DX12 and 6+ CPU cores: savior or irrelevant?

[Orangey]

IIRC a fair few people were falling short of the magical 3GHz mark! 3.8 on air must have been platinum!

 

[humbug]

I think an X5650 would Torpedo an 8 Thread Haswell in DX12, thats if you can find a Motherboard for it.

 

[setter]

IIRC a fair few people were falling short of the magical 3GHz mark! 3.8 on air must have been platinum!

There was a thread on here years ago for those who hit 4ghz 8 hour prime stable. Quite a few hit this magic number. I ended up swapping mine for a Q9550. 3.8 from it on 1.200v.

I think an X5650 would Torpedo an 8 Thread Haswell in DX12, thats if you can find a Motherboard for it.

I considered getting one to replace the 930 in my second system. P6 x58 de board. But holding off at the moment as I need the cash for an X99 setup to replace the one in sig.

 

[Kelt]

I think an X5650 would Torpedo an 8 Thread Haswell in DX12, thats if you can find a Motherboard for it.

Hope so!

Loving my X5650.

Chews up any multi-threaded game.

 

[jigger]

I'm sure DX12 and Vulkan will benefit from more cores in the right developers hands. The question for me is how many devs are upto the task and how long will it be before a decent amount of games show the performance.

If I was building a system just now, I would go with more mores cores over clockspeed even if it meant taking a small hit in DX11 performance.

 

[spoffle]

IIRC a fair few people were falling short of the magical 3GHz mark! 3.8 on air must have been platinum!

Really?! I had one that would hit 4Ghz on air. It wasn't 100% stable, but I could run a few bench marks at that speed.

 

[Silent_Scone]

2x factor at its best more often than not no difference.

You said DX11 is single threaded, which is a false statement. I corrected that statement, the legislation is there to be read in my post. If you meant solely draw calls, no matter how they're implemented add driver overhead deficit. This is why developers and even GPU vendors alike encourage batching of similar calls.

The limitation isn't simply from being thread limited. Effectively you still want as small a number of draw calls as possible. This is equally as important as efficient buffer management, not least of all constant buffer management.

API calls is the spin-sell for DirectX 12, there's a lot more groundwork that developers have been asking for that is equally if not more important than this within it.

 

[r22snapper]

IIRC a fair few people were falling short of the magical 3GHz mark! 3.8 on air must have been platinum!

I used to run a B3 at 3.2 on air. Used to warm the whole room up

 

[8 Pack]

Cores or no X99 is faster in games at the high end. Lashings of extra L2 help, as do various other things. Including the extra cores in some games.

So forget the DX12 argument, because it is rather silly unless you plan on keeping the rig 10 years and just focus on the fact that it's faster for same money.

Agreed !!!! Agreed!!!!!!!!!!!! AGREEEEDDDDDD!!!!!

 

[setter]

I used to run a B3 at 3.2 on air. Used to warm the whole room up

My GO was hitting 82c at 3.8, lapped it and the TRUE cooler I had, knocked it down to 68c max.

 

[Mauller]

You said DX11 is single threaded, which is a false statement. I corrected that statement, the legislation is there to be read in my post. If you meant solely draw calls, no matter how they're implemented add driver overhead deficit. This is why developers and even GPU vendors alike encourage batching of similar calls.

The limitation isn't simply from being thread limited. Effectively you still want as small a number of draw calls as possible. This is equally as important as efficient buffer management, not least of all constant buffer management.

API calls is the spin-sell for DirectX 12, there's a lot more groundwork that developers have been asking for that is equally if not more important than this within it.

Dx11 is single threaded when it comes to batch submission. but batches can be prepared on other threads then submitted to the gpu through the main thread.

Having the very low overhead on calls is as important as explicit memory and state control. Having low call overheads means developers don't have to jump through hoops to restrict calls and they can submit far more unique calls allowing them to create more unique scenes. This means calls/batches can be submitted faster and the code can be simplified.

Restricting the number of calls by batching is not an optimisation per say. It is a forced necessity for working within the api's limitations.

 

[Orangey]

The TRUE was the Corsair H100i of its day, everyone had one.

 

[Silent_Scone]

Restricting the number of calls by batching is not an optimisation per say. It is a forced necessity for working within the api's limitations.

It's not just the CPU this is applicable to. The more calls, the more driver overhead. As I explained - I'm not impeding the benefit of increased draw call limits

 

[Mauller]

It's not just the CPU this is applicable to. The more calls, the more driver overhead. As I explained - I'm not impeding the benefit of increased draw call limits

Oh yeah, but with the application now managing memory and state. the driver overhead is near non existent compared to what it currently is.

But the driver overhead is more redundant now when you can send 15 - 20 million calls compared to 600k - 2.0 million.

you can spend less time in driver batching up those 20 odd calls to render an object and instead send them directly to the GPU. which in the end will lower latency further. Or you can make many smaller batches overall if you wanted to try working in a lower call limit for lower end systems.

 

[humbug]

You said DX11 is single threaded, which is a false statement. I corrected that statement, the legislation is there to be read in my post. If you meant solely draw calls, no matter how they're implemented add driver overhead deficit. This is why developers and even GPU vendors alike encourage batching of similar calls.

The limitation isn't simply from being thread limited. Effectively you still want as small a number of draw calls as possible. This is equally as important as efficient buffer management, not least of all constant buffer management.

API calls is the spin-sell for DirectX 12, there's a lot more groundwork that developers have been asking for that is equally if not more important than this within it.

I didn't use those words at all...

The problem with DX11 is "Draw Calls" its the way the CPU communicates with GPU and this is very important, as it stands DX11 can only communicate through a single CPU thread, for some years now this is not enough, there is so much communication going on between the CPU and GPU that this "single thread" gets choked-up, the result of that is the GPU must wait on the CPU thread to catch-up, in waiting what the GPU is doing is slowing down, the result of that is lower FPS.

I don't know how you are understanding that ^^^ from what you are accusing me of but they are not the same thing.

Its also obvious that in that context i was talking solely about Calls, firstly talking about GPU communication and going on to say....

This is aside from other things the CPU does in games, like Physics and AI calculations which it can do across multiple threads.

Stop chasing me around the thread with this, its ridiculous.

 

[keyser]

DX12 isn't going to make GTAV any faster (boo).

I get how DX12 has less overhead, I don't get how it will automagically use all 6 or 8+ cores of the CPU unless the devs actually code it like that, in which case all 6/8 cores would have worked under DX11 too..

I get how DX12 could use the iGPU to boost frames a bit - of course that will mean hotter CPUs..

 

[sastusbulbas]

Personally, all I see is more tech, more claims, and most systems from the last few years still playing games as well as these new platforms.

X99, DX12, Skylake. Fact is, for current and gaming through 2016 and maybe more, none of them offer much if anything above the old tech from 2600k and upward. Waxing lyrical about a few fps here and there is nothing, as is PCI lanes when it's been shown to offer no tangible benefit currently with the majority of multiple GPU set ups of average users.

For me, from a gaming system perspective, with anything under 3 GPU's there is no point upgrading from an i5/i7/Z87/Z97 to any of the current new platforms. The FPS gains are usually single figure and not noticeable.

Skylake's M.2 SSD is a worthwhile boost in benchmarks, and Skylake should help drive down DDR4 costs, but M.2 at skylake speeds with DDR4 will not benefit average PC gamers enough to justify an upgrade for the forseeable future.

What I do hope to see is a drop in DDR3 and standard SSD's both of which have been overpriced due to demand, but are now clearly second generation and old tech.

Clearly DX12, and M.2 will have more to offer, but it will be a year or more before any advantage is made of them, and hopefully GPU's will have caught up by then, enough to show a big benefit with 16 PCI lanes over 8.

 

[Silent_Scone]

Oh yeah, but with the application now managing memory and state. the driver overhead is near non existent compared to what it currently is.

But the driver overhead is more redundant now when you can send 15 - 20 million calls compared to 600k - 2.0 million.

you can spend less time in driver batching up those 20 odd calls to render an object and instead send them directly to the GPU. which in the end will lower latency further. Or you can make many smaller batches overall if you wanted to try working in a lower call limit for lower end systems.

That's something that would have be to be tested so I couldn't comment either way.

I didn't use those words at all...



I don't know how you are understanding that ^^^ from what you are accusing me of but they are not the same thing.

Its also obvious that in that context i was talking solely about Calls, firstly talking about GPU communication and going on to say....




Stop chasing me around the thread with this, its ridiculous.

Nobody is accusing you of anything Humbug, you made a paragraph which said DX11 was single threaded, it's not obvious what you were talking about at all. At least with posting the legislation people won't have to take your word for it.

 

[MDPlatts]

The TRUE was the Corsair H100i of its day, everyone had one.

Still have mine in its box - beautiful object of desire.

 

[opethdisciple]

Not conclusive, but it seems more physical cores is the way to go, when it comes to DX 12 gaming.

---

In contradiction to the above if you look at this video review of Ashes of the singularity, you find the 6700k faster than the 5960x.