A new week and a fresh Monday brings with it the hopefully welcome prospect of a new episode of DF Direct Weekly – and this week, the team sit down to discuss topics as diverse as Starfield’s planned performance upgrade for Xbox Series X, the latest Switch 2 rumours, more path tracing in Capcom RE Engine titles and some new information on the PlayStation 5 Pro’s GPU – such as a max clock speed of 2.35GHz.
It’s the PS5 Pro graphics details I’m going to concentrate on today because the information casts an interesting new light on the upcoming console – and may deliver clarity on some of the question marks surrounding GPU performance and backwards compatibility with the existing PS5. Leaked specifications, derived from Sony’s developer portal, suggest that the PS5 Pro has 30 WGP (Work Group Processors) delivering 33.5 teraflops of performance. This is up against the standard model with 18 WGP offering up an equivalent 10.23 teraflops.
On the surface level, that’s an extra 227 percent of performance, except that the same Sony documents suggest only an extra 45 percent of actual game throughput. Part of the explanation comes from the RDNA 3 architecture with its dual-issue FP32 support, which doubles the amount of instructions processed, but which does not typically double game performance.
- 0:00:00 Introduction
- 0:01:41 News 01: Bethesda announces big Starfield update
- 0:23:05 News 02: AMD sees massive gaming revenue decline
- 0:40:14 News 03: New PS5 Pro GPU details!
- 0:58:27 News 04: Switch 2 rumour roundup
- 1:07:40 News 05: RTX Remix getting DLSS 3.5 Ray Reconstruction
- 1:16:02 News 06: Resident Evil titles get path tracing!
- 1:27:14 News 07: AMD Strix APUs pack great integrated GPU performance
- 1:35:06 Supporter Q1: Could Nintendo games use DLSS 2 to reach 4K output on Switch 2?
- 1:40:23 Supporter Q2: What features would you add to Switch 2, if you could pick anything?
- 1:45:25 Supporter Q3: Could the potential merging of Xbox and PC development hurt Xbox?
- 1:50:13 Supporter Q4: Could developers run a game’s logic at high rates to improve responsiveness, while keeping frame-rate untouched?
- 1:54:19 Supporter Q5: Gaming handhelds theoretically seem as fast as a Series S. Why are they slower in practice?
- 1:57:40 Supporter Q6: Could Valve build a viable console platform to compete with Sony?
- 2:03:16 Supporter Q7: If you had to pick between #StutterStruggle and FSR 2 artifacts, which would you pick?
However, beyond that, there has been confusion about backwards compatibility support with the standard PS5. We know that the PS5 has 36 compute units (two CUs per WGP), running at a maximum of 2.23GHz. However, ‘reverse-engineering’ PS5 Pro’s 33.5TF figue suggests a 2.18GHz clock from its 60 CUs, lower than that of the standard model. Something doesn’t quite make sense then, with some suggesting that the new console actually has 56 CUs, with four disabled, which would deliver a higher clock to hit that 33.5TF – and perhaps in the process provide a hardware balance that’s a better fit for PS5 game compatibility.
Recent details we’ve seen – a part of Sony’s recent developer disclosures – seem to tidy up the discrepancies to a certain degree. The standard PS5 features 18 WGP over two shader engines/four shader arrays in a 5-4-5-4 configuration, while the Pro does indeed feature 30 active WGPs over two shader engines/four shader arrays in an 8-7-8-7 set-up. Both consoles, therefore, feature four deactivated CUs – and this is confirmation of 60 CUs in the new machine.
This means that the stated 33.5TF does indeed suggest a slightly lower clock-speed for the GPU in the region of 2.18GHz – which may be the case in general operation, but the new information also reveals that the PlayStation 5 Pro can boost higher than its standard counterpart, to a maximum of 2.35GHz (a theoretical maximum of 36.1TF). However, similar to the original PS5, system performance is limited by a power ceiling, so it’s rather rare for the GPU to hit that maximum and only certain games will boost that high. Bearing in mind the surfeit of compute power for standard PlayStation 5 games, we must assume that the slight reduction in general clock speed (just under three percent) likely makes no difference, while the ‘ultra boost’ mode should instead deliver much more graphics throughput for existing titles.
The cache structure of the new GPU changes in some areas. The 4MB of L2 cache per WGP remains the same, while L1 doubles from 128KB to 256KB to accommodate the larger numbers of compute units per shader engine. L0 cache also improves from 16KB to 32KB, which Sony says is to accommodate higher ray tracing performance.
Elsewhere, the PlayStation 5 Pro’s GPU evolves to include DirectX12 Ultimate features that were omitted from the original console – so, hardware support for variable rate shading is included, along with extra features for hybrid MSAA. The primitive shader features found in RDNA 1 and the vanilla PS5 are augmented with full support for mesh shaders, which should hopefully see more widespread adoption of a very useful feature. Sony points out how the stricter requirements of this feature up against PC APIs are no longer an issue.
Stepping back from the specifics, it’s still surprising to me that a 67 percent increase in compute units only translates to around 45 percent of extra performance, but similar to PS4 Pro vs PS4, increased compute does not scale in a linear fashion with increased game performance. Memory bandwidth only scales by around 29 percent from PS5 to Pro, for example. I think we’ll need to look at the Pro more holistically, judged by its actual results, and PSSR upscaling could be just as transformative for Sony as it has been for Nvidia DLSS.
Extra compute power paired with machine learning-based reconstruction have proven to be a potent combination in the PC space and both working in tandem should show a clear and noticeable improvement over the standard console in terms of both image quality and performance. Sony also appears to be offering developers an easier route forward in retrofitting existing PS5 games with PSSR – and I would expect its required inputs to be similar to those used by FSR 2. So, extra resolution and a more potent reconstruction technique could be game-changing – especially for those 60fps performance modes that the vanilla PS5 is struggling with. The ability to add PSSR to existing games without migrating onto the latest Sony SDK is also going to help immensely in revisting those older titles to add support.
There are clearly lessons learned in the Pro’s design too: I rather liked the PlayStation 4 Pro in its day, but there were issues and challenges for developers. Too many games simply delivered the same, or slightly tweaked experiences as the vanilla PS4 version – just running at a higher resolution (1440p vs 1080p proved commonplace). Developers I’ve spoken to were dismayed at having to service a 4x denser resolution with only an extra 512MB of memory, and weren’t entirely convinced by Sony’s checkerboard rendering tech. PS5 Pro looks like an altogether more solid package: enough compute power to increase resolution if needed, but a superior reconstruction technology that should hopefully be able to deliver convincing 4K results from 1080p inputs.
This time around, without a new wave of displays to justify a console upgrade, Sony has it all to prove in pitching the idea of a PlayStation 5 Pro – but I also remember being impressed with the results of PS4 Pro’s 4.2TF GPU once I saw the demos in the flesh and I genuinely can’t wait to see what this new machine is capable of.