• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

On PC (Nvidia) vs PlayStation, and why PS6 has to have 30GB VRAM

The PS6 SOC is known, the GPU is a node shrink/die shrink RX 9070 (non-XT), probably nearly identical to the RX 1060 or 1060XT (successor to the 9060/9060XT). It's going to be well behind an RTX 6070.
Well thats an absolute potato in potential performance better to throw it all in the dumpster and aim for 2030 with a real generational leap.
 
The real question is with that ps6 spec, is $599 price tag achievable maybe on the year 2029 for ps6?
 
Last edited:
Your assumption of 5070 performance is incorrect based on leaks. Estimates we have so far point towards roughly 9070XT level in compute. Based on leaks so far the closet Nvidia GPU would be the 5070Ti.

Assuming 3Ghz clocks (might be higher), the PS6 is going to be 40TF. Between the 9070 and 9070 XT, with the key difference is that VOPD enhancements in RDNA5 will give it a marked improvement over RDNA4, especially in a closed system like a console. So 9070 XT in compute (with Ada/Blackwell RT/PT at least thanks to the new Radiance cores and SER/OMM support) seems likely. Thus, 5070 Ti is the closest Nvidia card.
 
Last edited:
Your assumption of 5070 performance is incorrect based on leaks. Estimates we have so far point towards roughly 9070XT level in compute. Based on leaks so far the closet Nvidia GPU would be the 5070Ti.

Assuming 3Ghz clocks (might be higher), the PS6 is going to be 40TF. Between the 9070 and 9070 XT

If anything, the clocks should be very conservative. It presumably doesn't have the cache that the 9070 has either.
 
Assuming 3Ghz clocks (might be higher)
The leaks say 2.6-3.0. And it has a 640 GB/s VRAM buffer with only 10MB L2. Pumping clocks wouldn't even work.

As for your later points, yes I am expecting Nvidia Level RT/PT. That's why I am using Nvidia cards as the equivalent and not AMD cards.
 
Last edited:
The leaks say 2.6-3.0. And it has a 640 GB/s VRAM buffer with only 10MB L2. Pumping clocks wouldn't even work.

As for your later points, yes I am expecting Nvidia Level RT/PT. That's why I am using Nvidia cards as the equivalent and not AMD cards.
The leak also says 2.5-3 times the performance of a PS5 in regular compute. Even the low end of that is roughly 9070 XT levels.
 
The leak also says 2.5-3 times the performance of a PS5 in regular compute. Even the low end of that is roughly 9070 XT levels.
That part is explicitly extrapolated by him, As he said. He also said PS6 has better ray tracing than 5090. You take in the info and you filter bullshit sort of deal.

Again, his document says that AT2 DT Targets 1.2x 4080, which is around 5080. Lets compare.

PS6 has 0.81x CUs, ~0.87x CLK, 0.75x ROPs, 0.71x MEM BW, 0.42x L2.

Bandwidth is noticeably worse than 70% (given big L2 cut), 70% of compute and around 67% of the front end.

PS6 as 67% of AT2 is a reasonable estimate. Which would be 80% of a 4080 if AT2 DT hits it's target.

5070 is 77% of a 4080, 9070 is 82%, 5070 Ti is more like 98%, etc.

Try comparing RX 9070 with PS6 for a sec.

They have similar TFLOPs and bandwidth, yet PS6 achieves similar or slightly less performance despite:

10MB L2 vs 8MB L2 + 64MB L3
160W vs 220W
3SE/96ROPs vs 4SE/128ROPs

That's what RDNA5 is actually about. Maximizing performance per area and minimizing SRAM use. Everything they are doing should be viewed within that design philosophy.
 
Last edited:
That part is explicitly extrapolated by him, As he said. He also said PS6 has better ray tracing than 5090. You take in the info and you filter bullshit sort of deal.

Again, his document says that AT2 DT Targets 1.2x 4080, which is around 5080. Lets compare.

PS6 has 0.81x CUs, ~0.87x CLK, 0.75x ROPs, 0.71x MEM BW, 0.42x L2.

Bandwidth is noticeably worse than 70% (given big L2 cut), 70% of compute and around 67% of the front end.

PS6 as 67% of AT2 is a reasonable estimate. Which would be 80% of a 4080 if AT2 DT hits it's target.

5070 is 77% of a 4080, 9070 is 82%, 5070 Ti is more like 98%, etc.

Try comparing RX 9070 with PS6 for a sec.

They have similar TFLOPs and bandwidth, yet PS6 achieves similar or slightly less performance despite:

10MB L2 vs 8MB L2 + 64MB L3
160W vs 220W

That's what RDNA5 is actually about. Maximizing performance per area and minimizing SRAM use. Everything they are doing should be viewed within that design philosophy.
I'm just going off the statement by KeplerL2:

PS6 looks around ~9070 XT and Xbox around ~5080

But it's all moot anyway, it's speculation on leaks based on an architecture that hasn't even got final clocks. But 5070 seems low for what Sony is trying to achieve, unless the RT acceleration is revolutionary.
 
PS6 looks around ~9070 XT and Xbox around ~5080
Yes and he later said that Magnus is between 1.3x to 1.5x better depending on the TDP.

5080 is 25% better than a 9070 XT. Not even within his range. Nor does it even make sense for Magnus to only be 25% faster.

Aka he oversold PS6 / was being kepler in minimizing the gap between Magnus and Orion.
 
Last edited:
Yes and he later said that Magnus is between 1.3x to 1.5x better depending on the TDP.

5080 is 25% better than a 9070 XT. Not even within his range. Nor does it even make sense for Magnus to only be 25% faster.

Aka he oversold PS6 / was being kepler in minimizing the gap between Magnus and Orion.
Or Magnus is faster than a 5080.
 
Or Magnus is faster than a 5080.
Look if you want to create fantasy scenarios where PS6 is that fast all the power to you. You can always conjure up enough to believe it.

On planet earth, the range for PS6 is roughly between RTX 5070 to RTX 5070 SUPER. Or from -8% of 9070 to +4%.
 
Look if you want to create fantasy scenarios where PS6 is that fast all the power to you. You can always conjure up enough to believe it.

On planet earth, the range for PS6 is roughly between RTX 5070 to RTX 5070 SUPER. Or from -8% of 9070 to +4%.
It being that fast is hardly fantasy, you make it sound as if we are talking about massive differences here. Even your "high end" estimate is within 10% of a 9070 XT. And the 5070 Super doesn't exist. Nor will it.
 
They are not. Also an auxiliary point so don't derail.

RTX 5070 is a very powerful GPU. The downside of the GPU on PC (only 12GB on such a powerhouse) don't exist here.

Also don't forget that in path traced games, PS6 will be astronomically faster than anything PS5.
None of that matter unless AMD gets their shit together with RT. Nvidia has dedicated hardware for it and it's FG technology on their GPU's chips, they're borderline SOCs at this point.
 
All this talk of a ps6 being around a 9070 or 5070 with better ray tracing performance is quite depressing really.

If that's truly the case, they better release it next year because it's really weak. If they delay till 2028 or 2030 as rumours are suggesting, they need to go back to the drawing board.
 
that would be crazy expensive no?
RFpWoLg88GansrOv.png
 
All this talk of a ps6 being around a 9070 or 5070 with better ray tracing performance is quite depressing really.

If that's truly the case, they better release it next year because it's really weak. If they delay till 2028 or 2030 as rumours are suggesting, they need to go back to the drawing board.

I think the specs were chosen, partially because to keep costs somewhat down. But mainly because the handheld is going to be a requirement, and that's going to be Series S like performance at best.

While Magnus will be a lot more powerful, I also think Sony is thinking FG will be what they will use to close the gap.
 
Last edited:
16 GB on PS5 to 20 GB on PS6 would be a pretty miserable generational leap. I think Sony would rather delay the PS6 in that case.
How would a delay even benefit the PS6 in anyway?
DRAM & NAND Shortages Will Extend Beyond 2030, Warns Phison CEO, Citing an Industry 'Structural Shift' That Will Destroy Consumer Businesses

Sony will most likely secure a long term contract deals below spot prices and raise game and subscription prices.

This may allow the PS6 with 30GB to launch at $700. I highly doubt it would be cheaper than the PS5 Pro launch price tbh.
 
What the hell are they gonna stick in all that memory? I guess a bunch of ray tracing data - but if you have much crappier RT performance - what's the point?
 
How would a delay even benefit the PS6 in anyway?
DRAM & NAND Shortages Will Extend Beyond 2030, Warns Phison CEO, Citing an Industry 'Structural Shift' That Will Destroy Consumer Businesses

Sony will most likely secure a long term contract deals below spot prices and raise game and subscription prices.

This may allow the PS6 with 30GB to launch at $700. I highly doubt it would be cheaper than the PS5 Pro launch price tbh.
they made a small loss on the 499 PS5 on release. The RAM cost them 50 bucks back then. 30GB would cost them 300+ at current industry rates, and that is only if this crisis doesn´t get worse.
And back then TSMC was nearly a 100% cheaper, so the rest of the SOC has gone up very substantially, too. And they`ve announced the next price increases already...Same goes for NVME storage.
700 bucks is not doable with the leaked specs if they are not willing to eat huge losses on each system or something drastic happens, like the AI bubble bursting and "resetting" hardware-demand in the process.

Also, there are no long term deals with fixed pricing in this industry. Contracts only guarantee that you have the chance to buy, but the prices get re-evaluated every quarter usually and they have to compete with all the big AI players here.
 
Last edited:
Reports indicate that Mark Cerny gave PS6 30GB of unified memory; A ludicrous quantity of premium memory.

It's highly unlikely that he did it for the reasons that AMD misinformation peddlers spout (memory is cheap, 8GB GPUs should be free, you will need 16GB VRAM to play games, game developers actually need it, etc).

It's a calculated Attempt to shut 16GB PC GPUs out of the high-end gaming tier entirely. He compromised on raw GPU performance (~RTX 5070) and CPU (~AI 7 350 level) to afford it. That trade-off is the whole game.

Here's the question worth asking:


It's 30GB.

Before we begin, some facts about PC one needs to Accept before. If you disagree with these, fair enough, but don't derail the discussion based on them as these are auxiliary.

With 93%+ market share, Nvidia is the only game in town. It doesn't matter if a niche budget brand (AMD Radeon or Intel Arc) offer a larger VRAM buffer at an affordable price, if nobody's buying them, they don't define the platform. PC Gaming is synonymous with GeForce.

AMD's Radeon has a long history of being delusional about where their GPUs actually land and how they perform relative to Nvidia. This is important because that's the frame of reference you need to adopt when translating the grand performance claims of console vs PC.
PS5 was not equivalent to a 3060 Ti whatsoever or even on the same planet as it. It was equivalent to a 6600 XT to a T.

Now that we got this out of the way, let's divide PC into multiple tiers.

(RTX 60 Series is guesswork/illustrative)

OBMHzTiVPlFkYaOL.jpeg


You can immediately see that Cerny tried a version of this last generation.

PS5 launched with ~RTX 3060 raster performance and ~9GB of usable VRAM on the high end, and the vision was this:



PC gamers were having none of it. A lot of them owned 8GB GPUs that genuinely outperformed PS5 in raw rasterization to high hell, and they weren't going to accept being told their hardware was substandard and that they should just run low settings at 400FPS. Despite the AMD influencer ecosystem (MLID, Hardware Unboxed's Steve, etc) pushing hard for Cerny's framing, developers blinked after multiple disaster launches.

They burned an extra gigabyte of RAM on PS5 CPU / underutilized PS5's SSD. because PC was genuinely a bigger platform than high end console at that point. Even though PS5 ate up Xbox, high end console market shrank while PC grew.

The result: the number of games where 8GB is not enough for PS5+ settings (slightly less on textures, significantly better in VRAM light settings) can be counted on two hands. Cerny lost that round.

PS5 Pro was a concession in a way. Since PS5 couldn't really ray trace, so it was being cut out of that side of the generation.

30GB Is the Rematch

This time the gap is massive. That's the entire moat of PS6.

18GB VRAM+ should be mostly safe given the sheer market mass. It's doable engineering work for developers to work around using more CPU side memory and tweak compression to get 20-22GB into 18GB.

16GB is the battleground because Nvidia has so many GPUs there. It ranges from mass market cards (RTX 6060 TI 16GB) to cards that dumpster the PS6 in raw perf (RTX 5080).

Cerny can't afford for those GPUs to be able to match PS6. Because otherwise PC has had RTX 4080, a PS6 equivalent, since 2022.

Which is why he has to wait for lower memory prices to launch with 30GB. He can't launch with 20GB

Nvidia's counter will be straightforward again: flood the market with 16GB cards fast enough and cheap enough that developers can't afford to ignore them. If 16GB becomes the dominant install base at the high end, developers will cater to it the same way they catered to 8GB last gen.
Get Over It GIF
 
they made a small loss on the 499 PS5 on release. The RAM cost them 50 bucks back then. 30GB would cost them 300+ at current industry rates, and that is only if this crisis doesn´t get worse.
And back then TSMC was nearly a 100% cheaper, so the rest of the SOC has gone up very substantially, too. And they`ve announced the next price increases already...Same goes for NVME storage.
700 bucks is not doable with the leaked specs if they are not willing to eat huge losses on each system or something drastic happens, like the AI bubble bursting and "resetting" hardware-demand in the process.

Also, there are no long term deals with fixed pricing in this industry. Contracts only guarantee that you have the chance to buy, but the prices get re-evaluated every quarter usually and they have to compete with all the big AI players here.
Where did I say it's going to be a fixed price?

The problem is many of you are thinking Sony is going to pay the same price as we see in stores.

Not only that, GPU prices barely increased from their launch prices. It's mostly DDR that have seen huge price increases due to AI demand.
 
Where did I say it's going to be a fixed price?
"Long term contract deal... "sounded like it, guess I misunderstood.
The problem is many of you are thinking Sony is going to pay the same price as we see in stores.
Nope. That's why I explicitly stated "industry rate" not consumer prices. And I'm already at the low end of the current fluctuations with my numbers.
Not only that, GPU prices barely increased from their launch prices. It's mostly DDR that have seen huge price increases due to AI demand.
Have you slept under a stone for the past 2 years?
The DDR is just the latest issue with Nvidia reacting by basically dropping GPU production instead of accepting a hit to their very hefty profit margins, which base console hardware usually doesn't have btw.
 
Last edited:
The leak also says 2.5-3 times the performance of a PS5 in regular compute. Even the low end of that is roughly 9070 XT levels.
Going by MLiD leak is still tricky due to the RT uplift.

tAxS8gTfhx5TTQSL.jpg

PS5 is more or less equivalent to a 3060 or 6700.
Performance vs 5090:
Raster: ~21%
Ray tracing: ~14%

PS6 (Estimated)
~2.5–3× PS5 raster performance
~6–10× PS5 ray tracing performance
Performance vs RTX 5090:

LEXRu1OhTUICQk5r.jpg

Rasterization:
Lower bound: 2.5 × 21% = ~52.5% = 9070XT
Upper bound: 3 × 21% = ~63% = 5080
Keep in mind this most likely includes RDNA5 IPC gains.

pOKtvhIRZl9HUInO.jpg

Ray Tracing:
Lower bound: 6 × 14% = ~84% high than 4090
Upper bound: 12 × 14% = ~168% higher than 5090

PT-weighted overall (35% raster, 65% RT/PT)
Lower bound overall:
Raster: 52.5% × 0.35% = 18.375%
RT/PT: 84% × 0.65% = 54.6%
Total: 18.375% + 54.6% = 72.975% ≈ 73% = 5080

Upper bound overall:
Raster: 63% × 0.35% = 22.05%
RT/PT: 168% × 0.65% = 109.2%
Total: 22.05% + 109.2% = 131.25% ≈ 131% (above a 5090, which is crazy to think)

Average estimate (most realistic)
Mid raster: (52.5% + 63%) ÷ 2 = 57.75%
Mid RT/PT: (84% + 168%) ÷ 2 = 126%

Raster: 57.75% × 0.35% = 20.2125%
RT/PT: 126% × 0.65% = 81.9%
Total: 20.2125% + 81.9% = 102.1125% ≈ 102%

PS6 overall vs RTX 5090:
Lower bound: 73% (5080)
Expected: 100–105% (5090)
Upper bound: 131% (5090Ti?)

Now this is obviously with all the bells and whistles (PSSR, Radiance Cores, etc.) enabled. And I can see why MLiD was saying the PS6 has 5090 RT performance. This is why PS6 RT uplift is tricky.

The real question is if this is all custom work by Sony or if this applies to Xbox Magnus and it is ~25% faster.

Imagine the 6090 RT performance.
 
"Long term contract deal... "sounded like it, guess I misunderstood.

Nope. That's why I explicitly stated "industry rate" not consumer prices. And I'm already at the low end of the current fluctuations with my numbers.

Have you slept under a stone for the past 2 years?
The DDR is just the latest issue with Nvidia reacting by basically dropping GPU production instead of accepting a hit to their very hefty profit margins, which base console hardware usually doesn't have btw.
Go and check GPU launch and current prices for yourself.

Nvidia dropping some of their GPU lineup has more to do with available chip wafer capacity than RAM availability imo.
 
Going by MLiD leak is still tricky due to the RT uplift.

tAxS8gTfhx5TTQSL.jpg

PS5 is more or less equivalent to a 3060 or 6700.
Performance vs 5090:
Raster: ~21%
Ray tracing: ~14%

PS6 (Estimated)
~2.5–3× PS5 raster performance
~6–10× PS5 ray tracing performance
Performance vs RTX 5090:

LEXRu1OhTUICQk5r.jpg

Rasterization:
Lower bound: 2.5 × 21% = ~52.5% = 9070XT
Upper bound: 3 × 21% = ~63% = 5080
Keep in mind this most likely includes RDNA5 IPC gains.

pOKtvhIRZl9HUInO.jpg

Ray Tracing:
Lower bound: 6 × 14% = ~84% high than 4090
Upper bound: 12 × 14% = ~168% higher than 5090

PT-weighted overall (35% raster, 65% RT/PT)
Lower bound overall:
Raster: 52.5% × 0.35% = 18.375%
RT/PT: 84% × 0.65% = 54.6%
Total: 18.375% + 54.6% = 72.975% ≈ 73% = 5080

Upper bound overall:
Raster: 63% × 0.35% = 22.05%
RT/PT: 168% × 0.65% = 109.2%
Total: 22.05% + 109.2% = 131.25% ≈ 131% (above a 5090, which is crazy to think)

Average estimate (most realistic)
Mid raster: (52.5% + 63%) ÷ 2 = 57.75%
Mid RT/PT: (84% + 168%) ÷ 2 = 126%

Raster: 57.75% × 0.35% = 20.2125%
RT/PT: 126% × 0.65% = 81.9%
Total: 20.2125% + 81.9% = 102.1125% ≈ 102%

PS6 overall vs RTX 5090:
Lower bound: 73% (5080)
Expected: 100–105% (5090)
Upper bound: 131% (5090Ti?)

Now this is obviously with all the bells and whistles (PSSR, Radiance Cores, etc.) enabled. And I can see why MLiD was saying the PS6 has 5090 RT performance. This is why PS6 RT uplift is tricky.

The real question is if this is all custom work by Sony or if this applies to Xbox Magnus and it is ~25% faster.

Imagine the 6090 RT performance.

There is no way in hell that it will reach anything close to 5090 real world performance even if it has that RT performance on paper (rest of the GPU won't keep up).

Same way that PS5 Pro that was rumored 2-4x better RT performance ended up with ~50% in the real world? I think that's what the AC devs presentation showed. Even with RDNA4 RT parts rest of the GPU will limit it.
 
Go and check GPU launch and current prices for yourself.

Nvidia dropping some of their GPU lineup has more to do with available chip wafer capacity than RAM availability imo.
It's both. In the end it's about the margin. GPUs had a ridiculous margin priced in from the get go which caught a lot of the bullshit so far but the RAM crisis was just one hit too many.
If you are limited by parts or production capabilities you choose the most profitable path which is AI hardware.
And of course this doesn't change the fact that the ridiculous price increases for RAM and parts in general hit everyone, including Sony which has to compete with the likes of Nvidia for parts and production capacity now. With the current prices and price trajectories and the leaked specs 700 bucks wouldn't be doable without heavy subsidies and there's no reason for that when you have no competition.
 
Last edited:
There is no way in hell that it will reach anything close to 5090 real world performance even if it has that RT performance on paper (rest of the GPU won't keep up).

Same way that PS5 Pro that was rumored 2-4x better RT performance ended up with ~50% in the real world? I think that's what the AC devs presentation showed. Even with RDNA4 RT parts rest of the GPU will limit it.
Did you read this part?

Now this is obviously with all the bells and whistles (PSSR, Radiance Cores, etc.) enabled. And I can see why MLiD was saying the PS6 has 5090 RT performance. This is why PS6 RT uplift is tricky.
 
Cancel no, delay maybe.

I'm assuming that NV is going to take most of the 3 GB GDDR7 capacity for Rubin CPX this year and next. It's going to be tough to compete for the leftovers.
When will you guys understand that DRAM prices will continue to rise for the next decade (over the next 10 years).

How is a delay going to help?

Just release the PS6 as a premium device since the PS5 life cycle is expected to be longer.
 
Did you read this part?

Now this is obviously with all the bells and whistles (PSSR, Radiance Cores, etc.) enabled. And I can see why MLiD was saying the PS6 has 5090 RT performance. This is why PS6 RT uplift is tricky.

You think 9070XT class GPU will reach 5090 levels when PSSR+RT is enabled vs. DLSS+RT on 5090?

If we covert 9070XT power to nvidia equivalent we end up with 5070ti:

wNk49OxElf22T138.jpg


That's the difference for raster. Here RT:

huX5W3LqOaFpxHiX.jpg


If we assume that RDNA5 is more peformant in RT than Blackwell (possible of course), do you think it's enough to cover almost 2x difference in power?
 
Top Bottom