The PS4 Pro is a complete failure

AF is poorly implemented in quite a few PC games too, particularly GTA V and Assassin's Creed from my experience. I force it via Nvidia Control Panel in most games.

At least Horizon: Zero Dawn is a good example of Pro support done right.

As for third party, The Surge is a good example.

But from now on I'll definitely wait for port reports, so I know how each version stacks up.

It really is. I mean if they were going to do a mid-gen upgrade, they should have waited and went big. But nope, they let MS handle them in that department. With that said, I think there's still plenty time to really fully use the Pro. While Pro enhancements may not be that stellar now, it could only get better in the future. But then again, the Scorpio will always be steps ahead. Sony might need to release a next gen console instead.

PS4 Pro APU size = ~321 mm2

Scorpio APU size = 360 mm2

12 percent difference in BOM cost.

MS selected improvement path similar to Nvidia's Maxwell i.e. reduce bottlenecks/lower latency (e.g. L2 cache rendering connection) and increase memory bandwidth. Polaris inherits double rate FP16 data elements processing.

Sony selected improvement path similar to classic AMD's high TFLOPS PR i.e. increase TFLOPS via double rate FP16. Polaris' Pixel Engine not connected to L2 cache.

I dont think he is refuting that Scorpio is superior, I think he attacking Ron's claims based a million slides and an example of console port of a specific game as "proof". Forza on Pc should not be used as an example of what to expect. Fact is that the game itself pegs two threads to 100% and the rest are underutilized past 4 threads. Which means that game itself at any resolution is not feeding the gpu the best it can. making the argument about gpu bandwidth moot.

Also Ron was wrong in the assumption that Scorpio's gpu would be VEGA and have double 16FP ability, giving it immense efficiency and processing power advantage over PS4 pro.we can not ignore Scorpio from being gimped by its cpu in the end, even with it having a dedicated command processor for the gpu queuing that command processor still has to wait on the cpu to get its tasks, so think of it as a buffer to not to stall the gpu as much as it would have without it.

I think we can expect Scorpio's gpu performance around a RX 580 which isnt far off from a GTX 1070, but again between the cpu being a jaguar or puma based architecture at 2.3ghz and that 8gb is shared between game and vram Scorpio is not 4k ready system, being able to provide the performance a detail to take 4k's true might.

PS4 Pro hardware was designed for Sony's processing model i.e. focus on 8 ACE units/Compute Engine and L2 cache tiling path which is a continuation from CELL's 6 SPU+local on-chip memory processing model. Pixel Engine path is just an afterthought i.e. they didn't care about Pixel Engine not connected to L2 cache bottleneck. Sony selected Vega's double rate FP16 over substantial memory bandwidth increase/bottleneck reduction (e.g. Pixel Engine connected to L2 cache) improvements.

Nivida Maxwell/Pascal GPU has Pixel Engine connected to L2 cache tile processing model but it can handle Compute Engine connected to L2 cache tile processing model. CryEngine 3 and Unreal Engine 4 has heavy Pixel Engine path usage.

Prey is powered by CryEngine 3. Nvidia optimized their Maxwell/Pascal hardware for major 3rd party 3D engines.

Without Pixel Engine being connected to L2 cache, effective memory bandwidth bottleneck will be the big factor.

Both MS and Sony started from Polaris baseline IP blocks and both leadership team selected different subset Vega IP blocks.

Exactly and just so he knows you and me almost never eveeeeeeeeeerrrr agree on anything,just so he doesn't think that you are covering for me or any crap like that.

I told Ronvalencia about using forza but you know how he is,he was wrong about Ryzen to being inside scorpio but what really made me mad was hes argument about FP16,he even claims scorpio would exceed a 1080GTX using FP16 because at FP16,Scorpio was 12TF,but when it was confirmed than scorpio didn't have FP16 double pumped,and the Pro did i turn the argument against him.

Since the Pro is 4.2TF with FP16 it was 8.4TF and would exceed scorpio,he went insane and started inventing excuses on why FP16 would not work for the pro ..Hahahahaa

Even that i was messing with him,in fact i was one of the few who doubt Mark Cerny claim about 8TF with FP16,but Ronvalencia use to quote it religiously when he though Scorpio had it,i even ask him to provide a link to where MS confirmed FP16 on Scorpio,and just because sony had something doesn't mean MS also has it..

Just like i asked for proof or Ryzen and Vega as well and he never did,he just misquoted Phil Spencer..

He doesn't even admit he was wrong.

Red herring with your FP16 argument.

Polaris already has increase data processing features with dual Fp16.

http://gpuopen.com/using-sub-dword-addressing-on-amd-gpus-with-rocm/

The following code shows how this is implemented using GCN ISA (going forward in this blog, all comparisons are against ISA capabilities of GFX 8 Fiji + Polaris):

In several Machine Learning algorithms, training the data on fp16 data types has been effective in increasing the execution rate.

SDWA operations seem nice, but how do they perform and how do you use them? In HIP, several SDWA math operations are implemented using the techniques described in this blog. A full list can be found here Link.

HIP is a portable higher level API that has similar syntax as CUDA which works on both AMD and NVIDIA GPUs. Module APIs from HIP can be utilized to load and run AMD GPU HSA code objects compiled offline

AMD's FP16 features designed to target CUDA GPUs with dual FP16 processing support (e.g. GP100, TX1 and TP1) and this feature to be enabled with incoming Shader Model 6.

Polaris dual subword FP16 doesn't increase math operation rate i.e. TFLOPS number remains the same but it has increase data element processing rate.

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

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

XBO version is already tiled render game and GPU's 2MB L2 cache was used to reach 4K.

Existing AMD GPU's pixel engines are not even connected to L2 cache.

@04dcarraher said:

I think we can expect Scorpio's gpu performance around a RX 580

Bullshit, RX-580 has memory bandwidth bottlenecks and you know it.

http://www.eurogamer.net/articles/digitalfoundry-2017-the-scorpio-engine-in-depth

"For 4K assets, textures get larger and render targets get larger as well. This means a couple of things - you need more space, you need more bandwidth," explains Nick Baker. "The question though was how much? We'd hate to build this GPU and then end up having to be memory-starved. All the analysis that Andrew was talking about, we were able to look at the effect of different memory bandwidths, and it quickly led us to needing more than 300GB/s memory bandwidth. In the end we ended up choosing 326GB/s. On Scorpio we are using a 384-bit GDDR5 interface - that is 12 channels. Each channel is 32 bits, and then 6.8GHz on the signalling so you multiply those up and you get the 326GB/s."

Scorpio's GPU has more than 300 Gbps memory bandwidth allocated to it. Polaris DCC recovers memory bandwidth lost to memory subsystem inefficiencies.

Yeah lets ignore Scorpio bigger case probably,lets ignore its more expensive cooling system,lets ignore Scorpio 4k drive which is totally absent on pro,lets ignore 4GB of extra ram which is also faster than Pro.

Lets ignore all that and pretend BOM cost is only based on the stupid Soc for both platforms.

It is post like this that make you look like a raging MS fanboy.

No hes argument is not red herring.

Scorpio DOESN'T HAVE FP16 DOUBLE PUMPED CONFIRMED BY MS ITSELF NEITHER DOES POLARIS,IS A VEGA FEATURE.

One thing is FP16 another is double pumped which scorpio can't do.

And yes Forza is badly optimized on PC which hold back GPU like the 1070gtx which make your comparison totally null.

The Vega feature that I'm after is the rendering to L2 cache feature.

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

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

XBO version is already tiled render game and GPU's 2MB L2 cache was used to reach 4K.

Existing AMD GPU's pixel engines are not even connected to L2 cache.

Polaris already has dual subword FP16 feature that increase data element processing.

Deal with it.

When a company reveal its brand new console specs and its looking this bad, what were anyone expecting? PsPro isn't an upgrade but an additional product Sony was able to sell under its huge dominance and popularity.

When you upgrade, you expect an actual upgrade, just like you would when upgrading your pc. No one upgrades their pc from playing on med settings/60fps to med settings/60fps in the same games.

I'm all for upgraded consoles every three years, but the upgrade must be just that, an actual upgrade. One that you're able to notice without DF videos.

Polaris dual subword FP16 doesn't increase math operation rate i.e. TFLOPS number remains the same but it has increase data element processing rate. Try again.

MS confirms Scorpio has baseline Polaris IP which includes it's sub-word FP16 processing features. Furthermore, Fury and Polaris native FP16 processing features to be enabled under Shader Model 6.

The Vega feature that I'm after is the rendering to L2 cache feature.

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

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

XBO version is already tiled render game and GPU's 2MB L2 cache was used to reach 4K.

Existing AMD GPU's pixel engines are not even connected to L2 cache.

MS has stated Scorpio has Vega features, but not Vega's double rate FP16 with math operation rate increase . MS selected Vega's second major feature.

http://www.ticgn.com/xbox/scorpio-price-possibly-leaked-retailer/

That works out to $428.74 USD at today’s conversion rates which means a price point of $399 to $449 is certainly possible if XtraLife’s information is accurate.

http://www.guru3d.com/articles-pages/geforce-gtx-1060-review,2.html

The GeForce GTX 1060 is your standard dual-slot solution product, its cooling is vapor chamber based.

If you can't notice the difference between a well code Pro game and a PS4 normal one you have eye side problems.

Horizon on Pro kick the living crap out of the PS4 version.

And considering that you are not dropping $400 for a new GPU and $200 more for a new CPU i think the balance is well made the Pro was $100 more than the PS4 but also came with a 1TB disc and more powerful hardware,the XBOS 2TB was $400 when the Pro was $400 as well how can that be,when the XBOS is a cheap console with a big drive.

So not only do i have to pay an additional $400, but i'm also hoping and praying 1 out of 10 games is well coded for the system. What a great deal. Xbone is a POS. No point of comparing the Pro to it.

The gpu received a x2 boost. If you're unable to do the same (and i'm aware that the cpu doesn't need to be as powerful as the gpu) with the cpu, call your new console as a replacement to the old models, not an upgrade. Because in reality, the Pro is a joke of an 'upgrade'.

Scorpio will have a more reasonable upgrade but even MS are still using the same mediocre cpu. How can anyone expect 60fps from an 'upgraded' console when even a simplistic looking title such as Destiny 2 is unable to run at 60fps (on the pro and most certainly scorpio)? Spending $400 to play at a stable 30fps instead of the usual 21-30fps isn't much of an upgrade because the original systems (ps4 and x1) were already supposed to achieve 30fps stable by default. That wasn't the case, clearly. Now, they're selling us 'pached up' consoles for $400+. Not much of an upgrade, more of a patch up.

With all the other bottlenecks in the system will put the gpu's performance below your "assumptions". You talk about memory bottlenecks and yet GTX 1070 also uses 256bit bus with 256gb/s rate the same as RX 480/580 and yet still performs better than Fury with HBM and competes and regularly beats 980ti with 337gb/s with similar TFLOP performance. Bandwidth is only part of the problem... where 480 barely performs better than 290 and yet 290 has 320gb/s bandwidth. your overlooking the other aspects..... ie pixel rate and texture rate

R9 290 has a pixel rate of 60 and texture rate of 151, while RX 480's pixel rate is 40 and texture rate of 182.....

With up to date charts At 4k with BF1 RX 480 only averaging 6-7 fps behind GTX 1070, but the point is that both gpus aren't strong enough to handle 4k to even average 40 fps.

then Doom Rx 480 is only around 8 fps behind a GTX 1070 and again 1070 barely passing 40 fps average.

I don't understand why people would want to save $100 on hardware while they are paying $60-70 on a single game, or even more

The ps4 pro will seem more valuable over time, especially for major titles. But in general it is not a successful system, although if you want to lower the bar since it is mid-gen upgrade then sure, go ahead.

You still dismiss effective memory bandwidth with memory compression/delta color compression (DCC) difference.

GTX 1070's Pixel Engine is connected L2 cache and superior Pascal DCC and it's capable rivaling GTX 980 Ti's 384 bit GDDR5-7012 + Maxwell DCC + Pixel Engine-to- L2 cache connection.

Remember, Scorpio's Forza M6 4K/60 fps result was superior to GTX 1070 and Richard still claims Scorpio being still GTX 1070/Fury X class.

Do you still dismiss memory compression difference?

For Scorpio's Forza M6 4K/60 fps results

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

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

XBO version is already tiled render game and GPU's 2MB L2 cache was used to reach 4K.

Current AMD GPU's Pixel Engine doesn't have 2 MB L2 cache connection i.e. RX-480 wasted it's 2 MB L2 cache.

http://www.eurogamer.net/articles/digitalfoundry-2017-the-scorpio-engine-in-depth

"For 4K assets, textures get larger and render targets get larger as well. This means a couple of things - you need more space, you need more bandwidth," explains Nick Baker. "The question though was how much? We'd hate to build this GPU and then end up having to be memory-starved. All the analysis that Andrew was talking about, we were able to look at the effect of different memory bandwidths, and it quickly led us to needing more than 300GB/s memory bandwidth. In the end we ended up choosing 326GB/s. On Scorpio we are using a 384-bit GDDR5 interface - that is 12 channels. Each channel is 32 bits, and then 6.8GHz on the signalling so you multiply those up and you get the 326GB/s."

Scorpio's GPU was allocated with "more than 300 GBps" of memory bandwidth.

For Scorpio's Forza M6 demo, the GPU memory setup duplicated 980 Ti's memory setup.

----------------

R9-290/R9-290X doesn't have 320 GB/s effective memory bandwidth!

RX-480 and R9-290/R9-290X has similar effective memory bandwidth, hence similar results.

Facts: R9-290/R9-290X doesn't have Polaris DCC! Deal with it.

@04dcarraher said:

R9 290 has a pixel rate of 60 and texture rate of 151, while RX 480's pixel rate is 40 and texture rate of 182.....

Useless without effective memory bandwidth context.

@04dcarraher said:

With up to date charts At 4k with BF1 RX 480 only averaging 6-7 fps behind GTX 1070, but the point is that both gpus aren't strong enough to handle 4k to even average 40 fps.

That's not my not argument.

Where's your "With up to date charts At 4k with BF1 RX 480 only averaging 6-7 fps behind GTX 1070" source links?

@04dcarraher said:

then Doom Rx 480 is only around 8 fps behind a GTX 1070 and again 1070 barely passing 40 fps average.

GTX 1070 has 11 fps advantage over RX-480 ie. 49 - 38 = 11 fps.

Doom 2016 Vulkan has heavy Compute Engine Path usage which has L2 cache connection on both AMD and NVIDIA GPUs.

@ronvalencia:

Your over exaggerating the gains from the bandwidth from going from 256gb/s to 300-320gb/s..... Plus your forgetting Polaris also incorporates memory compression methods as well.

The Forza 6 example is flawed do to the fact that game is a meh port on pc. It does not gauge the effect of "bandwidth" being sole the factor for 4k.

Also pixel fill rates is not dependent totally on memory bandwidth. ie fury X's 8TFLOP range with 512GB/s being beaten by GTX 1070 with half the bandwidth and 2 TFLOP less. however 1070 having nearly 40% higher pixel fill rate. however losing to fury x texture fill rate by nearly 50%.

you need to look at more current RX 580 results to look driver maturity for RX 480 with same games,

@04dcarraher: The bandwidth is a game to game basis too, not all games require the same

@04dcarraher said:

@ronvalencia:

Your over exaggerating the gains from the bandwidth from going from 256gb/s to 300-320gb/s..... Plus your forgetting Polaris also incorporates memory compression methods as well.

The Forza 6 example is flawed do to the fact that game is a meh port, that only allocates two cores fully with any cpu with more than 4 threads.

Also pixel fill rates is not dependent totally on memory bandwidth. ie fury X's 8TFLOP range with 512GB/s being beaten by GTX 1070 with half the bandwidth and 2 TFLOP less. however 1070 having nearly 40% higher pixel fill rate. however losing to fury x texture fill rate by nearly 50%.

you need to look at more current RX 580 results to look driver maturity for RX 480 with same games,

My FarCry Primal benchmark graph has RX-580 (6.17 TFLOPS with 256 GBps physical memory bandwdith) along with R9-290 (4.8 TFLOPS with 320 GBps physical memory bandwidth) as a point of reference.

The reason for why I selected Far Cry Primal is due to NVidia Gamework sponsorship.

Both BF1 and Doom are AMD sponsored titles.

http://www.geforce.com/whats-new/articles/far-cry-primal-gears-of-war-ultimate-edition-game-ready-driver-released

In reference to your posted graphs, it doesn't make any major difference, since RX-580 with 256 Gbps physical memory bandwidth still shows major memory bottlenecks since R9-390 only has 5.1 TFLOPS while RX-580 Nitro+ Boost has 6.5 TFLOPS.

Increase in TFLOPS has a minor increase frame rates (less frame time completion via higher FLOPS) but it doesn't negate memory bottlenecks (feeding the ALUs via TMU).

My argument is based on PS4 Pro programmer's POV. http://gamingbolt.com/ps4-pro-bandwidth-is-potential-bottleneck-for-4k-but-a-thought-through-tradeoff-little-nightmares-dev

PS4 Pro has 4.2 TFLOPS with 218 Gbps physical memory bandwidth before DCC and memory subsystem inefficiencies are factored in.

That's 51 Gbps physical memory per 1 TFLOPS ratio.

RX-580 Nitro+ Boost's 6.5 TFLOPS is 39.39 Gbps physical memory per 1 TFLOPS ratio which is worst than PS4 Pro's bottleneck.

Scorpio's 6 TFLOPS is 54.3 Gbps physical memory per 1 TFLOPS which is better than RX-580 Nitro's physical memory bandwidth : TFLOPS ratio.

Furthermore, BF1 is a well known game with higher Compute Engine to L2 cache usage via compute shaders i.e. it's AMD sponsored game.

You haven't factored in workload type with Forza M6's wet track example i.e. tons of alpha and transparency effects. Pixel Engine's ROPS path handles blending functions.

I down clocked my GTX 1080 Ti to 6.5 TFLOPS with memory setting kept as is and it still delivers 4K 60 fps Forza M6 wet track results. The majority of Scorpio's improvements deals with increase memory bandwidth and reduce bottlenecks improvements. MS demonstrated Scorpio's Pixel Engine improvements over RX-480 OC and R9-390X OC.

I have factored in Polaris DCC.

Notice RX-480's 264 GBps compressed vs 194 Gbps non-compressed.

Polaris compression ratio is 1.36 compressed to 1 non-compression i.e. 264 / 194 = 1.36

Games like Doom Vulkan has heavy Compute Engine path usage, hence both AMD and NVIDIA are even with L2 cache usage.

Forza M6 wet track smashed current AMD GPU's Pixel Engine bottleneck.

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

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

XBO version is already tiled render game and GPU's 2MB L2 cache was used to reach 4K.

Current AMD PC GPU's Pixel Engines runs into memory controller bottleneck!

Forza M6's 2 CPU core usage is not a large problem with Haswell at 4.5 Ghz overclock.

-------------

Fury X's 512 Gbps physical memory bandwidth is not effective memory bandwidth.

Your argument with pure physical memory bandwidth is flawed since Fury X's doesn't get 512 Gbps effective memory bandwidth in the first place. Fury X's compression is garbage with too many exceptions.

For scorpio you have to pay even more and the same apply as well scorpio only increase resolution and frames on games that allow it,and any extra effects and quality increase need to be coded and on Scorpio side it is say to cost even more $500.

But that is the point you complain about the pro when the XBOS was $400 with a 2TB HDD when it launched last year same as a Pro with weak sauce hardware.

It is an upgrade because it has a more powerful GPU and faster GPU as well as more ram,and bigger standard drive.

So yeah it is an upgrade,just because you chose to change your GPU for something double as powerful doesn't mean you didn't upgrade it that make no sense.

Scorpio will also have a higher price 1 year latter which isn't huge,you cry about the Pro being just 2X the PS4 but Scorpio is just 43% more than the PS4 Pro with the same 1GB drive,so you will pay $100 more for 43% more power while the Pro gave you 100% more power over the normal PS4 and bigger drive and was $100 more.

BY your own logic Scorpio is not a big jump over the Pro.

Not all games can run at 60FPS even on hardware that is more powerful than scorpio,everything depends on what you do and there are 60FPS games on Pro.

Is an option over both current gen consoles,and i would certainly like more 1800p than 720p,even more when Pro have also higher visual fidelity.

This is the problem with you,you use speculation to counter peoples points.

The price is unknown.

Scorpio has a more expensive cooling solution that Pro because it produces more heat from both of its main components..

Scorpio has more memory.

Scopio has 4k drive.

Scorpio will probably be bigger than Pro case wise.

Scorpio probably will feature a power brick.

This year scorpio Soc is relatively more costly that Pro simply because it is new and have clocks never achieve before on GPU on consoles,the yield are also lower,this year sony can shave money from scorpio cost because by now the Soc has mature which could make even wider the gap without incurring in loses.

So trying to claim cost is just 12% more on Scoprio based on a Soc that you know nothing about is a joke and show how skew your arguments are,even 1 dollar more on scorpio side in manufactoring cost over the pro mean 10 million more dollars in just 10 million units,now imagine if the cost is $30 more than the pro and multiply that by 10 millions,what seem small to you really isn't.

Scorpio has more than 300GB/s so does any PC with a RX480 and DDR4,remember PC has 2 bandwidth not just 1 no matter how you try to deny it.

Consoles are budget machines.

And nobody is going to develop for tiny userbases.

The pro was designed to prevent playstation players from moving to PC platform in mid gen / end gen of its life span. Because the PS4 simple isn't providing a good enough experience anymore for those people.

The problem with this reasoning is.

1) Somebody that pays extra money for a console to play the same games just for better performance will most likely already sit on a PC solution already. Which makes the pro completely useless.

2) people that play pc games and console games mostly do it on the cheap solutions. The cheapest playstation and the cheapest PC ( laptop ). A pro and a gtx 1080ti is not even interesting for them even remotely. So making a pro for those people is totally useless.

There requirement is mostly, i just want to play it. Can it run it?

Not really, most people don't even know what fps are, or what high / mid settings are. They just want to run the game.

4k actually makes sense towards them, because that showcases itself on tv screens.

So a 4k playstation with a 4k screen makes sense. 60 fps over a 40 fps on a game doens't make sense towards them.

Keep in mind your average consumer is uninformed and dumb as a rock.

Higher framerate isn't interesting towards them, as i wouldnt'be shocked that 90% of the playstation userbase isn't even aware on the differences of framerate.

1. Vapor cooler...

http://www.guru3d.com/articles-pages/geforce-gtx-1060-review,2.html

The GeForce GTX 1060 is your standard dual-slot solution product, its cooling is vapor chamber based.

Sony followed mainstream AMD's GPU cooling solutions. MS is looking at NVIDIA camp, not just Maxwell architecture.

http://www.coolermastercorp.com/manufacture.php?page_id=9

Vapor Chambers provide far superior thermal performance than traditional solid metal Heat spreaders at reduced weight and height

https://www.amazon.com/Dynatron-R23-Bridge-Heatsink-Cooler/dp/B009PAM8SS/ref=sr_1_3?s=electronics&ie=UTF8&qid=1496585167&sr=1-3&keywords=vapor+cooler

$26.49 retail for Sandy Bridge EP vapor cooler solution. Cheaper at BOM cost.

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

"So really, between trying to target a compact design, and also strike overall efficiency and minimise power use, we do a lot of things that are special," says Leo Del Castillo, general manager of Xbox hardware design. "One of the things we do is we basically fine tune the voltages for each of the chips and optimise them so the chips are getting exactly what they need to get the job done... That drives a much higher degree of efficiency into the system and allows us to get rid of a lot of wasted power that would otherwise come out as heat."

It's a technique that Microsoft calls the 'Hovis method', named after the engineer who developed it. Every single Scorpio Engine processor that comes off TSMC's production line will have its own specific power profile. Rather than adopt a sub-optimal one-size-fits-all strategy, Microsoft tailors the board to match the chip. Regardless, Scorpio is still a powerful piece of technology, so there is still the issue of dissipating its heat. Here again, the solution is revolutionary for a console.

Finer energy delivery control to reduce heat generation.

2. Minor cost increase.

Based from https://pcpartpicker.com/products/video-card/#sort=price&c=391 with RX-580 4GB vs RX-580 8 GB difference.

It's another $34 at retail price. BOM cost is less than $34 difference. Some AIB sells RX-480 8 GB as RX-480 4 GB, hence minor cost difference.

3. Minor cost increase.

http://news.ihsmarkit.com/press-release/technology/ihs-markit-teardown-analysis-microsofts-xbox-one-s-brings-significant-value

During year Nov 2016, $33.50 BOM cost for UHD BD-ROM player vs $18 conventional Blu-Ray drive.

$15 increase.

4.

http://www.gamesradar.com/xbox-scorpio-looks-smaller-than-xbox-one-and-some-other-details-from-the-dev-kit-reveal/

Scorpio dev kit is smaller than XBO.

http://www.product-reviews.net/2016/11/04/ps4-pro-unboxing-size-vs-xbox-one-s-compared/

5. Wrong. https://www.windowscentral.com/project-scorpio-hevc-internal-psu-4k-game-dvr

Project Scorpio' Xbox will pack internal PSU

6.

RX-570 has faster clock speed over RX-470 with similar price. ~1 year silicon maturity yields faster clock speed at similar price.

R9-390 has faster clock speed over R9-290 with similar price. ~1 year silicon maturity yields faster clock speed at similar price.

7. You don't know shit.

8. You don't know shit. Intel Sandybridge Core i5-2500 at stock clock speed with 128bit DDR3-1600 (25 Gbps) still murders PS4 Pro results.

PCI-E version 3.0 16X is required to exceed PS4's 10 + 10 GBps links, hence Intel Ivybridge is required.

Seriously, 30 fps and 60 fps AI/Physics are low targets.

Parts of GPU command list build are transferred to custom Direct3D12 command processor with microcode (mini-CPU).

Just for you, I set my secondary PC (with GTX 980 Ti GPU), 128 bit DDR3-2133 down to 128 bit DDR3-1600 (25 Gbps) and it still murders PS4 Pro in Tekken 7. Again, You don't know shit.

Game console AI and Physics are low targets.

Hint: When PC benchmark is not a target (i.e. not targeting e-penis PC master race benchmarks), beating PS4 Pro is easy.

Remember, gaming PC with dGPU card doesn't have fusion links, no matter how you try to deny it.

@ronvalencia: damn you rekt the **** outta Tormentos.

Actually we dont know that if its faster that RX 580 by 20% or what ever . We need the full performance specs. If the pixel fillrate is no higher than RX 580 then we are effectively looking at on par performance. We have to look at the other bottlenecks from within the system as well which will hold the gpu back from full potential. All we know for sure RX 580 and Scorpio are 6TFLOP gpus based on virtually the same architecture. So seeing sizable gains between them is unlikely.

Your i7 960 is still much faster than the jaguar/puma based cpu in Scorpio. But also we have to look at the fact that Scorpio 8gb dedicated for the games will still be split between game cache and vram which means the gpu will not be able to use 8gb. Ratios will most likely be 3/5gb or 4/4gb or 2/6gb types of usage. Playing games with a slew of medium, high and some ultra settings at 4k at 30 fps is a waste of resources when you could cut the resolution in half run everything on high/ultra and get near 60 fps.

Tormentos never knows what he is talking about. He gets all his info from google and Neogaf and posts it like he came up with it.

Thats why he is constantly getting owned, resulting in him calling everyone a lemming.

There are differences between 6 TFLOPS Scorpio Engine and 6 TFLOPS RX-580 (Polaris 20)

1. More on-chip registers/more L1 cache/more LDS on Scorpio Engine e.g. 40 CU vs 36 CU.

2. Scorpio Engine has Vega improvements without double rate FP16 (which doubles the operation rate) i.e. Vega's non-CU improvements. Sony selected Vega CU instead of non-CU improvements.

3. 44 CU GCN has a wider internal bus over 36 CU GCN.

4. More than 300 Gbps memory bandwidth allocated to Scorpio Engine.

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

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

Forza XBO version is already tiled render + texture tiling (PRT) game engine and GPU's 2MB L2 cache booster was used to reach 4K performance. This sounds like Vega's second major improvement area.

Scorpio Forza M6 has L2 cache booster for it's 4K resolution rendering while PC AMD GPU's Pixel Engine directly smashed into memory controller bottleneck. MS recycled the technique from Maxwell.

Forza M6's wet track has tons of alpha and transparencies effects which are heavy Pixel Engine based effects. MS selected a workload that crippled AMD's current PC GPUs and it wasn't bad on GTX 1070 and Fury X. The purpose was to demonstrate Scorpio Engine's Pixel Engine improvements.

I tested my theory with 6.5 TFLOPS GTX 1080 Ti and mostly Pixel Engine path improvements.

There little performance difference between NVIDIA's Compute and Pixel Engine paths while AMD's GPUs has wider performance difference between Compute and Pixel Engine paths.

6. Design goal for Scorpio Engine is to speed up XBO processing model and that's with less ACE units than PS4. XBO games are not like PS4 games with 8 ACE units.

7. Scorpio Engine hardware was profiled against MS first party games and Unreal Engine 4 dominates MS first party games.

8. Scorpio Engine is equipped with bake in DirectX12 API command processors with microcode support (mini-CPU).

https://en.wikipedia.org/wiki/Microcode

Microcode is "a technique that imposes an interpreter between the hardware and the architectural level of a computer".^[1] As such, the microcode is a layer of hardware-level instructions that implement higher-level machine code instructions or internal state machine sequencing in many digital processing elements.

@ronvalencia:

Your making too many assumptions.... The Forza example is flawed.... too many factors as in poor cpu scaling and other areas make M6 as an example moot.

Again we need actual performance numbers from the gpu to gauge what it could do, speculating on the increased gpu cache and a wider memory bus, with a revised command processor which still is fed by the main cpu so it acts like a buffer to keep the gpu from stalling. will help but it boils down to that fact that the gpu is still a 6 TFLOP gpu still vastly based on polaris, still being held back by 2.3ghz amd cpu based on jaguar or puma.... then has less vram buffer for assets.

Scorpio is not a true 4k ready gpu, the performance will be between RX 580 and GTX 1070, at best.

50 GB/s more isn't going to make a sizable difference on a 6 TFLOP gpu. Especially if the pixel rate isnt in the right league for 4k.

Memory overclock (bandwidth increase)will help texture fill rate alot more than pixel fill rate, it wouldn't increase performance that much. Even with an effective 1000mhz overclock on memory ( 9000mhz vs 8000mhz stock) on a RX 580 along with a 210mhz core overclock. vs a stock RX 580 average gain is around 7% at 1440p. The memory overclock increase brings the 256gb/s to like 288 gb/s.

The deciding factor in how well Scorpio performs at 4k will be primarily what we find out on its pixel fill rate.

Actually no your i7 960 would perform better in DX11 than Scorpio cpu..... DX11 is limited to deferred multithreading, your cpu would still smash a 2.3ghz AMD Jaguar/Puma based cpu into the ground. Jaguar architecture is much slower than AMD's old Phenom 2's clock per clock. Even in DX12 your aging i7 960 will give Scorpio CPU a run.

Tormentos is a major fanboy and you should have seen the arguments him and me had about the PS4 and Pro. My assumption on hardware performance range for PS4 and Pro came pretty close to what I predicted.

Scorpio wont be running alot games at 4k at 60 fps unless there are alot of compromises in quality is put into affect. Even top tier gpu's like gtx 1080 or even 1080ti in some games aren't even 4k 60 fps ready using high/max settings.

1. CPU scaling is not a large problem when you have 4.5 Ghz overclock Intel Haswell/Skylake/Kabylake CPU. The bottleneck is GTX 980 Ti GPU not my i7-4790K at 4.5 Ghz i.e. it's not hitting 100 percent in each of the two CPU threads. FM6's wet tracks with heavy alpha and transparent effects are not heavy CPU based effects i.e. they are Pixel Engine heavy effects. On dry tracks, my 980 Ti can do 4K/60 fps without problems. Your argument on CPU scaling is moot.

I overclocked my i7-4790K to 4.6 Ghz for FM6 at one time. At 4.6 Ghz, CPU's power consumption exceeded 126 watts with MT Super Pi CPU stability test, it crashes after 10 minutes and motherboard VRMs are hot. That's how far I played with my i7-4790K i.e. I need a new motherboard with better VRMs.

2. For GPU command list building, the custom command processor can reduce CPU workload up to half.

If a console game has 60 hz Physics and AI at 1.6 Ghz CPU, your argument is moot.

3. Being a Polaris means nothing just as being a Maxwell also means nothing i.e. there are crappy Maxwell GPUs. You still missed the following

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

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

Along with "more than 300 GBps" memory bandwidth, Scorpio GPU's 2MB L2 cache booster was used to reach 4K.

Existing AMD GPU's Pixel Engines are not even connected to L2 cache and that includes Polaris 10/20.

4. That's not my argument. My argument is Scorpio Engine is better than RX-480 OC and R9-390X OC while inferior to RX-Vega 10.

The Jaguar argument...

Intel CPUs has almost half the L2 cache latency from Jaguar. This problem affects real world fetch rates from L2 cache. MS claims lower latency improvements for CPUs. Jaguar has room for improvement.

Ryzen includes both lower latency L2 cache latency and 2X wider L2 cache. MS's lower latency improvement is a half way house to Ryzen's full L2 cache improvements.

Jaguar's real world IPC vs theoretical IPC

Jaguar has theoretical 2 IPC with real world 1.10 IPC for typical apps. Jaguar has room for improvement in this area.

Again, Hawaii XT has the synthetic numbers with real world bottlenecks e.g. Pixel Engine not being connected to L2 cache, missing Polaris DCC, missing Polaris triangle culling and 'etc'.

DCC is important to reduce memory bandwidth consumption hence improving concurrent GPU workloads e.g. concurrent ROPS and TMU operations.

Pixel Engine connected to L2 cache reduces external memory hit rates and lower latency between ROP and TMU ping-pong loop.

Holy shit. @ronvalencia is still posting charts lol.

@Juub1990: I can't even read these threads after a couple of pages. It becomes pages of quote chains filled with charts and incomplete sentences and people screaming "fanboy" and "get a clue" at each other. We all have out preferences and such but damn :-P

As a request could you close this thread? I created it and would like to kill it too.

@Juub1990: sure thing. Sometimes they spin off the rails