Rumor: Xbox Project Scarlett target specs more than double Xbox One X in many areas

Both Volta and Turing shares CUDA compute capability 7.x . https://docs.nvidia.com/cuda/turing-tuning-guide/index.html

The Turing Streaming Multiprocessor (SM) is based on the same major architecture (7.x) as Volta, and provides similar improvements over Pascal.

With RTX 2080 Ti, we have Volta+ CUDA 7.x compute capability without usable FP64 and HBM 2.

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

Volta has CUDA Compute capability version 7.0 to 7.2

Turing has CUDA Compute capability version 7.5

Optimizations for Volta is applicable for Turing. There's only minor CUDA compute capability differences between 7.0/7.2/7.5 versions just as AMD's GFX 9.x Vega family.

In terms of features, Volta is similar to GTX Turing with extra Tensor cores . RTX Turing has RT cores and Tensor cores.

From programmer's POV, Turing is treated like a Volta.

Developing AI apps for NVIDIA's Volta CUDA CC v7.0/7.2 embedded platforms via RTX Turing equiped workstation PCs are valid. There's dev kit hardware inside a gaming PC.

So basically close to earlier rumors? The GPU will be the same as the PS5. DF said that with the new AMD architecture 10.8Tflops will be the same as 12Tflops of the old one.

@Martin_G_N: "DF said that with the new AMD architecture 10.8Tflops will be the same as 12Tflops of the old one."

Is that good or bad?

We have to wait and see but you are sure its going to be huge,,,,OK

@ronvalencia: You should stop you die hard lobying of MS.

The Xbox one X lose the RX580 in many instances period.

So yeah you don't need 12tf to double the Xbox one X power and once again MS always portrait the best scenario or scenarios that don't hold up. Aka True 4 no compromises and 60fps.

Indeed.

"We could be looking at a generational performance improvement of 40-50 percent or more for RTX 3080, with a chip that's still 15 percent smaller than the RTX 2080."

https://www.pcgamer.com/looking-forward-to-nvidias-rtx-3080-and-the-next-generation-7nm-gpus/

1. X1X beats reference RX 580 on games that matter for 4K e.g. Far Cry 5 4K, Gears of War 4 4K and Forza Motorsport 7 4K wet tracks.

2. X1X wasn't designed for 1440p high frame rates which is bias towards CPU bottlenecks. X1X was designed for Digital Foundry's XBO vs PS4 resolution gate debacle.

Similar to Vega 56 at 1710 Mhz with 12.1 TFLOPS beating Vega 64 at 1590Mhz with 13 TFLOPS.

Higher clock speed improve chip's internal bus, L1/L2 cache bandwdith, geometry and ROPS.

Digital Foundry didn't use R9-390X (5.9 TFLOPS) which has better 4K GPU when compared to memory bandwdith gimped RX-580 (6.1 TFLOPS).

RX 5700/5700 XT has RX Vega II's quad geometry units and 64 ROPS backend (Z-buffer and color pixel read/write), delta color compression and 448.0 GB/s memory bandwdith which is higher than Vega 56's 410 GB/s memory bandwdith.

Vega 64 has 484 GB/s and AMD's HBM v2 memory controllers seems to have higher latency when compared to AMD's GDDR5 memory controllers.

The main difference between NAVI 10 and Vega II on geometry as follows

NAVI 10 ASIC codename: 8 input, 4 output per cycle. Fixed back face geometry culling.

VEGA 10/20 ASIC codename: 4 input, 4 output per cycle, needs compute shader workaround to cull back face geometry.

NAVI 10's wave32 single cycle and wave64 two cycle latencies vs GCN's wave64 four cycle latency eqates to improved branch handling and less need for preparing four wave64 to hide GCN's wave64 four cycle latency.

To support hardware VRS feature, Geometry and ROPS units needs to be change again when Vega introduced multi-MB cache with ROPS as per Maxwell/Pascal GPU designs.

X1X GPU already supports VRS like feature.

From https://www.techpowerup.com/261328/nvidia-readying-geforce-rtx-2080-ti-super-after-all

"RTX 2080 Ti Super" in year 2020...

The main bottleneck is ROPSand memory controller coupled design.

Adding extra TFLOPS would be nearly pointless without addressing front end geometry and back end ROPS I/O issues. TFLOPS by itself doesn't include I/O.

With RTX 2080 (TU104), NVIDIA manage to fit six GPC units with 64 ROPS and 256 bit bus.

TU102 and TU106 follows GPC/ROPS/bus width ratios like Pascal GPUs e.g.

Six GPCs with 96 ROPS and 384bit bus RTX Titan (TU102)and GTX Titan Xp (GP102)

Six GPCs with 88 ROPS and 352 bit bus, RTX 2080 Ti (TU102 cutdown) and GTX 1080 Ti (GP102 cutdown).

Four GPCs with 64 ROPS and 256 bit bus, RTX 2070 (TU106) and GTX 1080 (GP104).

Turing includes Async compute, 2X L2 cache storage, delta color compression, split INT/FP pipelines, VRS and rapid pack math improvments when compared to Pascal.

TU104 is a standout design that didn't follow Pascal s GPC/ROPS/bus ratios.

I expect future GPUs to scale from TU104 e.g.

Titan Ampere with 8 GPC, 96 ROPS and 384 bit bus.

RTX 3080 Ti with 8 GPC, 88 ROPS and 352 bit bus

@ronvalencia: ”X1X was designed for Digital Foundry's XBO vs PS4 resolution gate debacle.”

I’ve never agreed with you more. Where was this line of thinking when you had those goons creaming themselves over Scorpio beating the 1070/1080? @kuu2 and @antwan3k might still be with us today...

In the speculation thread, my rapid pack math speculation has an IF conditional statement, but I welcome alternative X1X GPU raster related improvements such as ROP's 2MB render cache and VRS like features.

Vega improvements involves more than just CU's ALU!

Picking between raster vs ALU improvmements, I'll pick raster hardware 1st! (reason, this is NVIDIA approach).

For Forza Motosport 7 wet tracks, X1X is superior to RX-580 and R9-390X. It's a no brainner to why X1X is superior for this particlar workload i.e. RX-580 and R9-390X ROPS doesn't have multi-MB cache with ROPS design!

MS overloaded alpha blending workloads to show X1X's design changes from AMD's PC GPUs with similar TFLOPS. Alpha blending can NOT be done purely on TMUs, hence AMD's TMU workaround wouldn't work.

Alpha blending can be done with compute shaders and TMUs, but it reduces the available compute shader resource.

Alpha transparencies also gimps DCC.

GTX 1070's 64 ROPS with 2MB L2 cache read/write has reduce need for TMU read/write workaround and there's already cases that shows RX-580 (6.1 TFLOPS) or R9-390X (5.9TFLOPS) approaches GTX 980 Ti and GTX 1070 (+6.5 TFLOPS) e.g. Titanfall.

NAVI 10 enables AMD to master 64 ROPS (read/write I/O and fix function effects) with 256bit bus PCB design!

GTX 1070 vs RX-580 vs X1X GPU

Pixel shader ROPS read/write to 2MB L2 cache, GTX 1070 = good (64 ROPS), RX-580 = bad (missing L2 cache link, 32 ROPS), X1X GPU = half good (2MB render cache with 32 ROPS)

Compute shader TMU read/write to 2MB L2 cache, GTX 1070 = good, RX-580 = good, X1X GPU = good

Async compute to 2MB L2 cache, GTX 1070 = half good, RX-580 = good, X1X GPU = good

PS; Xbox 360 GPU already has basic async compute feature, but not many pipelines and queues depth when compared to GCN 1.0.

Turing's double L2 cache storage improvement is not paper weight.

NVIDIA already mastered 128 ROPS with Titan V HBM v2 and 96 ROPS with 384 bit bus Titan RTX.

AMD is not a threat to NVIDIA's raster technology leadership.

@ronvalencia: You want me to dig up you saying the Xbox One X would be playing AC Odyssey and Shadow of the Tomb Raider at 4K/60fps, chump?

Are you thanking you're alt LOL.

Why on earth would I have an alt? You jelly because you get no love?

Hahahahaaaaaaaaaaaaaaaaa...

So now the few games were the XBOX X beat the RX580 are the games that matter..lol

Like FH which is holded back on purpose by MS to help its xbox machine?

Which by the way is lead platform is the damn xbox.

And is fun that you mention wet track crap fact is the xbox one X beat it by little while the RX580 without wettrack beat it by quite more.

You should stop your bullshit lobying man is pathetic.

The xbox one X was advertice ad TRUE 4K gaming 60FPS and no compromises that means ULTRA effects not low or even LOWER than low settings on PC because that my blind friend is a damn COMPROMISE.

By the way the rumors running is that MS took an xbox one X route with the xbox series X,so the xbox series X will have higher compute unit count than the PS5 probably 8 or 10 more,but clocked considerably lower than the PS5 ones,whichs is why supposedly the PS5 is performing better.

According to rumors the PS5 is running at 2ghz i find it hard to believe,but who knows.

Reading some of this thread its the same shit. "The games going to PC day and Date makes an Xbox pointless". Except to people that don't want to build, own, upgrade, upkeep, clean, dust, a PC. Some of us do not care. Its really fun to decide for oneself what is of value and to allow others the same keeps the "Im a little bitch" at bay. Give it a try folks, you'll smile more

Also @tormentos good to see you, old timer

@cainetao11: Good to see you to bro too much work lately for me.

@tormentos:

Try again

PS; RDNA 2's 7nm+ EUV has 20 percent improved transisior density over TSMC's 1st gen (fake**) 7nm process.

PS5's 239 mm2 at 7 nm+ EUV translates into 286.8 mm2 at 1st gen 7nm.

XSX's 312 mm2 at 7 nm+ EUV translates into 374.4 mm2 at 1st gen 7nm.

For comparsion, RX Vega II has 330 mm2 at 1st gen 7nm which is Tonga/Tahiti class chip area.

Both RDNA 2 GPU designs are larger than 251 mm2 size NAVI 10 build on TSMC's 1st gen (fake**) 7nm process.

**Density still inferior to Intel's 10 nm EUV IceLake.

Both RDNA 2 GPUs has RT hardware.

Sony stays with Pitcairn/Polaris 10/NAVI 10 class chip area RDNA 2 replacement. PS5 GPU could be indicative for next year's PC RX 6700 Pro/XT SKUs.

Don't worry about this year's RX 5700 Pro/XT and AMD could rename RX 5700 series into RX 6600 OEM series for PC OEM fleet builders.

MS selects Tonga/Tahiti class chip area RDNA 2 replacement.

CPU bottlenecks from Jaguar CPUs wouldn't be a problem with 8 cores "Zen 2" upgrade. AMD's PC Ryzens move to "Zen 3" next year.

MS released beta version of DXR 1.1 with hardware physics on dedicated RT cores

MS also patented 'Direct Physics' API and another patent about hardware support of physics.

NVIDIA's RTX cores needs to increase. NVIDIA Ampere RTX needs to be release ASAP.

For 7nm era generation, AMD hasn't released server/HEDT GPU with ~495 mm2 size range RDNA 2 which replaced Vega 64/Fury X size chips.

PS; Mobile phone SOCs moves to TSMC's 5 nm EUV next year.

XSX's case size is similar to Apple's Mac Pro bucket.

@ronvalencia:

Try again what dude?

I actually claimed that MS will have the bigger GPU but Sony GPU speed by leaks is a bit higher.

Rumors point at Ms having 64 cu at 1400+MHz,while the ps5 would have 52cu clocked at 2.0ghz or more those are the rumors.

Again, at 4K resolution.

The real bullshit comes from you.

PC's FM7 has option to scale towards 8K resolution (200 percent resolution render at native 4K, effectively Super Sample AA) 60 Hz with RX Vega 56 OC which reduce the need for MSAA.

Your PC argument injection is a red herring when all hardware has limits.

Reminder for RDR2 PC's release date, PC GPU power levels has moved on to another level e.g. 12 nm Turing with near 1.5X chip size over Pascal equivalents.

12 nm is half node jump, but NVIDIA's Turing deliverd near 1.5X area size chips against Pascal SKU market segments.

"Gears 4" PC build was designed with 14/16 nm era PC GPUs era.

RDR2 PC's forest simulator and artwork design are not above Forza Horizon 3 or SW Battlefront 2!

PC Ultra settings are suffering diminishing returns i.e. throwing many TFLOPS at the problem with minor graphics detail improvements. A game of Pong can waste many TFLOPS with many iteration shading loops.

No improvement with gameworld scope size and it plays like "curren gen" game console title, hence it's wasting my 8C/16T Intel CoffeeLake's and 8C/16T Skylake X CPU's potential.

Do you why PC gamers are also looking forward to PS5/XSX? Hint: removing Jagauar CPU game world design limitations.

PS5/XSX defines CPU minimum which includes dual AVX2 256bit FMA3 vector math performance profile i.e. it's about time AVX2 is used for gaming! 8 core Zen 2 has a total 16 AVX2 256bit FMA3 vector math performance profile.

...and ron is back vomiting out numbers and specs that will probably end up being wrong as usual

Try again.

ARE you still stuck with my IF conditional statement with rapid pack math when X1X GPU has alternative raster graphics enhancements?

Vega's improvements includes MORE than "rapid pack math" fool!

RPM is hardly being use fool!

@ronvalencia said:

@drlostrib said:

...and ron is back vomiting out numbers and specs that will probably end up being wrong as usual

Try again.

ARE you still stuck with my IF conditional statement with rapid pack math when X1X GPU has alternative raster graphics enhancements?

Vega's improvements includes MORE than "rapid pack math" fool!

RPM is hardly being use fool!

all that word vomit is just proving my point

Seriously. Those specs are like going back in time 10 or more years.

This is why I always say PC is at least one generation ahead of consoles.

@ronvalencia:

Bullshit Ms advertise the Xbox one X as a machine without compromises and on the damn freaking launch already was doing compromises. There is a difference between a 2 year old console and a brand new machine already doing compromises.

And on launch wasn't true 4K either in all games.

By the way I don't know the Coalition which is an Xbox studio was an expert on PS5 hardware of that Had that info.

But once again sources claim the ps5 GPU speed is quite faster than the new Xbox.

I see you already started you marathon quote of the same bullshit now is The coalitions turn..lol

@drlostrib:

Yep he's at it again,now he is on a non stop Coalition quoting spree.🤦

Digital Foundry in the double performance of the Xbox One X claim,

“However, if we're dealing with performance in like-for-like workloads, the Navi architecture's improvements deliver more perf for your teraflop than prior GCN tech, and our tests suggest that a 9-10TF Navi GPU can provide the requisite doubling of performance up against Xbox One X even before other new architectural features come into play.”

Did folks seriously just double 6TF from the Xbox One X and run with it?

At least he's giving you free bumps lol.

@BassMan: It better deliver on the 40 -50% since its critical to your claim.

I am not worried about it. 7nm Nvidia cards will be good. I just hope we see a substantial leap in RT performance. I usually turn off RT because I would rather have the higher frame rate.

Reality is we don't know what kind of performance the SeX will deliver. If MS really is using a 12tflop RDNA2 GPU than we can expect near 2080 ti performance, if they are saying its 12tflop GCN than we can expect 5700XT performance. We don't know how much more efficient RDNA2 is compared to 1st gen if at all. Also, rumor is that PS5 is clocked much higher but with fewer CUs, so does clockspeed or CU count = better performance. All in all though for the money, and I predict both consoles to be $499 it is far superior bang for the buck compared with PC.

@goldenelementxl:

Yeah I stated that a few days ago as well,you don't need 12tf Navi to double the Xbox one X.

Yes they did and people like Ron with his "theoretical" approach are just circle jerking the idea that 12TF Navi COULD happen.

Unfortunately in a world were Navi struggles to hit 11 TF without hitting 220w with a after market 3 fan vapour chamber cooler and runs at 84c... There is NO likely scenario that a 12TF Navi card WITH ray tracing cores will be in a console with a 240w PSU and single fan cooling system for the entire SOC only a year later of the 5700XT's release.

Also people don't seem to understand or comprehend just how little a 2x jump in GPU power actually is for a generational jump in a console, its by far the smallest jump ever(excluding Nintendo).

This console will be running game's at the same performance of a 5700 and in some games close to 5700 XT. WHICH is absolutely great for the price of a console even if they sell it for $550-600 which is what I suspect the price will be as I doubt they will sell this for the same price as a X1X when it launched.

I have no idea what @ronvalencia is saying most of the time, but I enjoying reading their posts

@Grey_Eyed_Elf: Not a fair comparison. The jump is 8x over baseline hardware. There was a mid-gen refresh which is a first.

@Grey_Eyed_Elf: why are you comparing the generation jump with Xbox One X instead of Xbox One? What are you expecting to happen a 5X improvement over One X? be reasonable

That's not the point. The Series X is being overhyped to the moon everywhere on the internet, when we have more than a good idea where the numbers will land. 12TF with a console TDP at a console price is lunacy. I thought the Scorpio/X hype was bad... Do any of you ever learn?!?!?!

And the X will be 3 years old when the Series X launches. Doubling the X's GPU output and matching a 5700XT isn't that impressive. And that comparison certainly doesn't match the 12TF with raytracing hype. So what the hell are we doing here? The 5700XT is not a 4K/60 card. So the Xbox Series X won't even meet some users hype they had for the Xbox One X!

I really wish Microsoft and Sony could wait 2 years to launch. We would get a proper generational leap that way. Launching consoles with 7nm and raytracing being brand new is stupid. In 2-4 years the consoles will handcuff devs more than they've ever been

@goldenelementxl: We will have to wait and see if it is 12TF GCN or RDNA2. Several sources including Digital Foundry and Windows Central who have better sources than me seem to believe it will in fact be 12tflop RDNA. We have to wait and see the improvements that 7nm EUV fab will bring. Even if it is basically a 5700XT it is still better value compared with a PC build.

5700XT - $399(cheapest)

R7 3700X - $329

8GB DDR4 - $60

that already puts it way past what these consoles will cost. Add in PSU, case, motherboard, cooling, SSD's, etc and you are easily looking at $1000 to match.

@nfamouslegend:

TSMC have already stated that 7nm+ EUV is going to yield upto 15% increase in transistor density and or increase performance per watt upto 15% over their current 7nm process. That is not enough to get a 12 TFLOP RDNA gpu to be under 200w, unless AMD has solved their efficiency curve issue with higher clocks. Now also I'm thinking if the new xbox is going to have that 3.5ghz 8 core, 12 TFLOP Navi, etc in that corsair one looking case it may be using AiO cooler .

@nfamouslegend: No no no. Digital Foundry said that 9-10TF is all that would be needed to double the Xbox One X performance. The 12TF number came from the double the X performance claim. 6x2=12. This is what happens when people who don’t know hardware start talking. Digital Foundry said 12TF isn’t needed to achieve that number, and in their tests 9-10TF hits that goal. 9-10TF makes far more sense for TDP, cooling and price. However, it’s not 4K/60 material. Then again, neither is 12TF.

Next gens secret sauce is a ssd. How dumb...

I really don't see how 12 TFLOPs RDNA is possible while keeping thermal constraint and price tag under control. It's still ambiguous what they mean by 2x the performance but I also suspect around 9-10 TFLOPs RDNA. 12x would put it on the same tier as a 2080 Ti which I don't think can be sold for a reasonable price unless Microsoft is ready to take enormous losses.

@goldenelementxl: I asked Microsoft for clarification on its current-gen multiplier comparisons with Series X, but the firm declined to commit in the here and now to a precise figure. However, our information is that the GPU is indeed 12TF and what are almost certainly well-sourced leaks from Windows Central back this up. The implications are quite remarkable because based on the processor image released at E3 combined with the sheer size of the case, even at a highly clocked 2.0GHz, we'd still require 48 compute units to get the job done - a 20 per cent increase over the Radeon 5700 XT, today's most capable Navi GPU. And that 48 CU allocation wouldn't include redundancy, where we'd expect a further four deactivated CUs to increase yield from the production line. Microsoft surprised us before in how it delivered a six teraflop GPU with Xbox One X, and I'm fascinated by the prospect of similar innovations for Series X. https://www.eurogamer.net/articles/digitalfoundry-2019-xbox-series-x-reveal-analysis

see the second line there. I'm not saying it will be 12TF RDNA, but this is just what they are saying. Also SSD has never been leveraged like they will be in the new consoles and they will have RT cores. How much do you want? I know you are usually always a debbie downer but let's be realistic here, this is a great value.

Did people really expect the performance next gen to be greater than 2x? GPUs have been struggling to double performance. The 2080TI maxes out at ~50% over the 1080 Ti. Why the shock?

@nfamouslegend: Windows Central? Might as well be Lemmings.com. Are you serious with this source?

I don't know you tell me. As for love lol, it must be heart warming knowing that there's more than you who's delusional af on here rofl.

IP_daily your an idiot and no one likes you and the stupid stuff you say. You will never get any love with how stupid and annoying you are. You are so dumb that you thought I was him. BassMan and I are smart and you are beyond stupid and annoying as hell to boot. You are the one that is delusional with all the delusional BS that you say around here. BassMan and I speak facts, something a moron like you will never know how to do. Now go F off, you little bitch.

@latech84: BassMan alt?

LOL, nice ownage. :)

No, I don't have alts and never will. I am an OG and keep it real. :)

Why am I?... Because that is how it works, not only that but that is how Phil Spencer measured the power of the series X.

What I am bringing up is what I have been since the X1X and Pro launched, is that releasing a DRASTICALLY more power SKU so close to the launch of a new generation RUINS the impact of a new generation.

Imagine if there was no X1X or Pro and they showed us 4K/30FPS with PC Ultra settings and Ray Tracing... It would blow peoples minds. The XSX and PS5 to most people is nothing new, offers nothing new and is as bland as sliced bread with a graphical jump that doesn't really spark excitement.

Also this is wrong...

A R7 3700X is a 4.4Ghz CPU... So yes its the cheapest 8 core Zen 2 CPU you can buy but its substantially better than a CPU that is rumoured to be less than 3.5Ghz. PC parts cost more because they are sold separately and are much higher quality parts by nature to bin parts with 1GHz lower spec would be a lot cheaper its why console's get those incredible price/performance releases.

As for your Digital Foundry believing it will be 12 TFLOP RDNA... On the contrary, the latest video on the XSX the talked about the 2x better GPU than X1X that Phil threw out and how vague it was and that they don't believe it will be a 2x 6TFLOP type of deal with RDNA as RDNA vs GCN is not 1:1 comparison in terms of TFLOPS.

If Phil said the XSX GPU is 2x the power of X1X and X1X is 6 TFLOPS of GCN then the likely scenario is that what ever gets you 2x the power on RDNA is what the GPU in the XSX will perform at which according to reviews is between the performance of a 5700 and 5700 XT so a it could be anything from 8-9.5 TFLOPS.

A 12 TFLOP Navi GPU is not a 2x X1X GPU it is much more powerful and if a 12 TFLOP Navi GPU was possible he wouldn't have said 2x the power of a X1X he would have said close to 3x the power of a X1X GPU.

Why would he under sell a consoles power?... Its bad marketing and hype if he did, chances are based on the game that it won't even give you 2x the performance even, due to not all games scaling in performance.

Its such a vague thing to say and when people like Phil are being vague its usually because he's hiding something that could look bad and judging how everyone here is completely delusional and or ignorant when it comes to teraflops then I say he made the right choice. You people expecting 12 TFLOPS on RDNA in a console are completely lost on the subject.