Forget Intel Haswell, Broadwell on the Way

Source even when mutlicore is enabled barely uses the second core...it is really sad

 

And then we have games like Flight Sim X that is ALL about single core speed. The "multi-thread patch" does jack and only dumps the non-essential background tasks to the other cores.

I distinctly remember someone saying to run Flight Sim X maxed out (especially w/ traffic density on max), you'll need a CPU capable of running 10GHz. I was just like "gg no re kthxbai" -_-

 

Heh. I remember when someone tried to play a SC4 city (loaded with 30GB of custom contents and mods) that had 8 million population with over 4 million commuters passing through the city from neighboring cities. It lagged even on a i5 2500K clocked at 5.2 GHz with two of its cores disabled.

I don't remember where to find that discussion thread though.

From what I've heard, SimCity 2013 is also single-threaded, and EA decided to severely limit the city size indefinitely citing performance issues with running larger cities. Not that I would want to buy that game anyways due to EA's exceptionally bad game launch.

 

Mantle and DX12 PLZ.

 

SC4 could have just used a better algorithm for its traffic to solve a lot of problems and traffic like that should be easy to task over several cores. It is a joke that they are so lazy with development like that

 

SC4 was released back in 2003, a year before AMD launched the first consumer dual-core CPU. Keep in mind that the "minimal" system requirements called for a Pentium III clocked at 500 mhz, 256MB of RAM, and a DX7 GPU (it directly mentioned Nividia TNT2, Matrox G400 and SiS 315).

It did not support anti-aliasing and simply used different texture models for each zoom in/out levels. It even supported software rendering, which transfers the graphics calculation from the GPU to the CPU if there are issues with the GPU. I believe SC4 was one of the last games to allow the CPU to handle all of the graphics processing.

EA attempted to reduce the traffic engine's load by setting the pathfinding to look for the most "direct" route instead of the shortest or more realistically, the fastest route. Which resulted in commuters using traffic jammed streets instead of an empty highway all because the highway was not the most "direct" route.

There's an mod called Network Addon Mod that make the pathfinding much more realistic at the cost of increasing CPU load.

On a side note, SC4 had major compatibility issues with Mac computers, until a partial patch was released in 2014. 11 years after launch, and 1 year after SC2013 was launched...


EA could've fixed the pathfinding issue with SimCity 2013, but from what I've read, the traffic engine is actually similar to the SC4's traffic engine in terms of the bugs. There were forum posts and videos about commuters ignoring empty roads, jamming up low-capacity roads, and in general taking seemly illogical routes. Seems awfully too similar to complaints about SC4's traffic engine...

I was expecting EA to do something stupid, but porting a decade old CPU-hogging programming function into a new software and only modifying it to support new features and a non-grid map is just laziness. And having to handle additional logic such as fire/police service only makes it even worse. Perhaps they were too busy trying to get the "always online" feature to work.

Although multi-core programming is tricky. All of the threads have to be in sync, and there aren't any programming languages that natively support multi-core processing. This is coming from a C programmer.


EDIT: Here's the Simcity 2013's traffic pathfinding failure, which is very similar to what 2003 Simcity4 suffered from: http://www.youtube.com/watch?v=g418BSF6XBQ

A SC4 version with refreshed graphics that looks cartoony, adjusted features, restricted city size, and a smaller library of custom contents?... Nah...

EDIT2: Cities XL (launched in 2009) also had problems with using only one CPU core and memory leaks. The original developer went defunct after causing an impressive backlash from trying to push players to buy an "multiplayer subscription" after charging $60 for the game. The current developer doesn't have access to the game's source codes, or it doesn't want to re-code the entire logic.

EA is still going because they're big enough to recover from any turd game launches.

 

Doubt if programs are enough smart to use cores like that. Hey Yo, you use that core I use this two cores?

 

You can manually force apps to use specific cores, which is what I'm assuming Loney111111 did. Windows is also quite smart in allocating tasks for the cores, depending on the settings (e.g. High Performance vs Power Saving), a laptop will usually prioritize Real cores currently active->Hyperthreaded cores -> inactive real cores.

 

my point was a better algorithm can be made and they can easily make traffic function in a different way. Why not make traffic algorithm for residential, industrial, pedestrian, and commercial? Each of them can be a completely different task that can be put on a separate core. You can even do sectors. There is no way that millions to trillions of calculations can't be rewritten in a more broken up manner. People in SC4 forum were like it is impossible to make traffic function better. The current algorithm is the best and can never be done better. It can never ever be split up no matter what....literally that is what they said and i call BS.

 

I used RadeonPro to force core affinity.

I think the people on the SC4 forum were saying that they can't directly modify the traffic engine because it's hard-coded and they don't have access to the source code. All they can do is make it "smarter" and add additional networks and functions onto it to allow players to build more realistic traffic networks.

Nobody is willing to spend the time to decompile SC4, and that would risk getting crushed by EA's lawyers if someone released an unofficial patch.

That's assuming the decompiler doesn't fail halfway through or Maxis programmers didn't add any hand-written assembly code.

I used to be on one of those SC4 forums. I read the Network Addon Mod's team's discussion threads before.

EDIT: I remember using a decompiler after a system crash wiped out my copy of source codes (200 lines), just leaving me with a .exe program. It was not fun.

 

Is there any concrete information on ~27-37W Quad cores in mobile Broadwell chips?

I was very interested in Broadwell because I thought ~27W-37W Quad-Core chips would become much more ubiquitous in 13" laptops. However, looking at recent news and the focus on the Y and U chips (M is dead?), it almost seems like the situation is going to be exactly the same as today with Haswell - meaning there might be a lower watt quad core but it'll likely be in a limited number of laptops.

 

yea no i explicitly (sp?) was talking on the forum on how i wish a modern day version one was made that allowed larger maps like 4-10 times the size and they went into how for a car that goes from block 1 has to pick from 256x256 choices. So they said a 1000x1000 map is impossible because the amount of places it would have to attempt to go is too hard. The stupid thing is that can easily be solves by forcing it to only select an option in 128 block range which just kept it limited to a very small location compared to 1000x1000 -_- That is only one option out of 100s of ways to make a better algorithm :/ They told me that you can never make a better algorithm and that would never work...this isn't talking about SC4 I am talking about if someone made an SC5. They were actually saying that you can never make a bigger map even with a new engine....morons I don't know jack squat about programing but i know that is utter BS.

You could easily make a set of rules that limit a car to X options. You could also easily make it multi-threaded and allow maps of 4-10 times the size with no issue with a little bit of care in designing some traffic rules. -_-

 

ArsTechnia: Broadwell is coming: A look at Intel's low-power Core M and its 14nm process
Anandtech: Intel Broadwell Architecture Preview: A Glimpse into Core M
The Tech Report: Intel's Broadwell processor revealed
Tom's Hardware: Introducing Intel's 14nm Node and the Broadwell Processor

Intel:

 

The thinner Z height on the broadwell package worries me, there is a small chance that the laptop H SKUs are pin compatible but will require new heatsinks.

 

Couldn't copper shims be used if the heatsink is too tall?

 

Aren't the H chips soldered (BGA package) anyway? Unless you feel like reballing you will pretty much have to get a new machine. Then there's always the question of chipset compatibility.

 

I think he meant the socket types.

 

The power improvements look promising for the thin-and-light segment. Ars reports that there's about 5% IPC improvement, similar to Sandy -> Ivy and Ivy -> Haswell. So, quite possibly not as exciting on the high end. Would probably be great if I wanted a 12" with insane battery life. But 5% IPC probably won't be enough to get me to leave Sandy Bridge, so I'll join Jayayess1190 in waiting for Skylake.

Actually, forget Skylake, Cannonlake is on the way.

 

Cannonlake is probably going to be even more delayed. Take a look a this timeline of Intel's prior releases: Intel Tick-Tock - Wikipedia, the free encyclopedia

Notice how it's progressively taking longer for each release. Laptop Broadwell won't arrive until spring 2015 since Apple won't be refreshing their laptop lineup until that time. Desktop Broadwell was rumored to arrive in fall 2015.

That means either Intel will also have to delay Skylake by a year, or release it alongside with Broadwell (and thus causing quite a bit of confusion).

 

And massive profit for Intel because <del>n00bs</del> general public

 

Finally Broadwell and Maxwell together. Alienware 13 is a small Broadwell and Maxwell gaming machine

Yupikayee

 

I see a massive fail coming due to the ULV CPU's... They should have put a full sized Haswell quad core like in the Clevo W230S and it would have been perfect...

 

Perhaps Intel is launching the ULVs first then the full-voltage ones?

After all, the first Broadwell chips are for tablets and smartphones.

 

Did people skip over this post?

Low performance U and Y crap come first, then soldered H chips come Q2 2015. No idea if socketed M chips will even be released.

 

Well, good news guys! I like the idea they shrinking the TDP packages and improving performance. I'm sure looking for to get a nice new Broadwell laptop, because 14nm is so exciting.


:thumbsup:

 

No thanks, not an another hairdryer again. Any hope the mobile maxwell GPUs will receive die shrink as well? Even if 860M today's most efficient GPU, I think it sucks a little too much power. At 20nm or lower would be better combination with Brodwell, that is my two cent .

 

That ULV chip is perfect, excellent for battery life yet powerful enough for most scenarios, the chassis is also sufficiently large for the heat to not be an issue (at least compared to ultrabooks). The reduced TDP simply allows more headroom for the GPU.
If you ask me, the W230SS is a little too overpowered on the CPU side with the full voltage i7s. Aside from the i7-4710MQ or weaker, most can only run the chip at full load for short periods of time before heat rears it's ugly head.

I would be happy with a ULV chip with zero throttling than a full voltage i7 that flirts with 95 degrees.

 

See:
AnandTech | Intel Disables TSX Instructions: Erratum Found in Haswell, Haswell-E/EP, Broadwell-Y

 

ULV CPU with unlocked multipliers would be a nice balance, and maybe quad-core ULVs 2-3 years from now when more software properly support quad-cores. From what I've read, the full voltage CPUs' additional idle power consumption is caused by the additional VRMs and power delivery circuits.

The HP Elitebook 850 allows overclocking of the ULV i7-4600U to 3.4 GHz for both active cores, making it almost matching the i7-4600M's 3.5 GHz. HWBOT also reported that someone reached 3.7 GHz with the i7-4600U: Intel Core i7 4600U @ HWBOT

If you go to the XTU link and look at the XTU hardware, it will say HP and the GPU as Radeon HD 8500M/8700M (the Elitebook 800s have the 8750M). The XTU setting says the multiplier is accessible, for an ULV.

The reasons why I didn't buy the Elitebook 800s is that they're very overpriced, and they throttle the 8750M severely at stock settings and operating temperature of 72C when a game such as Dota 2 is running, possibly to limit power consumption.

EDIT: If the W230SS had an ULV, I would've bought it since the battery life would've been significantly better (6 hours idle and 4 hours WiFi/video was not acceptable for me). The 62.6 Whr battery is larger than most ultrabooks' 50 Whr batteries, and such ultrabooks either come without a dedicated GPU, or they're more expensive than the W230SS.

 

i personally would dig a 25-35 w chip in that with unlock multipliers. 25-35 w is fairly easy to cool and gives great performance for the power. It also doesn't eat battery life any faster than a 17w chip unless on a load, which is good to me since i rather drain battery and have a usable computer Also the real issue with ULVs is the lack of single thread in my eyes you can get 4Ghz out of a chip with only 35w (IB). I would take a 35w chip that has a 4GHz clock any day. You would get 4GHz for single thread and 3 GHz on 4 with TDP throttling. I honestly don't knoww why they don't release chips like that to allow them to get max single thread performance. If my m17xR4 BIOS would work right i would force 45-50w TDP on my 3920xm since the cooling can only handle 45-50w :/ and give it a little turbo voltage boost so i could enjoy a 4.2-4.5 single thread speed and throttle to 3.5 GHz for 4 threads/cores. Right now i am stuck with 4GHz and a 55w TDP and setting ThrottleStop to throttle at 95C so it doesn't melt.

This is why i can't wait to get a desktop so i can have good enough cooling to get a CPU at +-5.5 Ghz. The single thread will be 50% faster than what i am currently using and will be so much nicer to use.

I am also hoping this die shrink again will allow getting single threads for laptops. IB was the first CPU that allowed 4GHz without some crazy cooling like an m18x with 3 pipes. If i had an m18x i could rock 4.5GHz with easy. Maybe broadwell will allow an m18x to rock 5GHz single thread.

 

LOL already done: http://forum.notebookreview.com/har...24059-forget-intel-broadwell-skylake-way.html

Maybe start thread "forget Intel Skylake, Cannonlake on the way" before OP beats you to it?

 

am on it ...

 

I don't think you have grasped the meaning of this thread...

 

I use a Lenovo X220 and have been since 2011.

 

well i did .. but forgetting each generations u end up with nothing .. then u can get amd products..

 

And no you didn't

 

well on other side u didnt get me too

 

Currently enjoying haswell over here btw

Sent from my Nexus 5 using Tapatalk

 

Holy Jesus. Have you guys seen what the desktop guys get to play with in September?
I know it is not Broadwell but look at this beast

8 cores @ 3.5GHz
20MB L2 cache
DDR4 support

All in a tiny little 355mm^2 package. Yep thats right, it is bigger than the GTX 880M
http://videocardz.com/51380/intel-haswell-e-i7-5960x-i7-5930k-i7-5820k-pricing-revealed

 

Ah, cloudfire, that's intersting, was gonna try to clarify & illustrate the significance of what you said, but realised I don't have the knowledge to clarify! What's the significance to comparing it against the size of the 880M? How much performance gain do you see vs the previous gen intel cpu?

 

It's really a meaningless comparison except to illustrate how monolithic a GPU is nowadays when you compare that to the die area and transistor count of something like a GK110 or Hawaii.

 

I was just amazed by the size of the thing. Its just a CPU. Its their first 8-core for consumers btw.

Die size:
880M 295mm2
5960X: 355mm2

Transistors
880M: 3.54 billion
5960X: 2.6billion

TDP:
880M: 125W
5960X: 140W

Another thing that makes me want to go to desktop. So much to play with there.

 

Actually I predict an epic fail for that octo-core 5960X. 3.3GHz stock turbo + Haswell means it'll probably be a struggle to get to 4GHz. Unless Intel deliberately underclocked it due to the heat, but in that case when you do reach 4GHz you'll need at the very least a good AIO to control the heat. Nevermind that very few games these days benefit from 6 cores, much less 8.

4790K might be the way to go for gaming if you don't fancy too much tinkering to get the most performance. 4.4GHz stock turbo is nothing to scoff at on a single core. (of course there you don't have much OC headroom since most of it had been used up at the factory already)

 

The new overclock edition of haswell supposedly solved a lot of the haswell failure with better transistors or whatever. I think it is out and posting good numbers. was on Techpowerup. also u have to have this chip if you want 3 plus gpus because of pci lane limits

 

If you mean 4790K yeah it runs about 6C cooler compared to 4770K. The absolute clockspeed is better simply because it's already clocked at 4GHz on 4 cores and 4.4GHz on single core from factory. But if you dig around most seem to top out around 4.8GHz, with 5GHz still requiring either ridiculous (ie definitely unsafe for 24/7) voltage or just plain unobtainable. So relatively speaking, the OC headroom is about the same or worse compared to 4770K, simply because most of it had already been used up at the factory.

I think if you really want to go Haswell-E and X99 get either the 5820K or 5930K depending on your setup. 5820K offers 28 PCIe lanes while 5930K has the full 40. If you don't ever intend to run 3x SLI/XFire setup then 5820K is a great deal because it's hex core and you can probably get it for under $400, assuming Intel prices it like it did the 4820K. If it wasn't for Intel deciding to force DDR4 on X99, I'd be building an X99 setup the day it came out lol.

 

Yeah but there's no easy way to implement DDR4 for mass consumption. Either way, there's going to be a huge price penalty initially. In fact, this is ideal since it gets the ball rolling so when they start fresh with Skylake, DDR4 is more ubiquitous.

 

Intel Core i7 "Haswell-E" Pricing Detailed | techPowerUp

its under 400 FYI...you should read techpower up more

Also as i was saying with the devils canyon or whatever its call, is that the actual chips didn't just come with better TIM. they actually had better circuitry. I tried finding the article on tech power up but couldn't find it. It came out a while ago. It might not have been called 4790k so thats why i might not find it via googling.

 

Yes it has more caps to "smooth power delivery to the die" according to Intel, but that still doesn't change the fact it OCs poorly compared to IB or SB. (most of the OC headroom is used up at the factory that's why its base clock is 4GHz and and can turbo to 4.4GHz)

 

AnandTech | Devil’s Canyon Review: Intel Core i7-4790K and i5-4690K

it appear 4.7 is the average. I must have read a single person getting 5 GHz on it. I thought a few people got that. Getting the various CPUs straight these days is getting hard. Maybe it was on the thread where this guy has a list of everyone overclocking haswell that i am getting confused with. A guy on there i know got 5 Ghz but he was the only one so maybe i am mixing the two things together. Read a lot of articles on Haswell/IB overclocking over the last 2 weeks so no surprise

 

Core M to become commercial brand for 4.5W Broadwell