D3D12 is Coming! AMD Presentation

[soccerballtux]

Yeah,the SP0 version was.

even SP3 was miles faster than Vista/Patched Windows 7

 

[Azix]

XP was lighter. tetris is also lighter than dragon age inquisition. Who careth?

 

[RampantAndroid]

even SP3 was miles faster than Vista/Patched Windows 7

OK, so? You now have much more advanced security (I don't think XP even got SHA256 until SP3?), more features, better driver models, new things like TRIM and USB3 built in...and Windows runs just fine on a Surface 3 which is a pretty underpowered tablet, all things considered. I remember running Windows 7 on a netbook with my work outlook acccount on it and it ran better than Vista or XP did in a similar situation (my exchange account has ~50GB of mail on it consistently.)

Why on earth should we be stuck saying that Windows (or any software) be built to run on a 13-14 year old Athlon XP processor? Because, hint, we have software like that. It's called Firefox...only the newer Nightly builds actually up the compiler flags to something slightly more modern.

 

[frozentundra123456]

I payed $220 for my 4C/8T Xeon E3 at Microcenter :thumbsup: Sure not 8 real cores but I'm liking my decision more an more everyday, bring on DX12!

Or just wait an see what AMD releases for Zen cuz you better bet they will have 8+ core offerings. If Zen is as good as AMD is making it out to be an 8 core Zen might blow a 4 core i5 out of the water with DX12 games, and likely be close to your $200 price point.

If, and it is a big if, Zen is that good, no way they will sell it for close to 200.00. Just look what they tried to sell the 9590 for before the market spoke.

 

[AtenRa]

If you have 4 cores, and each core is 2-3x more powerful than each console core, you can put 2-3x as many threads on that core and see no degradation of performance (a simplified example, but mostly true).

Given the same overhead 6 threads running on 6 xbox one cores would run significantly faster on a Sandy Bridge 2500k overclocked at 4.2 ghz. Probably faster even at stock.

That is true only if you have something running in the CPU alone. In games that the GPU also do a lot of work, if your DX-12 game engine can scale to 6-8 treads, in order to feed the GPU more work at any given time, 6-8 slower cores will give you higher fps than 4 faster cores.
That is because you manage to feed the GPU doing more work per given time.

example, game engine 6 threads.

Fast Quad Core CPU,

Cycle 1 : CPU Thread - Thread - Thread - Thread
Cycle 2 : CPU Thread - Thread
Cycle 3 : GPU
Cycle 4 : CPU Thread - Thread - Thread - Thread
Cycle 5 : CPU Thread - Thread
Cycle 6 : GPU + CPU AI/Physics ---> Thread - Thread - Thread - Thread
Cycle 7 : CPU AI/Physics ---> Thread - Thread - Thread - Thread
Cycle 8 : CPU Thread - Thread - Thread - Thread
Cycle 9 : CPU Thread - Thread
Cycle 10 : GPU

Slower 6-core CPU

Cycle 1 : CPU Thread - Thread - Thread - Thread - thread - thread
Cycle 2 : GPU
Cycle 3 : CPU Thread - Thread - Thread - Thread - thread - thread
Cycle 4 : GPU + CPU AI/Physics ---> Thread - Thread - Thread - Thread -Thread - Thread
Cycle 5 : CPU AI/Physics ----> Thread - Thread - Thread - Thread - Thread - Thread
*Cycle 6 : CPU AI/Physics ----> Thread - Thread - Thread - Thread - Thread - Thread
Cycle 7 : CPU Thread - Thread - Thread - Thread - thread - thread
Cycle 8 : GPU

* Slower 6-Core will need Three cycles to finish AI/Physics calculations vs the faster Quad.

Quad core APU with mGPU support

Cycle 1 : CPU Thread - Thread - Thread - Thread
Cycle 2 : CPU Thread - Thread
Cycle 3 : GPU
Cycle 4 : CPU Thread - Thread - Thread - Thread
Cycle 5 : CPU Thread - Thread
*Cycle 6 : GPU + CPU AI + iGPU Physics ---> Thread - Thread - Thread - Thread
Cycle 7 : CPU Thread - Thread - Thread - Thread
Cycle 8 : CPU Thread - Thread
Cycle 9 : GPU

* iGPU will only need a single Cycle to finish Physics
Its over simplify but you get the picture.

 

[ShintaiDK]

There is still a scaling penalty. And it doesnt work that way.

Just look at the API test.

 

[TheELF]

If you have 4 cores, and each core is 2-3x more powerful than each console core,

That is true only if you have something running in the CPU alone.
example, game engine 6 threads.

Fast Quad Core CPU,

Cycle 1 : CPU Thread - Thread - Thread - Thread
Cycle 2 : CPU Thread - Thread
Cycle 3 : GPU

Slower 6-core CPU

Cycle 1 : CPU Thread - Thread - Thread - Thread - thread - thread
Cycle 2 : GPU

Yes let's look at that,
first of something very basic,cpus need hundreds of cycles to do the most basic stuff,that's why we have giga(giga=1 billion) hertz(one cycle per hertz) cpus.
Now let's say,for arguments sake,that you have a quad with exactly double the core speed of the sixcore.
While the hexa does 6*thread=6 units of work the quad will do 4*thread*2=8 units of work,in each cycle.

Slower 6-core CPU

Hundreds of Cycles : CPU Thread*1 - Thread*1 - Thread*1 - Thread*1 - thread*1 - thread*1
Hundreds of Cycles : GPU

Fast Quad Core CPU,

Hundreds of Cycles : CPU Thread*2 - Thread*2 - Thread*2 - Thread*2
Hundreds of Cycles : GPU

The quad will need ~30% less cycles to reach the point where it has enough stuff to start sending it to the vga.
And that's if there are actually 6 threads to be run all the time,if there are only 4 the hexa looses another ~30%.

 

[AtenRa]

There is still a scaling penalty. And it doesnt work that way.

Just look at the API test.

There is no scaling penalty here, if you have 6-8 draw calls simultaneously and you have 6-8 core CPU you will use 6-8 cores per cycle.

 

[AtenRa]

Yes let's look at that,
first of something very basic,cpus need hundreds of cycles to do the most basic stuff,that's why we have giga(giga=1 billion) hertz(one cycle per hertz) cpus.
Now let's say,for arguments sake,that you have a quad with exactly double the core speed of the sixcore.
While the hexa does 6*thread=6 units of work the quad will do 4*thread*2=8 units of work,in each cycle.

Slower 6-core CPU

Hundreds of Cycles : CPU Thread*1 - Thread*1 - Thread*1 - Thread*1 - thread*1 - thread*1
Hundreds of Cycles : GPU

Fast Quad Core CPU,

Hundreds of Cycles : CPU Thread*2 - Thread*2 - Thread*2 - Thread*2
Hundreds of Cycles : GPU

The quad will need ~30% less cycles to reach the point where it has enough stuff to start sending it to the vga.
And that's if there are actually 6 threads to be run all the time,if there are only 4 the hexa looses another ~30%.

You cannot have more than one Thread per Core per Cycle. You also assume that the faster Core will do 2x the work all the time. Not all threads need 2x the performance.

The faster Core does the same work vs the slower core in less time, it doesnt do 2x more work per cycle. In order to do more work per cycle you need an extra core.

 

[TheELF]

You cannot have more than one Thread per Core per Cycle.

I never said that you could.

You also assume that the faster Core will do 2x the work all the time. Not all threads need 2x the performance.

It's not a matter of need but a matter of if the vga(or any other component) will limit the core at a lower speed.
In you'r example you also assume that the hexa will always have 6 threads available.
Still the quad will definitely not be slower.

But than again we(headfoot was anyway) are talking about the 6 available 1,5ghz cores of a console,so twice the speed of that will always be welcome,even if not necessary(to reach the target 30FPS of the consoles)

The faster Core does the same work vs the slower core in less time, it doesnt do 2x more work per cycle. In order to do more work per cycle you need an extra core.

No you don't.
It's like reading a book,do you need an extra head to be able to read twice as fast as someone else?
No, you just need to be faster.
If a core has twice the speed it will (be able to ) do twice the work.

 

[PrincessFrosty]

"You buy a console and the games you see look awesome"

No by most objective standards they do not, juddery movement due to low frame rate targets and inconsistent frame times. Lack of high resolution and generally any decent kind of Anti-Aliasing, very low texture quality and almost no capacity for large scale physics.

"And they look awesome years after you bought it"

They maintain a static standard of quality relative to games developer by peers because the hardware cannot change and therefore nothing better can ever be released, only through relatively minor optimization due to understanding the hardware better can developers in future offer better games.

"And on the PC at the same time you're getting PC's that are way faster than the console and the games are still not looking like way better"

Er excuse me, I have a realtively medium setup with only a single GPU which isn't even the fastest single GPU you can get, and I still play my games in 4k all the settings cranked way higher than consoles. If 4k over 900p, much higher resolution textures, large scale physics, proper AA and texture filtering doesn't constitute "way better" than I don't know what does.

The vast majority of the reasons that PC games look close to console games is because developers are targeting the lowest common denominator which is the console and releasing largely the same game for both platforms, it has absolutely nothing to do with the lack of power on the PC.

Does he actually say anything sensible during any of this presentation? I switched off after 2 minutes, what an oaf.

 

[Paul98]

You cannot have more than one Thread per Core per Cycle. You also assume that the faster Core will do 2x the work all the time. Not all threads need 2x the performance.

The faster Core does the same work vs the slower core in less time, it doesnt do 2x more work per cycle. In order to do more work per cycle you need an extra core.

Sure you can't do more than one thread per cycle, but that doesn't matter as there are a huge number of cycles for even small operations.

It won't wait for one core to finish doing a thread before starting on the second thread. If you do two of the same threads on a single core both threads will finish around the same time. This will be similar as doing two threads on two cores, both will finish around the same time.

So in fact if you have two threads where one needs more performance than the other. and you only have one core, it will finish the one that needs less performance first, then will have it's full speed to finish the second.

Edit: And since you would be comparing 2 slower cores vs 1 faster core, the 1 faster core would actually finish those two threads quicker, the first one both would finish about the same time. Then the fast single core would finish the other thread quicker. Where as the two slower cores would be limited by it's second thread, where 1 core is idle while waiting for that thread to finish.( unless it can start another low performance thread again. and only then it would give similar performance to the single core.)

 

[BSim500]

Its over simplify but you get the picture.

"Perfect 100% thread scaling with a constant 6 threads per clock for FX-6xxx whilst an i5 will only have enough threads to load 4 cores half the time, and every instruction takes only one CPU cycle" is not merely over-simplified, it's completely backwards to the point of being totally broken. You only have to look at per core-loading on your average game to see overall CPU usage is often as low as 33-55% per FX core on hex & octo core CPU's (vs your mythical '100% on every clock' example). Overall that FX-6300 has total average 43% CPU usage, the FX-8350 has 46% and the FX-4300 has 68%, ie, a CPU which has barely half the IPC of a Skylake is still only being half-utilized, showing how bad thread-scaling is in most games after the first 4 cores. You think that's going to "bottleneck" an i5 giving FX-6300's magically higher fps because of +50% "MOAR CORES" when an FX-8350 vs an i5-6600K in 100% thread-loaded, core-utilized x264 encoding work out the same under 100% load vs 100% load scenarios? Any game which manages to cripple an i5 to the point of being unplayable by quadrupling the existing CPU workload will do exactly the same thing to AMD's current consoles, and FX & APU lineups.

Most PC games are being held back far more by consoles whose 8x notebook cores (6x usable) combined are barely on par with the slowest Haswell i3. If you're "waiting" for games to massively CPU-bottleneck an i5 all the way down to 20fps, then you're waiting for 2020-2027 era consoles (assuming they magically triple to quadruple the consoles current CPU horsepower, stick in a huge GPU that doesn't GPU-bottleneck that, all whilst keeping wattage low enough to not end up requiring full sized ATX PSU's in PC cases with huge 140mm CPU + Twin Frozr sized heatsinks or watercooling, and ending up a $700 / +400-500w console PC in all but name) at which point every current AMD FX chip would be equally obsolete. People can make mega-scale deliberately over-the-top DX12 games right now, but if they required such huge CPU power that an i5 could only manage 20-30fps, then current consoles would chug along at 10-18fps and end up "review bombed" for being "broken" on every single platform.

What you seem to be quietly desiring is intentional "proving a point" crippling of Intel quad's via wildly over-swinging from one design extreme (single thread draw calls) to another (draw call over-saturation) divorced from the reality of the target market install base. Great for a tech demo, but no-one's going to write such a game that won't work on i5's or it'll be a spectacular commercial flop given only 2% of the gaming market have 6-8 core CPU's. Two percent (and that figure includes Intel hex-cores). A serious reality check for the "FX-8350 will be the DX12 target design CPU of choice and i5's will become obsolete, any day now... any day now..." crowd. Even if you were correct, what developer is going to throw away +90% of their PC target market by waiting for a 300w 16x core AMD CPU to come out just to run his game at more than 25fps due to an obsession with novelty DX12 draw call spamming? That's sounds more like bad game design & epic target market research failure than a glorious "DX12 + MOAR CORE = Intel is dead" 'future'.

 

[TheELF]

Most PC games are being held back far more by consoles whose 8x notebook cores (6x usable) combined are barely on par with the slowest Haswell i3.

Lol no,it's two athlon 5150s and the 6 cores are hardly any faster than the slowest haswell celeron.
http://www.techspot.com/review/806-amd-kabini-vs-intel-bay-trail-d/page4.html

 

[shady28]

What DX12 brings to the plate is the ability to use more cores in games, it doesn't mean that they'll perform better.

Maybe more to the point, when you're in a game that doesn't stress your GPU the the quad core APUs still top out way before a dual core i3 :

But the APUs are able to use their cores more effectively and get to the point of being GPU limited when the game can push the GPU that hard :

Star swarm however is not necessarily reflective of all games.

RTS type games where you can have hundreds or thousands of units on screen at one time are going to get some particular benefit, because for each of those to be animated as an individual unit they require their own skeleton and draw calls.

This makes games like star swarm a particular beneficiary of DX12. I think most other games won't see anywhere near those performance increases.

It might even be slower for other games, because currently devs group things like a crowd of people milling around, a flock of sheep, etc into a single skeleton and make a single draw call. That's very efficient, as opposed to making 50 birds 50 draw calls which DX12 will supposedly enable.

 

[soccerballtux]

so even with D3D12 we're looking at CPU limitation on AMD

however, that's a pretty slow CPU compared to my 4.3/4.6ghz, so I think we'll, as AMD users, actually be fine.

additionally, it's possible that was only the critical section of non-parallelizable code in StarSwarm that was slowing down the AMD.
So, going forward, as long as AMD can hit 60fps in the game engine's critical section of code (aka the low settings in StarSwarm), AMD will be fine, as everything else can get parallelized.

 

[railven]

so even with D3D12 we're looking at CPU limitation on AMD

Current CPUs, Zen is a stone throw away and if that article I just read is any indication - they are moving closer Intel's designs and should be back in the running come 2016/2017.

 

[AtenRa]

I never said that you could.

You said Threads*2 per cycles. That means 2*Threads per core per cycle, which cannot be done.

No you don't.
It's like reading a book,do you need an extra head to be able to read twice as fast as someone else?
No, you just need to be faster.
If a core has twice the speed it will (be able to ) do twice the work.

Reading a book is a serial job, you need to read one page at the time, much like DX-11. So the faster your single Core performance the faster you read each Book page.
What DX-12 will allow is to simultaneously use 6-8 or more Cores (example with the 6 lights on the Video). It will be like reading 6-8 Book pages simultaneously but at a lower speed. That is way faster than having to read a single page at 2x times faster.

Also, you mistake the difference of work done in time units (example, over 60 secs) vs work done per CPU cycle.
Not only that, but you forgot that in games you also have the GPU. As was showcased in the Video, in DX-12 you can simultaneously use 6-8 or more cores to compute multiple draw call simultaneously to feed the GPU with more work per cycle.

That means that if you have 6 draw calls and each draw call needs one cycle, then the Quad core will need 2 cycles when the 6-core will only need one cycle.
Now, If the 6-Core CPU needs two cycles for each Draw Call (that means the Quad Core single Core performance is 2x faster), then the 6-Core CPU will also need two cycles to feed the GPU.

let me demonstrate,

Quad core 2x Faster per Core than 6-Core (Draw calls per Core per Cycle)

Cycle 1 : CPU ----> Draw Call - Draw Call - Draw Call - Draw Call
Cycle 2 : CPU ----> Draw Call - Draw Call
Cycle 3 : GPU

6-Core CPU

Cycle 1 : CPU ----> Draw Call - Draw Call - Draw Call - Draw Call -Draw Call - Draw Call
Cycle 2 : CPU ----> Draw Call - Draw Call - Draw Call - Draw Call -Draw Call - Draw Call
Cycle 3 : GPU

That is also very simplistic but i believe it is enough.

 

[destrekor]

what really surprises me is Microsoft not stepping up to bat to come up with an even more bloated OS.

Windows 7 is back to the speed of Vista, once you install the service packs. Rinse and repeat with W10...

personally, I'd be happy to have an XP64 install. XP was so fast...

xp up to service pack 1 was fast; service pack 2; made it about as fast as vista and 3 made it slower. *not that vista was terribly slow*

7 is faster than vista to a point; and 8 is faster than 7; same with 8.1. 10 is faster than 8.1.

but with repairs and services packs you do get some slow down

I'm not even certain what I am reading here.

How can anyone claim XP is faster than 7? Vista, sure, but Vista was faster than XP (any SP) in more than a few performance areas. Vista was a fairly-radically new kernel so there was some new bloat that needed to be addressed by both Microsoft and platform manufacturers/driver teams, as the driver stack was what saw the most change. Get the right drivers, Vista was plenty fast.

Right off the bat, Windows 7 was the fastest thing Microsoft released, save for maybe one or two test areas. Windows 8 continued that trend.

Sure, there were areas where XP remained faster than others, but often the scores were minimally different, to the point that the difference was not important or felt. Other times Vista even performed faster than 7, and plenty of times Windows 7, and then Windows 8, performed faster than everything else.

In short, in the ways that mattered the most, where the differences were largest and felt most readily, the 6.x kernel was king (with Vista, 7, and 8 swapping the lead in some areas), in very few areas did XP have a noticeable performance lead, as most XP benchmark wins were negligible, like 50fps vs 49.5fps. That's not an OS win, that's a driver likely not coded well for the newest OS at time of benchmark. You have mature OS with mature drivers versus newer OSes in most of these benchmark tests.

Anyone holding onto the notion that XP, any SP, was king... is delusional. Get out of the past just because you're afraid of a little change. Everyone made a huge deal about the changes Vista brought to the Windows platform, but developers have fully embraced and adapted to it at this point.

The only source of annoyance I have with the 6.x kernel is the fact that the audio driver model was changed so much that hardware audio effect processing was all but killed off. AMD is trying to get back to that but it requires a very specific approach that risks market fragmentation more than EAX had ever created, however, I doubt AMD's audio effects tech will be widely implemented, and, as it isn't vendor-neutral in that all PC users can't have it due to it being an AMD GPU solution and not an audio expansion board solution, it won't be reach the same market penetration.

Thankfully multi-threaded game developed is a thing these days, as the CPUs have the capability to handle the audio effects in their own thread pipelines.

 

[Paul98]

You said Threads*2 per cycles. That means 2*Threads per core per cycle, which cannot be done.




Reading a book is a serial job, you need to read one page at the time, much like DX-11. So the faster your single Core performance the faster you read each Book page.
What DX-12 will allow is to simultaneously use 6-8 or more Cores (example with the 6 lights on the Video). It will be like reading 6-8 Book pages simultaneously but at a lower speed. That is way faster than having to read a single page at 2x times faster.

Also, you mistake the difference of work done in time units (example, over 60 secs) vs work done per CPU cycle.
Not only that, but you forgot that in games you also have the GPU. As was showcased in the Video, in DX-12 you can simultaneously use 6-8 or more cores to compute multiple draw call simultaneously to feed the GPU with more work per cycle.

That means that if you have 6 draw calls and each draw call needs one cycle, then the Quad core will need 2 cycles when the 6-core will only need one cycle.
Now, If the 6-Core CPU needs two cycles for each Draw Call (that means the Quad Core single Core performance is 2x faster), then the 6-Core CPU will also need two cycles to feed the GPU.

let me demonstrate,

Quad core 2x Faster per Core than 6-Core (Draw calls per Core per Cycle)

Cycle 1 : CPU ----> Draw Call - Draw Call - Draw Call - Draw Call
Cycle 2 : CPU ----> Draw Call - Draw Call
Cycle 3 : GPU

6-Core CPU

Cycle 1 : CPU ----> Draw Call - Draw Call - Draw Call - Draw Call -Draw Call - Draw Call
Cycle 2 : CPU ----> Draw Call - Draw Call - Draw Call - Draw Call -Draw Call - Draw Call
Cycle 3 : GPU

That is also very simplistic but i believe it is enough.

Sorry, that is in no way realistic.

 

[RampantAndroid]

You cannot have more than one Thread per Core per Cycle. You also assume that the faster Core will do 2x the work all the time. Not all threads need 2x the performance.

The faster Core does the same work vs the slower core in less time, it doesnt do 2x more work per cycle. In order to do more work per cycle you need an extra core.

Not true? Hyperthreading works by duplicating sections of the CPU in hardware. Not ALL sections, but some. So some parallel work is allowed. And there are other dual issue CPUs out there.

 

[Enigmoid]

You said Threads*2 per cycles. That means 2*Threads per core per cycle, which cannot be done.

He is talking about the quad running at twice the frequency. Poorly worded but I believe he means that the quad runs effectively two threads per core per cycle (cycle of the half speed 6 core chip).

I have no clue why you are so interested by this. Games are dozens of threads and you simply do not need that kind of granularity.

Games generally have a large working set and scale well with memory. By and by most of the time threads are waiting on data. This is why hyperthreading scales so well on an i3 vs. pentium in some games and why broadwell's L4 cache makes such a huge difference in games. I can see where you are coming from but your analogy simply doesn't work.

Also, you mistake the difference of work done in time units (example, over 60 secs) vs work done per CPU cycle.
Not only that, but you forgot that in games you also have the GPU. As was showcased in the Video, in DX-12 you can simultaneously use 6-8 or more cores to compute multiple draw call simultaneously to feed the GPU with more work per cycle.

That means that if you have 6 draw calls and each draw call needs one cycle, then the Quad core will need 2 cycles when the 6-core will only need one cycle.
Now, If the 6-Core CPU needs two cycles for each Draw Call (that means the Quad Core single Core performance is 2x faster), then the 6-Core CPU will also need two cycles to feed the GPU.

let me demonstrate,

That is also very simplistic but i believe it is enough.

And here is the problem. Draw calls take hundreds of cycles. Look at the synthetic DX12 benchmarks. ~18M draw calls a second on a 6 core 4.2 ghz IVY-E CPU. Or around 10M per 2 cores at 4.2 ghz. This is around 840 cycles per draw call on average with efficiency decreasing as you add more CPU cores. 4 cores seems to max out at around 16M calls/s or around 1050 average cycles/call. Even on an application designed specifically to do nothing but drawcalls you are already seeing diminishing returns.

You want draw calls at the expense of everything else and that simply is not going to happen. No matter how efficient, draw calls are still going to chew through CPU time, especially if you up the number by an order of magnitude. Batches will still be more efficient than a ton of individual calls so batches will still be used.

Take a game that easily fills 4 CPU cores such as crysis 3. DX11 only uses 1 core for rendering. Thus it is immediately obvious that the other 3+ cores are doing game logic. Its also immediately obvious just how much game logic exists.

 

[TheELF]

You said Threads*2 per cycles. That means 2*Threads per core per cycle, which cannot be done.

Like I didn't explain it.

"While the hexa does 6*thread=6 units of work the quad will do 4*thread*2=8 units of work,in each cycle. "

if the hexa runs a thread at 100% load and you consider this 1 unit of work then the quad with twice the single ipc speed will do twice the work.

"Reading a book is a serial job, you need to read one page at the time, much like DX-11. So the faster your single Core performance the faster you read each Book page.
What DX-12 will allow is to simultaneously use 6-8 or more Cores (example with the 6 lights on the Video). It will be like reading 6-8 Book pages simultaneously but at a lower speed. That is way faster than having to read a single page at 2x times faster. "
Yes but the quad will not be reading only one page it will still read four pages simultaneously at twice the speed per page while the hexa will read 6 pages at halve the speed per page,You still end up with the faster quad reading ~30% more/faster.

Ypur opening argument wasn't that dx12 is better then dx11 but that a slower hexa will be better than a faster quad.


And you still cling to the notion that anything gets done in one cycle,this just doesn't happen.

 

[AtenRa]

Heh people dont understand, we are not talking only about how many draw calls each CPU can issue per second.
We are talking about command list(Buffer) and how much more work can we feed the GPU simultaneously per cycle.

Watch the video in the OP again.

also a few slides,

ps: This has nothing to do with AMD vs Intel (for those of you that mentioned it), its purely about Multi-Core CPUs and Throughput.

 

[Azix]

Are there other dx12 slides/presentations talking about these things besides AMD slides? Or are they just too technical. Some MS slides for example. Always an AMD logo on these things.

directX 12 presentations.