Hot Pants conference, Xbox One X reveal with Vega non-NCU hardware (LOUNGE)

http://www.anandtech.com/show/11740/hot-chips-microsoft-xbox-one-x-scorpio-engine-live-blog-930am-pt-430pm-utc

From Hot Chips conference in Cupertino

• Conservative occlusion query. Perhaps Conservative Rasterization.
• OOO (Out Of Order) Rasterization aka Rasterizer Ordered Views. Can boost transparency/alpha/blend ROPS performance and reduce shader workaround methods.

For reference

GCN 5 = Vega.

• OOO (Out Of Order) Rasterization similar to Rasterizer Ordered Views

https://msdn.microsoft.com/en-us/library/windows/desktop/dn914601(v=vs.85).aspx

Rasterizer Order Views

This enables Order Independent Transparency (OIT) algorithms to work, which give much better rendering results when multiple transparent objects are in line with each other in a view.

• Conservative occlusion query. Perhaps Conservative Rasterization.

https://msdn.microsoft.com/en-us/library/windows/desktop/dn914594(v=vs.85).aspx

Conservative Rasterization

Conservative rasterization is useful in a number of situations, including for certainty in collision detection, occlusion culling, and visibility detection.

From http://www.anandtech.com/show/11740/hot-chips-microsoft-xbox-one-x-scorpio-engine-live-blog-930am-pt-430pm-utc#post0821123606

12:36PM EDT - 8x 256KB render caches

12:37PM EDT - 2MB L2 cache with bypass and index buffer access

12:38PM EDT - out of order rasterization, 1MB parameter cache, delta color compression, depth compression, compressed texture access

X1X's GPU's Render Back Ends (RBE) has 256KB cache each and there's 8 of them, hence 2 MB render cache.

X1X's GPU has 2 MB L2 cache, 1 MB parameter cache and 2MB render cache. That's 5 MB of cache.

https://www.amd.com/Documents/GCN_Architecture_whitepaper.pdf

The old GCN's render back end (RBE) cache. Page 13 of 18.

Once the pixels fragments in a tile have been shaded, they flow to the Render Back-Ends (RBEs). The RBEs apply depth, stencil and alpha tests to determine whether pixel fragments are visible in the final frame. The visible pixels fragments are then sampled for coverage and color to construct the final output pixels. The RBEs in GCN can access up to 8 color samples (i.e. 8x MSAA) from the 16KB color caches and 16 coverage samples (i.e. for up to 16x EQAA) from the 4KB depth caches per pixel. The color samples are blended using weights determined by the coverage samples to generate a final anti-aliased pixel color. The results are written out to the frame buffer, through the memory controllers

GCN version 1.0's RBE cache size is just 20 KB. 8x RBE = 160 KB render cache (for 7970)

AMD R9-290X/R9-390X's aging RBE/ROPS comparison. https://www.slideshare.net/DevCentralAMD/gs4106-the-amd-gcn-architecture-a-crash-course-by-layla-mah

16 RBE with each RBE contains 4 ROPS. Each RBE has 24 bytes cache.

24 bytes x 16 = 384 bytes.

X1X's RBE/ROPS has 2048 KB or 2 MB render cache.

X1X's RBE has 256 KB. 8x RBE = 2048 KB (or 2 MB) render cache. X1X has hold more rendering data on the chip when compared to Radeon HD 7970.

http://www.eurogamer.net/articles/digitalfoundry-2017-project-scorpio-tech-revealed

We quadrupled the GPU L2 cache size, again for targeting the 4K performance."

X1X GPU's 2MB L2 cache being used for rendering in addition to 2 MB render cache.

For comparison, Vega 56 has 4 MB L2 cache which is directly accessible by RBE/ROPS and TMUs.

http://www.anandtech.com/show/10446/the-amd-radeon-rx-480-preview

RX-480 has 5.7 billion.

https://www.allaboutcircuits.com/news/microsofts-scorpio-system-on-chip/

X1X GPU has 7 billion transistors.

I layman's terms?

English please

Am I reading that right in the link that it will do 4K60fps game DVR? If so, awesome.

@endofaugust said:

More GPU hardware features to make graphics processing efficient.

@ronvalencia said:

@endofaugust said:

Jesus Christ Ron, no, just no... What are you trying to convey to us with this information?

ronbot 2.0

lol the closest translation I can come up with is x1x gpu is vega indeed? The reason is microsoft's latest and greatest dx12 features are supported by there new console matching vega+ gpus GCN 5. XB1X gpu is Vega take that eltormento? lol joking here of course.

Ron is just very happy that xb1x gpu supports latest and greatest dx12 features.

- 285 GB/s peak in the lab for mem bandwidth..

Hahahahaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa......

But but but but 326GB/s even more with DCC...

I guess the overhead of using a single pool of ram were even bigger in scorpio after all..

Note it doesn't say 285Gb/s for the GPU is says PEAK which mean total bandwidth for both,which mean even less than 285GB/s for GPU and which completely destroy any comparison you make with the RX580 which has 256GB/s just for it self.

Hey @goldenelementxl check that out...lol

Also note how FP16 is totally absent yeah you can bag the whole subword crap the XBO X lack FP16 double pumped support.

I wonder what you will do..

A-Admit you were WRONG.

B-Go into a rant quote 15 charts of totally irrelevant crap unrelated to what i am discussing.

C-Throw a Tantrum like always and tell me to try again even that i was right...

MS selected Vega related improvements that benefits the graphics pipeline.

Sony selected Vega related improvements that benefits compute pipeline. Sony is more focused on building their CELL ideology from AMD's IP blocks.

Both MS and Sony are taking care for their own interest but MS uses 3rd party 3D game engines such as Unreal Engine 4 and Unity3D.

285 GB/s peak in the lab for mem bandwidth is similar to my GTX 980 TI in pure CUDA mode without DCC.

It's slower for RX-470 and RX-480.

Without the memory compression, RX-480 has 194 GB/s.

285 GBps / 326 GBps = 0.87. That's 87 percent efficient which is superior to Hawaii's 82 percent efficient.

My MSI Gaming X 980 Ti has better GDDR5 chips hence it's higher than Techreport/Beyond3D's 980 TI's uncompressed numbers.

.

That in bold is as moronic and stupid as it can be,there is no such thing and Cell ideology buffoon,Sony simply pick something useful for them.

And the fact that use to claim that the damn xbox one X would exceed a damn 1080GTX only based on the fact that you BELIEVED that it would have FP16 double pumped and that with it Scorpio increased to 12TF really show how hypocrite and double sided your arguments are now the few vega improvements MS got are the best one even that FP16 double pumped is one of the most hype features of Vega.

Not only for double power for resources in process that run on FP16 like TResFX but does so without incrementing bandwidth usage.

Irrelevant what you 200 different GPU have for bandwidth.

285 GB/s peak in the lab for mem bandwidth..

What this ^^ means is that.

@ronvalencia said:

For Scorpio: (326 GB/s x Polaris memory subsystem efficiency 77 percent) x Polaris 1.37X memory delta compression = 343.90 GB/s estimated effective memory bandwdith.

This ^^^^^ is WRONG...

The xbox one X has 285GB/s peak even giving the CPU 20GBs that put the XBO with 255Gb/s actually 1GB/s slower than the RX480 and RX580...

This is from MS own documents is IRREFUTABLE....

You should learn to say i was wrong it would do wonders for you.

I seriously want to see a reality show starring Ron talking to people at parties. I would subscribe to cable TV for the first time in 20 years just for that.

To quote myself again

Vega NCU enabled Scorpio could gimp current PC GPUs without double rate FP16 feature.

The word enabled is conditional.

The word could is conditional.

Too bad for you, Polaris's double subword FP16 is still feature superior to GP104 (GTX1080/GTX1070), regardless of missing Vega NCU.

Vega 56/64 runs like larger scale PolarisPlus GPUs with current 3D game engines wasting Vega's double rate FP16 feature..

From 12:32PM EDT - 326 GB/s mem bandwidth

285 GBps / 326 GBps = 0.87. That's 87 percent efficient which is superior to Hawaii's 82 percent efficient. I was working on lesser Hawaii XT (82 percent) and PS4 (76.7 percent at 135Gbps from 176 Gbps) 's memory efficiencies. LOL.

285 GBps x 1.36 Polaris DCC boost yields 387 Gbps. hahahahahahahahahahah

Actually, 285 - 20 = 265.

255 GBps x 1.36 = 346.8 GBps hahahahahahahahahahah

265 GBps x 1.36 = 360 GBps hahahahahahahahahahah

RX-480 has 264 Gbps with memory compression and 194 Gbps without memory compression.

It's even better than my estimates.

Try again

Oh boy another RonTorm slap fight! This should be fun to skip over!

@Xabiss said:

Oh boy another RonTorm slap fight! This should be fun to skip over!

commence the multi-quotes

You're all wrong, here's why.

and with that the Xbox One X is the most powerful console on Earth.

Little less than 3 months and no more charts, graphs, or meaningless opinions.

I can't wait.

@kuu2: 20 bucks says your wrong. One or the other will post the same charts, graphs, and meaningless opinions proving the other right or wrong.

We will have actual games and pics. No amount of the other will matter. Proof will be in the games which is all that anyone should ever want.

285 GB/s peak in the lab for mem bandwidth..

PEAKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKKK...

285GB/s there is no 385GB/s bullshit.

I wonder what you will do..

A-Admit you were WRONG.

B-Go into a rant quote 15 charts of totally irrelevant crap unrelated to what i am discussing.

C-Throw a Tantrum like always and tell me to try again even that i was right...

So MS just admit that is 285GB/s PEAK (AKA no further bandwidth possible)

And you did what i say you will in Option C throw a tantrum and tell me to try again even that i am right and you were WRONG...lol

The XBO X doesn't have FP16 or subword crap it doesn't and not only MS state so in DF interview but hot chip presentation confirms its no talk about FP16 what so ever or even mention.

So save your damage control the Xbox one X lack FP16 period.

So what do you guys think of The Defenders?

@ronvalencia

IM THROWING YOU OUT OF MY PLANE, BIG GUY??

██]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]] 10% complete…..

████]]]]]]]]]]]]]]]]]]]]]]]]]]] 35% complete….

███████]]]]]]]]]]]]]]]] 60% complete….

███████████] 99% complete…..

?ERROR!? ?BIG? Guys are irreplaceable?

I could never be bigger than you, Bane!?

Send this to ten other ?Big Guys? who crashes your?plane?Or never get to call ittin☁again❌❌??❌❌

If you get 0 Back: Extra charge for them ???

3 back: You werent alone☁️?

5 back: You're banes hired gun???

Wrong

12:32PM EDT - 326 GB/s mem bandwidth

12:34PM EDT - 285 GB/s peak in the lab for mem bandiwdth

Lab's 285 GB/s peak is synthetic practical benchmark number just like my posted Techreport/Beyond3D suite's GPU memory bandwidth synthetic benchmarks.

That's 285 / 326 = 0.87 or 87 percent efficient which is better than Hawaii XT's 82 percent efficient.

Try again.

...SW was a mistake

This thread is the obvious front runner for thread of the year and it's not even SW related. And how Ron can come away from those slides saying, "yup that's Vega" is beyond me.

"Direct3D Optional Feature Support" slide sent Ron from 6 to midnight. Unfortunately it doesn't mean what he's saying it means.

My cat's breath smells like cat food.

LOUNGE RULES:

That means it has FIV and now so do you

I'm sorry

@ronvalencia said:

LOUNGE RULES

Dickiness quotient

Politics discussion

Good job

You posted some decent charts for once

MS/DF specifically stated PS4's double rate FP16 being missing from X1X. Polaris double subword FP16 version doesn't increase math operation rate.

To quote myself again

Vega NCU enabled Scorpio could gimp current PC GPUs without double rate FP16 feature.

The word enabled is conditional.

The word could is conditional.

Try again.

Too bad for you, Rasterizer Ordered Views is about out-of-order Rasterization.

https://msdn.microsoft.com/en-us/library/windows/desktop/dn914601(v=vs.85).aspx

Rasterizer Order Views

This enables Order Independent Transparency (OIT) algorithms to work, which give much better rendering results when multiple transparent objects are in line with each other in a view.

https://msdn.microsoft.com/en-us/library/windows/desktop/dn914594(v=vs.85).aspx

Conservative Rasterization

Conservative rasterization is useful in a number of situations, including for certainty in collision detection, occlusion culling, and visibility detection.

@ronvalencia: Please consider the following. I think this will show a pretty clear cut answer.

Seriously dude? It says no politics in the rules

What just happened?

@ronvalencia: I came here to see hot pants. There are none. I am severely disappointed.

Just let it happen

Get your camera, @charizard1605

Conservative Rasterization has nothing to do with politics.

Read https://msdn.microsoft.com/en-us/library/windows/desktop/dn914594(v=vs.85).aspx

Conservative Rasterization

Conservative rasterization is useful in a number of situations, including for certainty in collision detection, occlusion culling, and visibility detection.

"Conservative occlusion query" encapsulates Conservative rasterization's occlusion culling situations.

I don't approve of such government policies

@charizard1605 said:

@Zero_epyon said:

What just happened?

There is not a single meme in existence that can express how hard I'm LOLing right now...

lol ok I guess.

I think it's incredible. I also think it sucks that it was so short. I mean it was a decent length but one episode was only 38 minutes long. It felt a bit rushed.