Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [pcgames] Zen engineering samples in the wild. Units up to the 32 core are being tested.
New Posts  All Forums:Forum Nav:

[pcgames] Zen engineering samples in the wild. Units up to the 32 core are being tested. - Page 12

post #111 of 212
Quote:
Originally Posted by epic1337 View Post


on the other hand, they could also resort to using the big.LITTLE architecture in which they pair 2 big cores with 4 small cores, resulting in a 2C+4c/6T processor.

I really hope this happens later on for x86 especially as core count increases; dedicate 4 for primary.
post #112 of 212
Quote:
Originally Posted by TheBloodEagle View Post

I really hope this happens later on for x86 especially as core count increases; dedicate 4 for primary.

I thought big.LITTLE was to improve relative power efficiency because there was no good middle ground. Are you talking about x86 going into mobile devices, then?

Otherwise...don't we always big cores on desktop?
Quote:
Originally Posted by Sean W. View Post

Please wreck my 5960X

Amen.
The 2013 Build
(14 items)
 
  
CPUMotherboardGraphicsRAM
--- SOLD --- --- SOLD ---  --- SOLD --- --- SOLD --- 
Hard DriveOptical DriveCoolingOS
Samsung 850 EVO 500GB ASUS DRW-24B3ST (24X DVD burner) Noctua NH-U14S w/ NH-T1 TIM Windows 10 Pro (x64) 
MonitorKeyboardPowerCase
ASUS VG248QE w/ Lightboost @ 120Hz Ducky Shine III (MX Brown w/ Red backlights) Kingwin LZP-1000 (1000W) PSU Corsair Carbide Air 540 
Mouse Pad
Razer Goliathus (Precision Control model) - Sta... 
  hide details  
Reply
The 2013 Build
(14 items)
 
  
CPUMotherboardGraphicsRAM
--- SOLD --- --- SOLD ---  --- SOLD --- --- SOLD --- 
Hard DriveOptical DriveCoolingOS
Samsung 850 EVO 500GB ASUS DRW-24B3ST (24X DVD burner) Noctua NH-U14S w/ NH-T1 TIM Windows 10 Pro (x64) 
MonitorKeyboardPowerCase
ASUS VG248QE w/ Lightboost @ 120Hz Ducky Shine III (MX Brown w/ Red backlights) Kingwin LZP-1000 (1000W) PSU Corsair Carbide Air 540 
Mouse Pad
Razer Goliathus (Precision Control model) - Sta... 
  hide details  
Reply
post #113 of 212
Quote:
Originally Posted by Lex Luger View Post

God I hope these server chips are great and give Intel a kick in the nuts.

Intel can get away with charging 4000 dollars for their top of line Broadwell-E Xeons (which is the only chip in the new 14nm Xeon line that doesnt have cores disabled) because there is literally zero competition right now.

Broadwell -E desktop is in my opinion just horrible. The 6 and 8 core chips come from a 10 core die, and the 10 core chips come from a 15 core die, and they have the balls to charge 1700+ even though it clocks like garbage and has virtually zero IPC improvement compared to previous generation. To top it off, these are the bad chips they didnt make the grade for the Xeons, so we are getting nothing but leftovers chips. Thats why so many cores are disabled.

Zen being good will be best thing to happen in PC's in last 5+ years. With a 32 core chip priced fairly (2000 or less) intel with lose market share, and that will make me smile. People will support AMD if they have a good product.

I feel the same way as you buddy, I'm REALLY hoping AMD comes out swinging this year again like they did with the Fury X that used HBM.. Zen will make Intel rethink their overpriced 6950X prices, and I didn't even know that it was based off a 15 core die that just had 5 cores disabled, what a ripoff, and 750$ for two extra cores. Same thing for NVidia and their new Titan X, not that big of a difference compared to the old Titan and it's not even the full Pascal GPU, it's simply a renamed 1080Ti but for 1300$ or even more. So yeah, I'm really leaning towards AMD this year so I can take advantage of Freesync on a 144hz monitor opposed to paying an extra 300$ for NVidia's overpriced G-SYNC Module, if AMD never comes back, Intel and NVidia will continue wrecking people's wallets without a fear they won't buy their product.
post #114 of 212
Quote:
Originally Posted by ikjadoon View Post

I thought big.LITTLE was to improve relative power efficiency because there was no good middle ground. Are you talking about x86 going into mobile devices, then?

Otherwise...don't we always big cores on desktop?
Amen.

No, not for mobile. But I mean the first 4 cores would have no HT but be high clock (5Ghz) and then rest of the cores, say the 4 others in an 8 core would be 3 or 4 way hyperthreaded (similar to Power8 but less) at a lower clock.
Edited by TheBloodEagle - 7/23/16 at 1:59am
post #115 of 212
Quote:
Originally Posted by f1LL View Post

Baseline is: we need both for games. Highest IPC/IPS we can get and most cores we can afford.

The tricky part when buying is to find the balance between IPS/cores/$.

That balance is usually much more tilted towards higher IPS, fewer cores than most people, even tech enthusiasts realize. Amdahl's law is a harsh mistress.
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
Insert Name Here
(14 items)
 
  
CPUMotherboardGraphicsRAM
6700k Asus Maximus VIII Hero Gigabyte Aorus Xtreme 1080ti Corsair LPX 2x8GB 3200c16 
Hard DriveHard DriveCoolingOS
Old Seagate HDD Samsung 850 EVO Thermalright Silver Arrow SB-E SE Windows 7 Home Premium 64 bit 
MonitorKeyboardPowerCase
Asus PG258Q (240hz + Gsync) WASDKeyboards.com v1 semi custom w/ mx browns, ... Superflower Golden Green HX550 Air540 
MouseMouse Pad
Logitech G Pro Qck+ 
  hide details  
Reply
post #116 of 212
http://www.anandtech.com/bench/product/1289?vs=1549
almost everything benefits from more single thread
My System
(13 items)
 
  
CPUMotherboardGraphicsRAM
E6850 evga 680i a1 XFX 5770 11.7_Drivers_Aug15-2011) 4GB DDR2 800 
Hard DriveOSMonitorPower
320 GB SATA Windows 7 Ultimate x86 22inch display 1680x1050 750W Silverstone 
Case
antec 900 
  hide details  
Reply
My System
(13 items)
 
  
CPUMotherboardGraphicsRAM
E6850 evga 680i a1 XFX 5770 11.7_Drivers_Aug15-2011) 4GB DDR2 800 
Hard DriveOSMonitorPower
320 GB SATA Windows 7 Ultimate x86 22inch display 1680x1050 750W Silverstone 
Case
antec 900 
  hide details  
Reply
post #117 of 212
Quote:
Originally Posted by ikjadoon View Post

I thought big.LITTLE was to improve relative power efficiency because there was no good middle ground. Are you talking about x86 going into mobile devices, then?

Otherwise...don't we always big cores on desktop?
no, rather it was meant to tackle 2 types of load at once.
on one hand we have workloads that are very sensitive to single-thread performance.
on the other hand we have workloads that can effectively utilize a lot of cores, regardless of how slow each are.

the resulting combination of big.LITTLE architecture is that, they can provide for both types of workload, yet maintain high power efficiency and small die size.
if they had made an 8 big core processor, it would be a massive, power hungry processor that wouldn't be faster than a 4big + 4small core at that particular mixed workload scenario.


we don't always need lots of cores on desktop, hyper-threading shows us this, as it contributes to some gains despite not being as fast as the main threads.
which means, we can adopt the same approach, that hybridize the cores within the processor, using 4 powerful cores, and 4 high-efficient cores.
the resulting effect would be that, the die would be much smaller and cheaper than an 8 powerful core processor, yet provides us with the same amount of performance.
Edited by epic1337 - 7/23/16 at 4:18am
post #118 of 212
Or you can just query the processor for its core mapping and assign threads to cores yourself. Assigning a time critical thread to its own core and ignoring the hyper-threading logical core associated with it. Loading another and its logical cousin with multiple low priority threads.. etc etc. Unless we're talking about a closed system like a phone or embedded device, I'd much rather have as many fast cores as possible and distribute the load myself.
Edited by Lee Patekar - 7/23/16 at 6:57am
post #119 of 212
Quote:
Originally Posted by Lee Patekar View Post

Or you can just query the processor for its core mapping and assign threads to cores yourself. Assigning a time critical thread to its own core and ignoring the hyper-threading logical core associated with it. Loading another and its logical cousin with multiple low priority threads.. etc etc. Unless we're talking about a closed system like a phone or embedded device, I'd much rather have as many fast cores as possible and distribute the load myself.

its not that its impossible to do, but its a waste of resource.
if you open up task manager, you'd see one or two threads fully filled, while the rest are only at 80% or less load.

even DX12 is still like this, core0 has a 44%~80% heavier load than the rest of the cores.

Edited by epic1337 - 7/23/16 at 10:13am
post #120 of 212
Quote:
Originally Posted by epic1337 View Post

its not that its impossible to do, but its a waste of resource.
if you open up task manager, you'd see one or two threads fully filled, while the rest are only at 80% or less load.

even DX12 is still like this, core0 has a 44%~80% heavier load than the rest of the cores.

I'm not arguing that load can be perfectly balanced, that would contradict all my other posts. What I am saying is that a homogeneous CPU design with multiple identical cores, with hyper-threading, is preferable to a heterogeneous design with large and small cores. Its far simpler for software developers to design code that can be used on a wide range of hardware (from nehalem to skylake and to zen) than to optimize code for every processor family and architecture.

You can use specialized cpu architectures like 4+2 on mobile because its a closed software ecosystem. But in the PC space such a special-case processor will not be utilized correctly by older software nor by newer software. I feel this reality, in part, explains the failure of the bulldozer architecture. The intel compiler fiasco certainly didn't help matters.

For a general purpose CPU to run a wide range of software and games, you absolutely want the best single threaded performance across as many cores as you can afford. Perhaps one day load balancing in games will be much better (like the DX12 graph you showed) but alas the cynic in me expects most titles to be badly optimized when it comes to load balancing.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Rumors and Unconfirmed Articles
Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [pcgames] Zen engineering samples in the wild. Units up to the 32 core are being tested.