xool
Member
Yeah that's what I meant - betting chances 3:1 on, roll over 1 with d4 etc
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
- Status
Yeah that's what I meant - betting chances 3:1 on, roll over 1 with d4 etc
SonGoku
Member
I agree 60CUs max (4 disabled for yields), we need however news on Navi to see how high (or low) it clocks before discarding 1700-1800 clocks, atm im cautiously optimistic my current expectations are
Low end: 11TF
Mid end: 12TF (most likely in between 12-13)
High end: 13TF
Last edited:
Insane Metal
Member
CyberPanda
Banned
With AMD heat is always an issue.
SonGoku
Member
ResilientBanana
Member
14TF 
People really think that we are getting a GPU with nearly maxed out CU’s? E3 is going to be so disappointing for you.
Be prepared for $1,000 consoles
People really think that we are getting a GPU with nearly maxed out CU’s? E3 is going to be so disappointing for you.
Be prepared for $1,000 consoles
Last edited:
SonGoku
Member
*FP16
Last edited:
CyberPanda
Banned
People need to temper their expectations, or mass disappointment will happen.
CrustyBritches
Gold Member
I was taking a trip down memory lane today with the pr cycle leading up to the release of Polaris 10. So many similarities with stuff like "historic leap in perf/watt", "densest fab available", "2.5x perf/watt", "this is only the beginning for Polaris perf/watt optimizations". Even Vega 56/64 roughly sit in the same price and tdp bracket as the 390/390x.
If you had a Polaris 10 card you were lucky given the mining performance, but the actual gaming performance and power consumption was not as ideal as marketing would have lead you to believe. I would just advise to remember AMD GPUs usually end up more power hungry than they initially hype.
If you had a Polaris 10 card you were lucky given the mining performance, but the actual gaming performance and power consumption was not as ideal as marketing would have lead you to believe. I would just advise to remember AMD GPUs usually end up more power hungry than they initially hype.
SonGoku
Member
Considering all that 6TF X was mighty impressive, that's why i feel 12TF is very likely.
StreetsofBeige
Gold Member
LOL.*FP16
PS5 is more powerful than Anaconda. Cerny: Gamers, PS5 is the most powerful system at 13TF. Sweet raw power. You'll be amazed at Knack 3
Anaconda is more powerful than PS5. Cerny: Gamers, calm down. It's not all about raw TF. If devs code their PS5 games using F16, it's more TF than Xbox
Last edited:
SonGoku
Member
Frankly i dont care who's more powerful but i hope both target the ceiling of what's posible in a console come late 2020.
Remember the higher the baseline the better.
btw both will support fp16 this time around so devs will rely on it more extensively, should bring in up 20% performance improvements.
Last edited:
CrustyBritches
Gold Member
I expect between Vega 56 and Vega 64 performance, so 12TF power capped and voltage tuned could fit into that performance bracket(1070ti, 1080, RTX 2060, etc).
SonGoku
Member
That's only taking into account peak compute numbers, when you factor in fp16, tbr and vrs 12tf would punch above its weight compared to those cards.
xool
Member
There's some cheap maths you can do to show that 8TF is actually a reasonable expectation for next gen.
I just compare roughly equivalent first AMD GPUs at each new node - FP32 roughly doubles from 28 to 14 nm
[everything in the third row is a guess]
| Card | TDP / W | FP32 / TF | node / nm | area / mm2 | |
| 7850 | 130 | 1.8 | 28 | 212 | PS4 'equivalent' |
| RX 470 | 120 | 3.8 | 14 | 232 | |
| ?? | 125 | 8 (estimated) | 7 | ~220 | PS5 'equivalent' |
That puts it in Vega56 teritory, but at a smaller size, so uses much less power.
You can swap the 7850 for a 7870 (2.5TF 175W) and the 470 for a 570 (4.8TF, 150W) to get next gen at 10TF ~ 160W GPU .. so maybe somewhere in between.
[I think these are low ball estimates though - should get an extra +10% TF for free just from clock increases at 7nm]
Something will have need to improved to get us to
Last edited:
bitbydeath
Member
There’s math for 14TF too.
xool
Member
For TF I just assume linear scaling for specs.. so from 28 to 14 nm TF increases 2.1x (3.8/1.8) - so for 7 nm I just multiply 3.8 x 2.1 ~=8 TF .. (transistor count also roughly doubles from 28 to 14nm)
I'm trying to compare like for like as much as possible - so I pick cards with similar die sizes / transistor count / power .. and assume that the PS5 follows on a straight line graph, but uses the approximately same die size and power usage as PS4 .. it's a very rough estimate
Last edited:
SonGoku
Member
bitbydeath
Going by the latest AdoredTV chart
RX 3090 60 CUs at 180W falls within the limits of what could be used on a console (The X equivalent part was 180W)
That would provide close to 14TF numbers though realistically i don't expect more than 13TF
bitbydeath
Member
Going by the latest AdoredTV chart
RX 3090 60 CUs at 180W falls within the limits of what could be used on a console (The X equivalent part was 180W)
That would provide close to 14TF numbers though realistically i don't expect more than 13TF
Yeah, I’m not sure expecting it since that is on the absolute high-end of things.
SonGoku
Member
I mean Sony would have to try really hard to make the GPU underperform to 8TF
A 60CU GPU (4 Disabled) will fit in nicely on a 350mm2 APU die (roughly the same size as launch PS4)
Navi would have to clock really bad to get 10TF let alone 8TF, that's why i feel ~12TF its the most likely outcome.
Last edited:
TLZ
Banned
Well Sony love their loud fans, so why not?
Negotiator
Member
Next-gen consoles better have vapor chamber cooling and Nidec fans as a baseline feature.
CyberPanda
Banned
WIth their BOM budgets good luck with that.
CrustyBritches
Gold Member
That was my point with my original comment.
Despite all the architectural advantages, 6TF Polaris 10 performed like ~5.2TF R9 390, it just did it at 167W instead of 282W.X1X is like a RX 480 overclocked, power capped, and voltage tuned to sub-160W power draw with more memory bandwidth.
Consoles will get a GPU equivalent to the 150-160W AMD listed TDP Navi cards(160-170W actual draw). That should land in the Vega 56/64, 1070ti, 2060 performance tier according to AdoredTV's updated chart.
SonGoku
Member
What do you base this on?
xool
Member
Well Sony love their loud fans, so why not?
They need to fix this. Also battery life. I bet they screw up at least one.
Going by the latest AdoredTV chart
RX 3090 60 CUs at 180W falls within the limits of what could be used on a console (The X equivalent part was 180W)
That would provide close to 14TF numbers though realistically i don't expect more than 13TF
Going by 'equivalent' watts (~150W max) and price (~$250) from previous Sony gen - puts us in RX3070/3080 territory again - another low ball estimate
True - they are absolute bottom estimates.
14nm Zen CCX (4 cores) is 44mm2, so 8 cores is 88mm2 (that's without IO stuff - but that's duplicated on GPU, so don't count twice) - reduces by ~2.5-3x at 7nm (estimate) - so a tiny 30 to 35 mm2 ..
It's shocking just how small 8 core Zen is compare to GPU
CyberPanda
Banned
These companies can only do so much with their BOM budgets. I’m sure the cooling solution for ps5 will be similar to xbox’s. Hopefully anyway.
xool
Member
I honestly thought we already had 3D audio - plenty games had it, and the APU already some sort of hardware audio support (supposedly AMD TrueAudio ASIC equivalent) [wikipedia] ..
[edit - this http://www.redgamingtech.com/playstation-4-audio-dsp-based-on-amds-trueaudio-technology/ ]
Not complaining about more sauce though.
Last edited:
CrustyBritches
Gold Member
I've had a shitload of AMD cards including R9 390 PCS+ and RX 480 Red Devil. 390 PCS+ has superior gaming performance.
You can get X1X's 4k/30fps FH4 result with RX 480 1305MHz(core)/2150MHz(mem), -12% power limit, and slight undervolt. Average core speed is 1125MHz with mem clock balls to the wall. 156W peak draw, average much lower. 150-160W peak draw is probably a best-case for next-gen GPUs.
xool
Member
This might mix things up a bit : https://www.anandtech.com/show/1422...echnology-7-nm-with-higher-transistor-density - they've got a 6 nm process that is supposed to by an easy migration from 7 nm ..
I wonder if either MS/Sony might start out with a BIG chip, on the assumption that they can get ~20% reductions as soon as 6 nm is ready.. makes me thing back to PS3/360 gen when both launched big and (too) hot on 90nm, and moved rapidly (1/2 years) to 65 nm ..
I hope we get a big jump to next gen like back then, but without consoles shitting the bed because too hot .. transition to this gen felt more like a hop than a leap power wise.
I wonder if either MS/Sony might start out with a BIG chip, on the assumption that they can get ~20% reductions as soon as 6 nm is ready.. makes me thing back to PS3/360 gen when both launched big and (too) hot on 90nm, and moved rapidly (1/2 years) to 65 nm ..
I hope we get a big jump to next gen like back then, but without consoles shitting the bed because too hot .. transition to this gen felt more like a hop than a leap power wise.
SonGoku
Member
But you can't go off these different type of scenarios to decide where it will fall
Sony doesn't go off retail price when they choose the GPU, pricing of the card is entirely dependent on die size, like pointed at before a 60 CU card will fit in nicely on that 350mm2 APU budget.
Final TF count depends on how high Navi clocks and the cooling solution Sony chooses.
Last edited:
Negotiator
Member
Jaguar was equally small at 28nm vs the GPU.
Don't be surprised if they reduce the L3 cache on consoles. It's gonna be Zen 2 Lite/Mobile most likely.
SonGoku
Member
No offense but im skeptical, this is the first time i hear of the X GPU under performing a rx 480 most (including DF) put it up there with the 580
The X is limited by its CPU so fps count wont tell the whole story without diving into every effect used and its effect on performance
Also despite better arch it makes sense for R9 to outperform the rx480 at 4k considering the bandwidth gap and similar TF count.
idk about reducing cache but its going to be a mobile variant for sure hence the conservative 3Ghz+
But Jaguar was trash netbook grade CPU, what makes it surprising this time around is that an actual decent CPU doesn't take much space.
Last edited:
CrustyBritches
Gold Member
No offense taken, my friend. Going back to my original trip down memory lane with Polaris 10, it's like deja vu all over again.
That's why I prefer this approach, because it's something that has precedent. X1X performs like a tuned RX 480 with more memory bandwidth. I have a 2016 Polaris 10, and the memory oc wasn't great on card I got. I get max 2135MHz mem clock, while good cards and later models(like RX 580) could sustain 2250MHz. RX 480's FH4 result is memory bandwidth limited. When you chase X1X's performance with higher core speed/compute, you find Polaris 10 has nasty oc power draw scaling. Like a 30% increase in power draw for 6% increase in performance.
Last edited:
SonGoku
Member
Remember the RX 480s 5.8 TF is not far off from 6TF, what you are describing is a fine tuned 580 (similar peak performance more bandwidth)
Thats true but i would put the blame on clocking these cards beyond the performance per watt sweet spot thus hitting diminishing returns.
Keep in mind yields affect clock speed as well, the 580 likely benefits from this along with the X releasing a year later and all.
Last edited:
Negotiator
Member
It doesn't take much space because 7nm has much higher density than 28nm. What's so surprising about it?
Imagine if AMD had Ryzen CPUs back in 2012 (when the console APUs were taped out). How many Zen cores do you think they could reasonably fit in a 28nm 350mm2 APU?
I say 2 CPU cores max (for a 28nm APU). 14nm APUs have a quad-core Ryzen and 7nm APUs will have an 8-core Zen 2 CPU.
2 Zen cores at let's say 2 GHz would yield much higher single-threaded performance, but multi-threaded performance would be equal or worse to 8-core Jaguar.
It's the same thing with dual-core i3 CPUs: https://gamefaqs.gamespot.com/boards/916373-pc/73290375?page=4 (probably even worse these days due to security mitigations)
SonGoku
Member
N
Negotiator
Yeah i get it, technology advances and all that jazz... still before knowing Zen2 at 7nm would use the same die budget as jaguar did on 28nm i was a bit worried these better cpus would eat into the GPU budget compared to launch PS4 or that they would only use 4 cores. They do not hence why im pleasantly surprised.
Last edited:
SonGoku
Member
Very interesting indeed, from the article:
As soon as is ready for high-volume manufacturing console manufacturers could port their current designs over with minimal investment.
| Launch die size at 7nm | "6nm" die size (15% reduction) |
|---|---|
400 mm2 | 340 mm2 |
390 mm2 | 331.5 mm2 |
380 mm2 | 324 mm2 |
360 mm2 | 306 mm2 |
350 mm2 | 297.5 mm2 |
Hidden CELL 2 confirmed?
Last edited:
Negotiator
Member
Now that you mention it, I'm wondering about its potential die size/TDP at 7nm (assuming it's scalable, last time I checked they stopped at 32nm).
Back in the day, Cell would have been an I/O processor in the next-gen console and BC would be guaranteed...
CrustyBritches
Gold Member
No, I'm describing a RX 480 with underclocked core(1125MHz/5.2TF), overclocked mem, and twice the l2 cache. Like a 390. That's why 180W RX 580 isn't a great comparison.
Anyway. Same loop. Let's bet on the performance. I will bet you I can get a Navi 10 card from 2019 to beat the 2020 PS5. Sound good?
SonGoku
Member
If you can prove a stock rx 480 surpasses the X or conversely a downclocked 480 at 5.2tf matches it. I'll take your word for it or alternatively a downclocked rx 580
Just to be clear, you are implying a navi card with lower tf count will beat ps5?
Last edited:
CrustyBritches
Gold Member
I thought you listed the adored charts and stated they will use navi 20 180W card. I'll just use Navi 10 cards from 2019. You're saying Radeon VII performance and I'm saying between Vega 56 and 64. For a month of NeoGAF gold. Oh yeah, baby.
SonGoku
Member
I expect a fine tuned RX 3090 but Sony could go for 56CUs to increase yields
The bet im willing to make is that PS5 GPU will match Navi cards at the same teraflop count
Last edited:
- Status