General Discussions

[colesdav]

Pricing information here:
Xbox Series S and Xbox Series X Launch November 10, Starting at $24.99 a Month with Xbox Game Pass U... 

What do you think?

Time to give up on Gaming on PC with AMD GPU and just buy the Console?

[black_zion]

For the price of an RTX 3080 you can get a console (X) plus 100 games, plus backwards compatibility with all your existing XBOX games (according to Microsoft)...That's even cheaper than Gamefly, so that is going to have a broad appeal.

That is, IF it lives up to the hype and too many corners aren't cut graphically to maintain even 30fps.

[leyvin]

In general most Xbox One X Games are quite comparable in terms of Graphics and Performance to an equivalent Windows PC.

Now this isn't because the Console wasn't capable of better performance when properly Optimised for., but more this was a case of games this Generation were Optimised / Developed for PlayStation 4 then ported to the Xbox One.

And the Consoles while closely related in terms of Hardware Components, do have notable differences...

In any case this meant that the Xbox One games tended to be less well optimised and would rarely use the unique Hardware advantages. i.e. ESRAM (similar to High Bandwidth Cache), that when used is capable of greatly improving Frame Pacing but also UHD Resolution Support & Performance. 

Many developers just skipped this because unlike the Xbox 360, requires specific development rather than simply "Enabling" it... this apparently was due to "Poor Drivers", at least that's what most Developers initially claimed; but in my experience it was due to how GCN differs from Terascale; primarily the lack of an Automated Hardware Scheduler.

So you had to manually Barrier / Load / Read / etc. which most Developers hadn't been doing and didn't do.

The Xbox Series S for £279.99 and Series X for £479.99 are actually good price points for these Consoles. 

Keep in mind that the Series S is essentially the One X with DXR 1.1 Hardware Support., while the Series X offers similar performance to an R7 3700 with RTX 2080Ti (albeit almost 2x DXR Performance).

So... we can expect to see UHD 30Hz HDR Games on the Series S., and Native 4K 60Hz HDR Games on the Series X., with very little effort from Developers for Optimisation.

Both Consoles also have much improved Latency and Bandwidth over a Desktop PC as well., so better Streaming Data Performance and Loading Times in general (unless you happen to have a High Performance SSD and RAM., which tend to be a bit pricey).

•

I've personally always recommended that Consoles in general are the best platform for Gaming the Latest & Greatest Games.

No, they're not going to offer the Best Graphics or Performance... but then keep in mind < 15% of the PC Gaming Market can even afford anything better than what Consoles can offer; and that isn't 15% of the acclaimed "1.5Billion PC Gamers" Market; because the PC Gaming Market IS NOT that big.

Want proof of that... actually look at PC Game Sales Figures., they typically barely break 10% of Console Sales Figures (for each Console).

Now as a note Steam is essentially the Biggest Gaming Platform on PC., presently boasting ~92 Million Active (unique) Users on a Monthly Basis (as of March 2020 according Statistica).

This is essentially your "PC Gaming Market"... so take note of the Steam Hardware Survey with that in mind.

Arguable it's actually a subset of that because you can "Opt. Out" and also keep in mind 30% of those figures are from Net / Gaming Café, so I'd argue they don't count. 

Still it's good Valve finally resolved the issue of Multiple Accounts using the same Survey Hardware.

That tended to hyper inflate Intel / NVIDIA numbers.

Still as it stands the most used GPU is the GTX 1060 (10.75% / 9.89M) followed by the GTX 1050Ti (7.48% / 6.88M).

Well... it somewhat showcases that the PC Gaming Market, esp. that capable of even equalling let alone providing BETTER than Consoles is substantially smaller than you'd think.

Those who can play games with better Graphics / Performance than the One X., is frankly a quite a small market as a whole... let along those who can enjoy Ray Tracing (RTX / DXR) Games, which has a smaller install base than the Kinect v2 that was discontinued after 12 months of launch.

Couple that with PC Gamers being less willing to pay "Full" Retail., and purchasing fewer games on avg. per year... well yeah, there's a reason why we tend to get Half-Arsed Ports on PC; and it's typically seen as a "Bonus Revenue" rather than a Focus Platform for most Studios / Publishers. 

I mean if you get a breakout hit, like say Counter-Strike, LOL, Fortnite, Minecraft, etc. sure... it's amazing as a platform; but notice how NONE of them need anything more powerful than a half-decent Rig from 2010. 

Says a lot about the PC Gaming Market, don't you think?

[colesdav]

RE: while the Series X offers similar performance to an R7 3700 with RTX 2080Ti (albeit almost 2x DXR Performance).
Where did you hear that one specifically the "RTX2080Ti" part.
My guess is a Yoyo Youtuber...

[leyvin]

The performance estimation is based upon a few details., specifically the leaked details on the Xbox Series S.

As that specifically provides (near identical) performance to the Xbox One X (which in turn is the same performance level as a Desktop RX 580 / GTX 1060) and this gives us an excellent starting point regarding estimating what kind of improvements we can expect to see from Navi 2X (RDNA 2.0).

As the Xbox Series S uses a 20CU 1536MHz RDNA 2.0 GPU.

Now this produces ~4.0TF., which generally speaking is a bit of a meaningless number outside of direct comparison of "Same Generation" Performance but in this case because we know that the Xbox Series S is a direct replacement in performance & price terms for the Xbox One X; well this means we can directly compare it to the RX 5500 XT (5.20TF) and RX 580 (6.18TF) as these are also near identical performance.

Now as a keynote., the FP32 TF Performance Metric, while a bit useless in measuring Gaming Performance; is a good metric for determining Raw Compute Performance... thus we can use said differences to measure Instructions Per Clock Gains (or losses, which is the case for all NVIDIA Architectures since Maxwell 2.0) and in turn then be able to estimate the comparable performance of GPUs based upon little more than their Est. Technical Specifications.

In this case Navi 1X offers ~17.22% IPC Gain over Polaris 20., and Navi 2X offers ~26% IPC Gain over Navi 1X (or ~42.9% IPC Gain over Polaris 20)

That means we can actually relatively accurately estimate the expected performance given the Core Count and Core Frequency. 

As we know that the Xbox Series X has 52CU at 1825MHz (12.15TF)… well this give us an excellent basis for the comparative performance we can expect from it using the RX 5700 XT as a base metric.

For example., Shadow of the Tomb Raider at 2160p Max Settings achieves 39.9 FPS Avg. 

52CU is 26% > than 40CU... while 1825MHz is 4.3% < than 1905MHz. (Now keep in mind we're not determining precise Clock Variance., as well my RX 5700 XT being "Golden" Silicon actually runs Shadow of the Tomb Raider at 2040MHz Locked; but also has higher FPS, and I'm not doing an hours worth of Benchmarks at 2am to give more precise estimations given it's only intended as a rough ballpark, so we'll assume more Avg. Silicon Quality) 

Now we can apply these in order to determine roughly what performance the game should have on Xbox Series X.

39.9 + 26% (Cores) - 4.3% (Frequency) + 26% (IPC) = 60.6 FPS

Now the RTX 2080Ti is capable of ~60.3 FPS Avg. on Shadow of the Tomb Raider 

(as a note I'm sourcing all of the data from Eurogamer / Digital Foundry, as they have reliable and comprehensive performance metrics).

As Shadows of the Tomb Raider is one of those games that doesn't really skew in favour of either NVIDIA or AMD in regards to performance., it does make it a good aggregate benchmark title to use; and I'd argue this does mean that we can expect to see similar performance to the RTX 2080Ti from the Xbox Series X.

And I don't really need to justify my claim about the CPU Performance as it's quite literally just an R7 3700 CPU Component., so should have near identical performance. 

•

Now as a note., and this is just my own personal speculation based upon Die Size Comparison... well I believe that it only half of the Texture Mapping Units are Texture Processing Units. 

What the difference between these is., comes down to the Texture Processing Units have the additional Fixed Function Support for Ray Mathematics (Ray BVH and Ray Triangle Processing); they're quite a bit larger than the Texture Mapping Unit and the problem with the Xbox Series X APU is well once you account for the Zen 2 8C/16T CPU... you're not actually left with enough Die Space for the whole Navi 22 (60CU) GPU or even a Custom 56/52CU GPU. 

It actually only fits the remaining space IF half of the TMU don't have the Ray Extensions., and are similar to the Navi 1X TMU size.

As such, I think it only has Half the DXR Processing Capacity of the Desktop Variant... and this does make sense further if you consider the Power Envelope that it has to fit within is more restrictive (165w Total Power Consumption). It's just not doing that with a "Full" Implementation., and while reducing the Frequency could potentially help there; it would be a bigger hit to Traditional Performance, with no guarantee that every game will have / support DXR 1.1 Features.

Based on my analysis (from the revealed Ray Box / Triangle performance metrics)., well the DXR 1.1 Performance should actually be roughly on-par with the RTX 2080 SUPER for the Xbox Series X. 

I will point out that I think that the Xbox Series S does have a "Full" implementation; so all of it's TMU are actually TPU instead... and so while there is approx. at 4x Performance Gap between the two in Traditional Graphics; in regards to DXR 1.1 Performance., there is likely to only be ~40% Differential; this would put it somewhere on-par with the RTX 2060 (maybe 2060) in regards to said Ray Tracing Performance.

As a note this bodes quite well for the RX 6800 (56CU) and RX 6800 XT (60CU) Cards. 

Reason I say this is because AMD have ~18% Efficiency Gain between the Generation (based on the PS5 information)., this means we could see up to 2250MHz at the same power consumption as Navi 1X... that would mean the RX 6800 XT would be a 295w Card at said "Boost" Frequency.

This would result in Shadows of the Tomb Raider (same settings as above) being 77.4 FPS., which if we go by the Digital Foundry (percentile) difference... we should be looking at ~84.6 FPS for the RTX 3080.

That's only an 8-9% Difference to the RTX 3080., while it would be a 25% to the RTX 3070. 

If AMD is sensible about this... they'll accept the bigger difference (sit it between the two at the same 1905MHz Frequency as Navi 10)., but price it aggressively at say $500... because at that point you're talking the same price as the RTX 3070, 20% Better Performance, and running at ~250w.

When we keep in mind that there is also the RX 6900 (72CU) and RX 6900 XT (80CU)., which will frankly sit somewhere between the RTX 3080 and RTX 3090 (which is a rebranded Titan with a massive price drop); but could easily be a $700 - 800 Card.

Well I'd argue I think it starts to make sense why NVIDIA actually rebranded the Titan to reintroduce their x90 Class. 

We'll have to wait and see until next month... but I'd argue things are looking very good, and I could be wrong about the RTX 30-Series performance.

It's very weird that they wouldn't allow direct FPS Metrics to be used yet.

In the next few weeks the embargo will lift and we'll see direct performance; but it really NEEDS to be in the ballpark that I've calculated otherwise NVIDIA could be in real trouble... remember the RTX 20-Series basically sold very similarly to the RX 5000-Series., which looks good for AMD but their figures didn't really go up or down; for NVIDIA it could be considered a Failure. 

In any case., the Xbox Series X looks like great value... and if the RX 6800 XT can launch at the same price point., again that's going to be deeply disruptive; similar to how Zen was when it first launched. 

[colesdav]

The Xbox Series X was reportedly giving ~ RTX2080 level rasterisation performance on Gears 5.
I have run Gears 5 on RTX2080, RXVega 64 Liquid, RX5700XT DX12 4K Ultra, Min FPS at Normal, (no dynamic resolution) and compared them.
Based on that and some "back of the envelope" calculations, given the new TSMC 7nm process only gives +10% performace boost at same power level.
AMD will have pulled off a miracle if they manage to beat a 2080Ti using 8GB of GDDR6 never mind 16GB.
I think it will be about RTX3070 performance level if they use GDDR6 and  better than RTX3070 if they use HBM2e.
That's on DX12 and Vulkan. DX11 performance will likely be the usual story, worse.

[leyvin]

TSMC N7 (7nm) to N7P (7nm+) does provide ~10% Performance Efficiency... but it should be noted that AMD moved from 7nm to 7nm+ EUV, with that providing a further ~20% Performance Efficiency.

The combination of which works out to ~17% Lower Power at the same Frequency, or Higher Frequency at the same Power Usage., the later of which works out to a practical Performance improvement of ~10%.

This however is the SMALLEST improvement between RDNA 1st and 2nd Gen.

As noted., basic calculations comparing architectures that have pragmatically the same gaming performance in regards to Compute Metric showcases the IPC Improvements in the Architecture... if we keep in mind that RDNA 2nd Gen isn't just a refinement of the architecture but also a removal of legacy GCN, which are no longer needed., an even bigger improvement in IPC wouldn't be surprising; and as noted above we're not (as we'd normally be doing) relying on only say a single leaked Futuremark or Passmark benchmark for an Engineering Sample., but instead we have Practical Metrics due to both Microsoft and Sony trying to showcase their Next-Gen Consoles.

Now as for Gears of War 5 showcasing similar performance to an RTX 2080... that's something I've not seen., but that would be an oddly low metric.

I'll put it this way... even RDNA 1st Gen increasing it's Core Count from 40CU to 52CU, at the Lower 1825MHz Frequency should STILL be comfortably out-performing the RTX 2080. So for it not to be doing so, would suggest there is some sort of performance bottleneck with the Coalition Engine and RDNA 2nd Gen Architecture.

[colesdav]

RE: TSMC N7 (7nm) to N7P (7nm+) ...

That was a very long way to say "given the new TSMC 7nm process only gives +10% performace boost at same power level. "

RE: . if we keep in mind that RDNA 2nd Gen isn't just a refinement of the architecture but also a removal of legacy GCN, which are no longer needed.,

Really? It is going to be a complete removal of the GCN instruction set?
I do not think that will happen.
Good luck with the AMD Drivers and games running if they do that.

All of this RDNA2 versus RDNA versus GCN seems to be a marketing attempt to get away from perception that "Vega 64 failed because it is a compute card" story.
Chip design is rarely a case of "abandon everything that went before", just because of the amount of effort involved to get something working.

It is usually successive refinement of what went before.

As far as understand it it, Navi 10 or RDNA mostly implemented architectural features that were supposed to be going into Vega 64 in the first place but did not make it. They still use GCN instruction set architecture.

They also increased on die cache to compensate for using cheaper GDDR memory rather than HBM2(e).
I think RDNA 2 will be RDNA1 ported to new 10% faster TSMC process, some improvements in texture handling, + new Ray Tracing features.

The reason RX5700XT is much worse at Compute than Vega is because it has far fewer cores and it is a much smaller die.
The RX5700XT should have been called the RX690.
Same die size as Polaris.

RE: Now as for Gears of War 5 showcasing similar performance to an RTX 2080... that's something I've not seen.
Gears 5 on Xbox Series X: The Tech Demo Analysed In-Depth!

[colesdav]

I think the XBOX Game Pass subscription will be one reason why PC Gamers are buying less titles.
Epic Games are also giving away Free Titles on a regular basis.

I will not be surprised if "Big Navi" turns out to be the XBOX Series X GPU in a dual fan cooler along with a 3 month free subscription to XBOX Game Pass and a couple of free headline game titles for ~ 350.