Ice Storm Unlimited performance is very CPU-limited. On Tegra K1, the fps in game test 1 is > 200fps and the fps in game test 2 is > 100fps, so this test is not particularly stressful on the latest and greatest ultra mobile GPU's.

The GPU perf. of this Core M variant seems to be within ~ 10% of the Core i5 variant in the Surface Pro 3. Naturally this is very impressive given the lower power consumption. But the GPU perf. difference between Core M and Tegra K1 may not be significant with a more GPU-intensive benchmark such as GFXBench 3.0:

65868.png
Hay I am aware of whats going on. But whats been shown so far is impressive. I new we would see tablets in OCt . But I was so hoping September. I just glad they are out . Intel 6 months behind on the release isn't a surprise to me I said at 22nm no way would intel have 14nm out on time . But I was thinking a year. Boy I hope Intel has 10nm figured out. and intel gets back on the time line . Tick is not bad tho for 8 years being off a year is better than all others in that period of time, I honestly didn't think the performance would be as good as what iot appears to be . The 4.5 tdp is impressive. Of course under heavy load it will use more watts . But so do arm soc . amd mullins really goes over TDP alot. Should have product reviews soon that will be nice to see what the guys have to say . Hot hardware is going a great job on IDF. Anand left for apple . But he in reality did that 3 years ago . We all seen his recent bios reviews on Apple. I tring to get my grandson to follow in anands Foot steps . He is 7 . and he is brillant . Last year he created a problem for the school district . I came right out and told them You have special classes for slow kids . I said were is the special classes for kids Like him . I be darned they listened . He was recently interviewd on TV. About the special phi classes just introduced this year. Last year he was in 1st grade this year he is in fourth . This kid is BIG I mean BIG not fat BIG. He never cried till he was 3 when he seen his brother cry . and get his way than he tried it . I put end to that in a hurry. He built his First PC at 6 . Of course I guided him but he got it right off the bat. So were setting up a Video set in my shop so he can do video hardware reviews for Kids . When intel sees and meets this Kid he will be set ,
 
Last edited by a moderator:
4.5W is what, 450% more than Apple A7?
... huh?

The global memory type seems to allow hardware-supported coherence between the GPU last-level cache and interface and CPU cores, in both directions, which on the face of it is significantly better than what AMD currently offers, or what seems to be in the pipeline anytime soon.
Right so LLC/eLLC sharing already happens in current parts (GPU page tables map to same physical pages as the CPU, etc), but BDW adds hardware to support shared virtual memory - i.e. using the same pointer/address.
 
I'm at the "The Compute Architecture of Intel Processor Graphics Gen8" session on IDF14 now...

Intel has published the "white paper" on the Gen8 architecture now... Here is a link to the PDF:
https://dl.dropboxusercontent.com/u...itecture of Intel Processor Graphics Gen8.pdf



Intel also published the Compute Architecture pdf. Already posted from pixelio.


However, the excellent Gen8 PDFs by Stephen Junkins just popped up in the last hour or two here.
 
Can't wait to see the top end Iris benchmarks vs Kaveri. And the fact that it seems to have taken HSA a step further as well looks like a double blow for AMD.
 
Cool, that confirms that it's a fanless design. I'm suddenly very interested in this. Never wanted a Haswell based system just because they all required fans for cooling.

That also seems like a far less exotic cooling solution than the magnesium used for the Tegra-K1 in the Shield Tablet. That seems to indicate it probably puts out less heat and thus less need for a more exotic cooling solution. Have there been any reports on how hot the tablet feels while in use and while benchmarking?

Regards,
SB
 
You'e gotta wonder what Intels capable of putting together if they decided to go all out on massive desktop gaming APU. Given that this dual Broadwell CPU + 24 EU graphics is well under 1/6th the die size of their latest 18 core Xeons and has 1/100th the TDP....
 
Are these processors (at least the notebook/sub-notebook variants) still using DDR3 @ 800 MHz?

While the efficiency improvements are nothing short of spectacular (my MacBook Pro Late 2013 can run Skyrim on its HD 5100 and with decent settings, too) 25.6 GB/s don't leave a lot of room for scaling resolution.
 
http://www.legitreviews.com/intel-core-m-5y70-broadwell-y-benchmark-numbers-run-at-idf-2014_150217

The Cloud Gate graphics score is almost as fast as a Surface Pro with an i5-4300U and 5x better than Z3770.


This is a chip I always wanted in my Dell Venue 11 Pro!
Now if only replacement model for Venue 11 comes with 4GB or better 8GB standard then I will upgrade :)

Really nice numbers and congrats to Intel for pulling this off.
I still hope AMD can somewhat compete to keep prices low, but without 16nm AMD is dead for tablets.
 
The performance of Core M is a big surprize for me @4.5 watts. For intels 10nm Phi I was thinking around 15-20 Gflops. Using broxton cores . Now I am thinking 30+Gflops using Skylake core m. I not real heavy into intel stock . But I am changing my position . Core M is a game changer . Not so much for what it can do right now . but it shows intels near future . If Amd is smart they will back away from their zen project its dead in the water befor it ever gets started. Amd can compete with present Atom server parts . But Zen is along way off. Core M is the warning sign for all others THIN ICE stay away . Apples A8 shows one thing and one thing only . This one many of us already new. 20nm isn't getting anyone a lot of performance. Witthout finfet these chips ARE HOT.:!:
 
is there any reason other than size that this doesn't have the 128 megs of eram ?

I would think it would help a lot power wise on these chips
 
Back
Top