Build Thread 2.0

[Bartimaeus]

Ran into an old problem where in idle states (AMD Powerplay, the card clocks down to 300MHz on the core and 150MHz on the memory, down from 1000/1250MHz respectively), the VRAM will clock down too low and cause corruption (and possible black-screening) in 2D applications like an Internet browser. Fortunately I'm familiar with the problem and the solution - using a third party program like ASUS GPU Tweak or MSI Afterburner to raise the 2D clocks - but it's frustrating that there's no such inbuilt functionality in AMD's software.

That's only...sort of true. There's no way to do it via the actual control panel, but you can manually edit some text profile...uh, file somewhere or another in your appdata directory: I know this because I messed around with mine a while back trying to unlock lower clock speeds for my 3D clock speed settings, and found the 2D settings as well as the "half speed" settings...and it annoyed me that, for the 2D settings, the memory clock speed was set to 157MHz instead of just 150MHz (making my OCPD flare up a bit, ), so I changed that, too.

Edited July 29, 2015 by Bartimaeus

[Sarex]

So I forced the win10 upgrade in the end. I'll probably do a clean instal tomorrow.

 

So far I don't really have an opinion on win10.

[Bokishi]

I'll wait until drivers mature more. I hear Oculus on Win 10 is a mess

[Gorgon]

I'm still waiting on the update offer... Also it's bs that you need to update the OS first before you can do a clean install.

It basically is a new install and it's keeping some stuff from the old installation. I seriously doubt old programs are still gonna run. It's kinda silly even using the term 'upgrade'.

[Sarex]

It basically is a new install and it's keeping some stuff from the old installation. I seriously doubt old programs are still gonna run. It's kinda silly even using the term 'upgrade'.

 

It's a dirty install, that is for sure. As for old programs working, I had no issues and I suspect most people won't have any as this is basically just a new build of win8 + a reskin.

[Humanoid]

 

Ran into an old problem where in idle states (AMD Powerplay, the card clocks down to 300MHz on the core and 150MHz on the memory, down from 1000/1250MHz respectively), the VRAM will clock down too low and cause corruption (and possible black-screening) in 2D applications like an Internet browser. Fortunately I'm familiar with the problem and the solution - using a third party program like ASUS GPU Tweak or MSI Afterburner to raise the 2D clocks - but it's frustrating that there's no such inbuilt functionality in AMD's software.

That's only...sort of true. There's no way to do it via the actual control panel, but you can manually edit some text profile...uh, file somewhere or another in your appdata directory: I know this because I messed around with mine a while back trying to unlock lower clock speeds for my 3D clock speed settings, and found the 2D settings as well as the "half speed" settings...and it annoyed me that, for the 2D settings, the memory clock speed was set to 157MHz instead of just 150MHz (making my OCPD flare up a bit, ), so I changed that, too.

 

 

I tried that but for whatever reason the numbers wouldn't stick. Either I'm doing it wrong (not unlikely) or for some cards the idle states might be locked in the BIOS or something. I know some people have flashed custom BIOSes with increased idle clocks.

 

As for Win10, well, the upgrade icon isn't appearing on the one system I want to test it on. I have four other PCs running either Win7 or Win8.1 and they all have the option offered, but no such luck on the spare box. I've checked and double checked the prerequisites and fulfil them all, and have tried the workarounds to force the update (all three workarounds offered by the handy dandy batch file offered here), but no luck. I can't be bothered clean installing Win7 over again to see if that can force it to appear, since it's not a PC I'll be using for anything, so I'll wait until my build.

[Bartimaeus]

It's offered by Windows Update KB3035583 (I know this since I've had to remove the update from several systems). If that Windows Update is already installed, then I'm not sure what's up.

 

Weird that the clock speeds wouldn't stick: it did in my case, but my GPU at the time was the 5770, which may be old enough that that kind of mucking about was still functional.

Edited July 30, 2015 by Bartimaeus

[Humanoid]

Yeah, that batch file linked has as the first option a check that all prerequisite updates are installed. There's another update required depending on whether you're running Win7 or Win8. IE11 is also required for whatever reason. Either way, I pass that requirement. Windows version is correct and eligible (and activated), Win7 Home Premium OEM, gives me the all clear. Fix #1 should work instantly if relevant, nothing happened. Fix #2 says to run it and give it 10 minutes, nothing happened. Fix #3 says it may take an hour, I left it overnight, nothing happened. Ah well, my only options now are either to wait until MS provide an installation option not tied to the GWX app, or to nuke the current Win7 install and start fresh (the parts in this PC including the OS install are from my old HTPC).

[Humanoid]

So sometime Saturday afternoon I wandered outside of the house to throw out some rubbish and found that $1500 worth of stuff had been left by a courier on my front doorstep god-knows-how-long-ago. Pretty disconcerting, but on the plus side it meant I could get my build done over the weekend. The physical build was easy enough, but by god, the process of clean installing Windows 10, starting with an old Win7 Family Pack licence, was an ordeal of nightmarish proportions.

 

Intel i7-6700K

Scythe Mugen 4

Asus Z170-AR

16GB G.Skill 3000MHz DDR4

500GB Samsung 850 EVO m.2

Pioneer BDC-207DBK Blu-ray Combo

Fractal Design Define R4

EVGA Supernova G2 850W

 

Plus carried over parts:

Gigabyte Windforce R9 290X

256GB Crucial m4

250GB Samsung 840 EVO

256GB Sandisk Ultra Plus

Topping VX1 DAC/Amp (actually using the DAC portion now)

 

 

Need to go back and do some overclocking of the CPU and some underclocking of the GPU. Really wish Arctic Cooling would do a deal with a graphics card vendor to get their coolers factory-installed on high-end cards.

Edited August 17, 2015 by Humanoid

[Bokishi]

Really digging my new x52 Hotas, along with the 1tb Samsung Evo SSD + 6tb mechanical NAS drive (currently at almost 20tb storage now). Debating a Skylake upgrade in the fall, although I'm not sure how such an upgrade would benefit games running in VR or 4K (heavy gfx punishers), money might be better spent on another 980ti. Decisions decisions #firstworldproblems

 

[Gorgon]

So DDR4 and a better processor, but how much better. Is it worth upgrading from. I'm not sure I remember what kind of I7 i have. How much do you really need in the processor department anyway.

 

When is that fabled 3dram going to be stuck on a new gen processor. 

[Humanoid]

Honestly I'd still say that if you have an overclockable Sandy Bridge or newer, Skylake does very little for you, assuming you're comfortable clocking it to ~4.5GHz which is about the same clock speed Skylake typically overclocks to. The gains from 1st generation (Nehalem, 7xx to 9xx) to 2nd generation (Sandy Bridge, 2xxx) were the last time Intel made any significant gains between generations, and that switch happened in 2011. I'd been running an i5-750 so the upgrade was worth it for me.

 

That said, if your current CPU isn't overclockable (i.e. not a -K version), then Skylake may be justifiable purely for the raw clock speed increase, since those CPUs would typically be running at only a little over 3GHz. There are other minor perks for upgrading such as USB3.0/3.1 support, SATA3, m.2 slots, and somewhat faster boot times, but they're just that, minor perks. Dollar for dollar, throwing more cash at graphics cards will yield better results every time.

[Gorgon]

Sandra says I7 4770K @3,5 ghz.  Would only consider a processor only upgrade anyway and if it were cheap and it's new so that's not likely. Things are moving slowly on the GFX end though at least at my price point, which would be around an R9, a notch below the pricy headliners like FuryX GFti and Titan. 

[Humanoid]

If a mysterious benefactor sneakily replaced your 4770K with a 6700K, you'd never even notice a difference, it's pretty much a current product still. Clock for clock it's a single-digit difference in even the most CPU-bound tasks.

 

But yeah, the graphics market has been pretty glacial too, which isn't surprising given both sides of the duopoly have been stuck at 28nm for years, and tellingly products that launched in 2012 are still sold today. Fortunately that's changing in about a year.

[Bokishi]

I guess I be waiting for Intel Kaby Lake and Nvidia Pascal for next year then. Pascal is said to be an epic gpu jump that's 10x faster than anything that's out now, but we shall see

[samm]

I guess I be waiting for Intel Kaby Lake and Nvidia Pascal for next year then. Pascal is said to be an epic gpu jump that's 10x faster than anything that's out now, but we shall see

I don't know about Kaby Lake, but Pascal 10x faster? Yeah, sure :D 2x Perf/W would be nice, due to the shrink that is long, long overdue plus HBM. I'd not expect quite as much as 10x perf, in order to avoid being disappointed.

 

(edited for English)

Edited August 21, 2015 by samm

[Keyrock]

I guess I be waiting for Intel Kaby Lake and Nvidia Pascal for next year then. Pascal is said to be an epic gpu jump that's 10x faster than anything that's out now, but we shall see

1) Everything coming straight from Nvidia's (or AMD's for that matter) mouth should be taken with a grain cup bucket of salt.

2) The 10X performance claim is mostly applicable to compute, which will by far benefit the most from the expanded bandwith and 3D memory.  Gaming will likely see some benefit, but it will be a fraction of the benefit that compute will see.

 

Anyway, my i7-4790K at 4.7 GHz and Titan X with generous oveclocking headroom should have me set for years to come.  I'll think about an upgrade 3+ generations from now.

Edited August 21, 2015 by Keyrock

[Humanoid]

It's worth noting that with Intel, their last couple of die-shrinks have actually resulted in power consumption going *up* for the desktop chips and performance/watt has pretty much stayed the same. Frankly I expect sub-50% performance gains from the next generation, and that the primary improvement will be on the DX12 factor and not the actual hardware. In time, they'll be able to pump out big-die 16nm GPUs which is where the performance will come from - i.e. the brute force approach of more transistors - but realistically, the first generation on that node will consist of comparatively smaller chips, probably with more or less the same transistor count as the current high-end.

 

Kaby Lake at the moment sounds like it'll be an improvement on the integrated graphics of Skylake and bugger-all change in the actual CPU part. A hex- or octa-core Skylake-E would be the thing to wait for I imagine if you really have an upgrade itch, but it'd only be of benefit for playing games that scale with number of physical cores. To be honest, I reckon any Haswell or Skylake i7 will easily last another 4-5 years.

[Bokishi]

Well Intel is ditching the tick tock schedule with Kaby Lake and Canonlake, so that could mean big improvements (or none at all!) from Skylake. A 6-8 core Skylake I expect coming in the next few months also sounds enticing. But I do agree we might be at the point where substantial performance boosts from processor upgrades are over. A far cry from Pentium 4 ---> Core 2 or Nehalem ---> Sandy

[Keyrock]

It would certainly help if AMD was actually relevant again on the mid-high end desktop CPU front (I miss the Socket 939 days *sob*).  What motivation does Intel have to sink money into r&d toward serious IPC gains when they're essentially competing against themselves?

[teknoman2]

i watched a video making a performance comparison between a combination of AMD and intel CPUs and a 390X and GTX980, by using the beta version of a game that is made with DX12 support

in DX11, the CPU was a major factor and the 980 had often almost double fps compared to the 390X

in DX12, CPUs did not make much difference in the fps and the 980 got a 2-5 fps increase, while the 390X got from 40 to 120% more fps, getting overall the same fps as the 980 (the margin was +-2 fps compared to the 980 accross all combinations)

[Keyrock]

Supposedly DX12 and Mantle are very similar, so I guess it makes sense that AMD would initially be able to squeeze more out of DX12, since they were the ones that created Mantle, kind of giving them a head start on DX12.  Also, since Vulkan is also heavily based off Mantle (in a way Mantle became Vulkan), that gives me hope that the similarity of the 2 APIs makes porting from DX12 to Vulkan (and vice versa) much easier than from earlier versions of DX to OpenGL.

[teknoman2]

well one thing that test proved is that AMD does not have inferior hardware... the limitations were in the software

[samm]

Supposedly DX12 and Mantle are very similar, so I guess it makes sense that AMD would initially be able to squeeze more out of DX12, since they were the ones that created Mantle, kind of giving them a head start on DX12.  Also, since Vulkan is also heavily based off Mantle (in a way Mantle became Vulkan), that gives me hope that the similarity of the 2 APIs makes porting from DX12 to Vulkan (and vice versa) much easier than from earlier versions of DX to OpenGL.

Interestingly enough, AMD might be the last to the game with Vulkan: Intel were the first with drivers promoted on the web, because their Vulkan driver development was outsourced to another company and that was used for much (all?) of the Valve Vulkan preview demos. nVidia are just now starting to promote their drivers to be ready with the first Vulkan releases. AMD has been silent so far, as far as I've read, about their Vulkan drivers.

 

On DX12 vs. Mantle: http://nextgenapis.realtimerendering.com/ Check the second presentation (Next-Generation Graphics APIs: Similarities and Differences, by Tim Foley of nVidia). However, the comparison is still quite high level. I guess that many basic principles used in Engines with Mantles can be applied to the same engines using DX 12 or Vulkan, however there are differences in all the surrounding stuff. Take Vulkan's flexible layer architecture for example, or its shaders SPIR V which again can theoretically be compiled from any kind of source, even if there are only Open CL and GLSL compilers currently, as far as I'm aware.

[Fighter]

390 is tempting at the cheaper price but the power consumption...