Intel's Future Chips: News, Rumours & Reviews

Page 27 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
these deserve a separate post.

Linux 4.0 Will Run Source Engine Games Faster For Intel Haswell
http://www.phoronix.com/scan.php?page=news_item&px=Linux-4.0-Intel-Haswell-Patch
originally: Major Performance Breakthrough Discovered For Intel's Mesa Driver
http://www.phoronix.com/scan.php?page=news_item&px=MTgzMDk
begs the rhetorical question - how underperforming and unoptimized were intel's drivers in the first place that it took them this long (2 yrs after hsw launch)? 😗

More OpenMP Support Has Been Hitting LLVM's Clang
http://www.phoronix.com/scan.php?page=news_item&px=LLVM-Clang-OpenMP-March
Learning More About The Intel Vulkan Driver, Linux Vulkan Plans
http://www.phoronix.com/scan.php?page=news_item&px=LunarG-Vulkan-AMA
■The day The Khronos Group releases the Vulkan specifications is when they(lunarG) plan to open-source their Intel Linux driver.
■About 600 lines of code are needed to write a basic triangle program in Vulkan. However, much of that line count will be boiler-plate code that can be re-used or setup for the programmer by various libraries.
 
begs the rhetorical question - how underperforming and unoptimized were intel's drivers in the first place that it took them this long (2 yrs after hsw launch)? 😗

Well, Dolphin has very limited support for intel HD series GPUs because of known driver side bugs. So considering Intel doesn't even process OGL 3 properly...
 


Now, this is HD6000 vs HD5000. The Broadwell-K series receives Iris Pro level graphics. 20--30% from Haswell-graphics to Broadwell-graphics, more 40% (from Iris Pro update) would give 70--80% better graphics for the K-series. Doesn't this means that Broadwell-K will caught Kaveri?
 


Not necessarily. It's a good jump on paper, but you need more benchmarks that actually pit the two of them.

Cheers!

PS: http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/14
 


Thanks by the link, Haswell Iris Pro is behind Kaveri in some benchmarks and ahead in others

http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/12
http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/13

Broadwell-K would be Kaveri level.
 


Intel is limited by the cache they have for the Pro models. If they were able to slap 1GB inside of the 128MB they have today (don't know how much Broadwell has), they'd be 100% of the time ahead. But you can say the same thing for Kaveri. In terms of what "iGPU" is better, AMD is hands down better. I don't think this is earth shattering news, but my point is that Intel has to go with the brutally expensive solution to match/beat AMD in the APU game.

Well, in practical terms, if they do manage to get more memory on chip, then Intel will be ahead, that's for sure.

Cheers!
 


That Intel Broadwell comes with Iris Pro graphics was mentioned before



I think Intel will match Kaveri graphics performance.
 


Not to crash on your party, but... Notice the DDR3L 1600 support for socket LGA1150. So, take the leaks with a lot of salt. Plus, it doesn't mention anything about on-board L4 cache for the iGPU.

Cheers!
 


We have known for a while that socketed quad-core Broadwell desktop comes with graphics upgraded to Iris Pro level. Iris Pro 6200 include 48 execution units and 128 MB of L4. The mentioned DDR3L support looks like a typo from copy/paste info from the BGA section.
 


A bit premature unless the cost of Iris Pro comes down significantly.
The cheapest Iris Pro part is $434. This is competing with $170 Kaveri.
 


Doesn't the TDP increase imply something more? I expect 60% IPC increase over Broadwell thanks to AVX-512.
 

Isn't TDP (Thermal Design Power) essentially just Intel's indication of how much heat the processor will output in a "worst case scenario"? It doesn't seem like a TDP increase is likely to give any sort of good indication as to how much of a performance gain we should expect from the next generation of Intel processors.

1st Generation Core-i7 130W TDP
(Gulftown)

2nd Generation Core-i7 95W TDP w/ ~15%(?) performance increase
(SandyBridge)

3rd Generation Core-i7 77W TDP w/ ~5-15% performance increase
(IvyBridge)

4th Generation Core-i7 84W TDP w/ ~5-15% performance increase
(Haswell)

5th Generation Core-i7 65W(?) TDP w/ ~5-15%(?) performance increase
(Broadwell)

6th Generation Core-i7 95W(?) TDP
(Skylake)


I have definitely generalized the performance increases there, but none of the CPUs has brought more than a 20% performance increase over their predecessor. The TDP of the processors has fluctuated completely independent of all performance gain numbers. So why do you think a 46% TDP increase from Broadwell to Skylake is indicative of a similar performance increase?
 
^^ Based on what I read, it sounds like the bench is simply measuring how many draw calls the engine can handle. Performance increases in games would be an order of magnitude lower, since devs have been intentionally using as few draw calls as they can because of the bottleneck that existed there.
 
However if you are stuck on sandy you would see a decent increase in performance per clock over haswell not to mention haswell improved HT i actually see a 40% boost in multithreading performance with it on vs off. In area's that are extremely sensitive to single core performance i'm guessing you would see a 25% boost in single core performance from sandy to haswell.

The days of 2 to 3 times faster per core are over even with strict competition, my guess is there's not a lot left to do with X86 in that area.
 


The reduction here just seems to be due to a Intel's preference to focus on lower power at all cost.

Note the fastest Broadwell i7 is only 3.3/3.8 Ghz.
The Haswell line i7 topped out at 4.0/4.4 Ghz

The reduction is likely due to the FIVR which can't be cooled directly as it sits under the package.
Good for reducing z-height and better regulation at lower voltages.
 
Status
Not open for further replies.