Support NeoGAF

[SonGoku]

It doesn't sound unrealistic if as a result of the 7nm EUV process, the power efficiency makes allowances for the 2GHz clock.

I don't know SonGoku, I pray that it's at least 48. But Leadbetter, DF, Gonzalo, it's beginning to add up with a fair bit of consistency.

Makes no sense to clock all the way up to 2GHZ and lowball CU count, complete opposite of console design paradigms
They'll get a more efficient design by compromising a little on clock speeds, say 1.8GHZ and go with 48CU or if that's too power hungry, 56CU - 1.6GHz

Not to mention EUV density increase enables a 64 to 72CU part.

Suit yourself but I'm fallowing komachi leaks. Things constantly add up but you choose to ignore it.

Its the other way around actually: mental gymnastics, speculation and constant mix and matching
I treat it for what is: rumor and speculation.

 

[llien]

Dude I don't know what the fuck you are talking about - the comparison was with a 2070 not 2070 super ..

No problem:

NVIDIA GeForce RTX 2070 Founders Edition Review

GeForce RTX 2070 is NVIDIA's most affordable RTX-capable graphics card today. We ran it through 23 games at three resolutions, and the performance results are encouraging. The card sits right between the GTX 1080 and GTX 1080 Ti, with pricing lower than for the GTX 1080 Ti, and you get all the...

www.techpowerup.com

...not continuing this ..

I'd appreciate if you'd stop the denial. and detach your ass from nvidia's.
At the moment it looks like they are entangled for some reason.

 

[ZywyPL]

36cu* 2ghz =9,2TF

9.2 RDNA TFlops, which is equivalent to GCN's ~14TF, so we're talking about full-fat Radeon VII performance level if true.

That would make sense for BC, since they are the clocks of the each "gen" on the platform/ecosystem.

Gen 0 profile is PS4 clock.
Gen 1 profile is Pro clock.
Gen 2 profile is PS5 clock.

Yeah, that seems more thsn obvious, so the only question remains whether the BC will use full amount of PS5 CUs to give us rock solid FPS and resolution, or the PS4/Pro games will use the exact same amount of CUs as on their original platforms (18/36), and drop performance as the end result?

 

[xool]

No problem:

I'd appreciate if you'd stop the denial. and detach your ass from nvidia's.
At the moment it looks like they are entangled for some reason.

My posts were in good faith. I'm not pushing Nvidia stock. Maybe if you had just stated/explained the facts like this :

You wouldn't know unless you've had one (and I don't mean that in an elitist way.) My nvidia GPU is listed with a 'boost clock' of 1680 MHz, but will genuinely run around 1845-1860 MHz not overclocked (and sliders left at default.)

Would have been a lot simpler

ok so technpowerup says an average 1860-1875MHz clock for the 2070 FE vs what I assume is 1725MHz for the RX 5700 .. so that's ~+8% clock advantage for Nvdia.

It's still not the home run I would like it to be.. against the 2 year old (is that all it seems like decades) [edit] 3 year old 1080 which is on an inferior node 16nm .. GTX1080 has become an iconic card for a reason I suppose.

..and yes even taking into account the 4 out of 40 CU's disabled on the 5700 the 1080's transistor count is still way down.. and it's clocking higher on an older node ..

I admit it - I'm not yet sold on RDNA .. considering the good power/price/performance ratio of something like the RX 570/580 at 14nm the new stuff doesn't seem like a step forward. Especially at 7nm.

 

[xool]

Just in

This thread, the game :

Build and repair virtual PCs on your PS4 with PC Building Simulator, out today

Run a PC repair shop and help customers or design your dream rig in Free Build mode

blog.eu.playstation.com

 

[llien]

ok so technpowerup says an average 1860-1875MHz clock for the 2070 FE vs what I assume is 1725MHz for the RX 5700 .. so that's ~+8% clock advantage for Nvdia.

No, it is also the case for AMD, 5700XT ref card hovered at around 1871Mhz.

considering the good power/price/performance ratio of something like the RX 570/580 at 14nm the new stuff doesn't seem like a step forward.

 

[xool]

No, it is also the case for AMD, 5700XT ref card hovered at around 1871Mhz.

The comparison was with the 5700 no 5700XT .. but aren't we now back we're I started - ie comparing performance of cards that have a similar effective clock - just this time it's 200MHz higher.

RX580 good, I'll stop, for your mental health..

 

[mckmas8808]

9.2 RDNA TFlops, which is equivalent to GCN's ~14TF, so we're talking about full-fat Radeon VII performance level if true.

It was said yesterday that 9.2 RDNA TFlops is equivalent to 11.5 non-RDNA TFs.

 

[sonomamashine]

PS5 Reveal Event Details and The Last of Us Part II Release Date May Have Leaked

Some big announcements may be coming from Sony fairly soon! Find out when we might learn more about the PS5, The Last of Us Part II and more.

wccftech.com

 

[SlimySnake]

It was said yesterday that 9.2 RDNA TFlops is equivalent to 11.5 non-RDNA TFs.

in terms of theoritical IPC gains, yes.

However, the 5700XT is destroying real world performance of a 12.6 tflops vega 64 and is coming very close to matching the 13.8 tflops vega 7 which is aided by a 1TB bandwidth. Notice how the higher bandwidth allows the radeon 7 to post slightly better performance on higher resolutions, but on 1080p, they are essentially the same.

 

[mckmas8808]

in terms of theoritical IPC gains, yes.

However, the 5700XT is destroying real world performance of a 12.6 tflops vega 64 and is coming very close to matching the 13.8 tflops vega 7 which is aided by a 1TB bandwidth. Notice how the higher bandwidth allows the radeon 7 to post slightly better performance on higher resolutions, but on 1080p, they are essentially the same.

So in theory RDNA it's a 1.25x gain, but in practice, it can be as much as a 1.4x gain? Maybe that's where I got the 1.4x gain from yesterday. A 1.4x gain for a 9.2TF RDNA GPU would be similar to 12.88 TF GPU of today.

 

[DeepEnigma]

So in theory RDNA it's a 1.25x gain, but in practice, it can be as much as a 1.4x gain? Maybe that's where I got the 1.4x gain from yesterday. A 1.4x gain for a 9.2TF RDNA GPU would be similar to 12.88 TF GPU of today.

Maybe that is why there are two TF camps of "leaks".

In the 9-10TF range and 12-13TF range. Early kits were probably Vega (12-13TF), and those were the targets they knew roughly Navi would match at a lower and more efficient TF count.

 

[HeisenbergFX4]

Several interesting people follow this account.

 

[Insane Metal]

2Ghz GPU? In an APU? Lol.

Edit: Gen0 = PS4 clock, Gen1 = PS4 Pro clock so Gen2 COULD theoretically be PS5. WTF.

 

[THE:MILKMAN]

2Ghz GPU? In an APU? Lol.

That's what Komachi is reporting they found. Don't you believe it then? I can't think of what chip would use exact PS4 and Pro clocks if it isn't the PS5 APU.

 

[Mass Shift]

Several interesting people follow this account.

Yeah, it's time to accept the reality. I mean if Sony wants to come out and surprise us all with even better than this at their reveal then I'm all for it. But it's probably wiser to start getting comfortable with THIS reality.

 

[DeepEnigma]

Yeah, it's time to accept the reality. I mean if Sony wants come out and surprise us all with even better than this at their reveal then I'm all for it. But it's probably wiser to start getting comfortable with THIS reality.

I'm okay with it as a new baseline over 1.31, Jaguar, slow ass ram/bandwidth and mechanical drives.

 

[sonomamashine]

Several interesting people follow this account.

 

[mckmas8808]

Yeah, it's time to accept the reality. I mean if Sony wants to come out and surprise us all with even better than this at their reveal then I'm all for it. But it's probably wiser to start getting comfortable with THIS reality.

What's wrong with this reality though?

 

[sonomamashine]

Increasing CU count over higher clocks, not buying this 2Ghz.

 

[Negotiator]

2 GHz GPU still doesn't make sense for an APU, even if it's a 7nm EUV chip.

Maybe it would make more sense if this scenario was true:

PlayStation 5 will feature full discrete GPU, not an APU

Exclusive: PS5 might feature discrete graphics, not a slower APU.

www.tweaktown.com

Unless Sony found an innovative way to cool it down (not just vapor chamber), this already sounds like PS5 JetFighter Edition to my ears.

 

[Mass Shift]

I'm okay with it as a new baseline over 1.31, Jaguar, slow ass ram/bandwidth and mechanical drives.

I'm absolutely sure it's going to be worth it. In fact I'm already excited for it.

 

[Mass Shift]

What's wrong with this reality though?

Absolutely nothing. I'm totally cool with this.

RDNA simply provides a better performance metric than GCN.

 

[DeepEnigma]

I would still be shocked if it were 2.0 in an APU.

 

[Panajev2001a]

Increasing CU count over higher clocks, not buying this 2Ghz.

If you could do it without significant sacrifices and without wasting a lot of performance due to clock mismatch with main RAM for example, raising clocks over wider designs is preferable, for the same performance differential. It would help improve branchy and compute heavy code quite a bit without forcing developers to parallelise the code as much as with a very wide and very slow design of comparable performance.

 

[CrustyBritches]

Those clock speeds are interesting indeed. I've been out of the loop, but what does this mean for Gonzalo? Is Oberon an iteration of Gonzalo since they connect to Ariel? 1.8GHz seemed high, but after Navi's reveal it's within possibility. 2GHz seems really high. Maybe it's like Polaris where the 480 hit the wall at ~1290MHz, but the following year's 580 was upwards of 1400MHz, or higher?

 

[SlimySnake]

Those clock speeds are interesting indeed. I've been out of the loop, but what does this mean for Gonzalo? Is Oberon an iteration of Gonzalo since they connect to Ariel? 1.8GHz seemed high, but after Navi's reveal it's within possibility. 2GHz seems really high. Maybe it's like Polaris where the 480 hit the wall at ~1290MHz, but the following year's 580 was upwards of 1400MHz, or higher?

oberon is also a shakespeare character like ariel, gonzalo and flute.

but the biggest indication that oberon is the ps5 gpu are the three clockspeeds of 800 mhz, 911 mhz and 2.0 ghz. the first two being the exact clockspeeds of the base ps4 and the pro.

even if gonzalo and flute werent the ps5, this thing definitely is.

 

[SmokSmog]

Yes Sony! Give me 2ghz 180w + zen2 8core 35w monster in cheap plastic box with cheap sheety cooling which sound like jet engine for only 399$, I need this!

 

[mckmas8808]

If you could do it without significant sacrifices and without wasting a lot of performance due to clock mismatch with main RAM for example, raising clocks over wider designs is preferable, for the same performance differential. It would help improve branchy and compute heavy code quite a bit without forcing developers to parallelise the code as much as with a very wide and very slow design of comparable performance.

Thanks for explaining the "pro" reason to have a 2GHz APU. I was wondering why Sony or anybody would want to go in that direction.

 

[Panajev2001a]

Yes Sony! Give me 2ghz 180w + zen2 8core 35w monster in cheap plastic box with cheap sheety cooling which sound like jet engine for only 399$, I need this!

Some people pay thousands to play with early RT and/or get a few frames per second more and some others tolerate anything more than pure silence to get a cheap and very powerful machine you will barely hear under the noise of your speakers most of the time anyways .

 

[TLZ]

I would still be shocked if it were 2.0 in an APU.

Guess they really want them high Fps counts.

Or was it higher res? I guess it's game dependent.

 

[DeepEnigma]

Guess they really want them high Fps counts.

Or was it higher res? I guess it's game dependent.

There should be less excuses for shitty framerates with a Zen2 in the box.

 

[HeisenbergFX4]

There should be less excuses for shitty framerates with a Zen2 in the box.

Until they shove Ray Tracing down our throats.

 

[SonGoku]

Several interesting people follow this account.

lol T thuway
paging statham

in terms of theoritical IPC gains, yes.

However, the 5700XT is destroying real world performance of a 12.6 tflops vega 64 and is coming very close to matching the 13.8 tflops vega 7 which is aided by a 1TB bandwidth. Notice how the higher bandwidth allows the radeon 7 to post slightly better performance on higher resolutions, but on 1080p, they are essentially the same.

Piss poor performance at 4k
Next gen needs 2080S performance minimum unless they target 1440p

 

[Polygonal_Sprite]

If they get anywhere near 2070 GPU performance along with a Ryzen CPU, 16GB's of memory and an SSD then the new consoles will be phenomenal (especially in first party developers hands).

 

[dalekjay]

The console will be with me for 5+ years so I wish they actually avoid 399 prices or at least come out with the pro version.

 

[mckmas8808]

Next gen needs 2080S performance minimum unless they target 1440p

You're forgetting about checkboard rendering. I imagine doing that next-gen will be even more refined.

 

[SonGoku]

Yeah, it's time to accept the reality. I mean if Sony wants to come out and surprise us all with even better than this at their reveal then I'm all for it. But it's probably wiser to start getting comfortable with THIS reality.

assuming this is real (pretty big if) there's no telling the performance target
40CU --> 10.2TF
44CU --> 11.2TF
48CU --> 12.28TF
52CU --> 13.3TF

Personally i think anything under 40CU is unrealistic
48CUs is the most likely config

Even lowballin with 44CU it meets my minimum expectation

 

[sonomamashine]

lol T thuway
paging statham

Piss poor performance at 4k
Next gen needs 2080S performance minimum unless they target 1440p

we gonna see a lot of 1440p CAS (Contrast Adaptive Sharpening) - 4k

If they get anywhere near 2070 GPU performance along with a Ryzen CPU, 16GB's of memory and an SSD then the new consoles will be phenomenal (especially in first party developers hands).

this is a more realistic spec, i salute you.

 

[SmokSmog]

Plot twist. There will be ps5 and ps5 pro.
PS5 36cu 399$ ps5 pro monster 72cu 550$

 

[SonGoku]

You're forgetting about checkboard rendering. I imagine doing that next-gen will be even more refined.

As did most devs even 1st party
9TF would be great for a 1440p machine

Plot twist. There will be ps5 and ps5 pro.
PS5 36cu 399$ ps5 pro monster 72cu 550$

Base PS5 1440p
Pro PS5 4K

 

[vpance]

Plot twist. There will be ps5 and ps5 pro.
PS5 36cu 399$ ps5 pro monster 72cu 550$

Probably, if they're going with a $399 launch price. But I don't think Pro will need to go beyond $449.

PS5 Pro will be a far better bargain for the time if true. Tiny and hot APU isn't such a great combo especially at launch.

 

[MadAnon]

Piss poor performance at 4k
Next gen needs 2080S performance minimum unless they target 1440p

XBX with its RX 580 equivalent does pretty well with native 4k in a lot of games while doing dynamic res scaling in others. But here you are telling that 5700xt level GPU (double the RX 580 power in combination with 4x the cpu power) in console would give piss poor performance at 4k. A gpu which beats vega 64 by the way.

Check DF video on Red Dead Redemption 2. What Rockstar Games managed to do on XBX is simply amazing.

 

[CrustyBritches]

I don't know who IsAParrott is, but there are already benchmarks with 40CU Navi at 2GHz and it's just about 5% faster than a 2070, which is fast. 36CU Navi at 2GHz would probably be slower than a 2070. 9.2TF Navi 10 is not "almost 2080 power".

 

[SonGoku]

XBX with its RX 580 equivalent does pretty well with native 4k in a lot of games while doing dynamic res scaling in others. But here you are telling that 5700xt level GPU (double the RX 580 power) in console would give piss poor performance at 4k. A gpu which beats vega 64 by the way. Check DF video on Red Dead Redemption 2. What Rockstar Games managed to do on XBX is simply amazing.

Double the X performance for the same rez target (dynamic 4k) is not enough for a next gen leap in visuals, its anemic
You can quote me on this: if there is a 9tf console, most devs will target 1440p with the occasional 1800p just like the PS4Pro

 

[Thedtrain]

Several interesting people follow this account.

This is just thuway. Remember how last time went with him?

Anyway he’s just reporting the rumor basically. Adds nothing new.

 

[SonGoku]

I don't know who IsAParrott is, but there are already benchmarks with 40CU Navi at 2GHz and it's just about 5% faster than a 2070, which is fast. 36CU Navi at 2GHz would probably be slower than a 2070. 9.2TF Navi 10 is not "almost 2080 power".

A wannabe insider from the old days (thruway) with a bit of a hard on for Sony
His "almost 2080 power" statement is just blind hype

 

[MadAnon]

Double the X performance for the same rez target (dynamic 4k) is not enough for a next gen leap in visuals, its anemic
You can quote me on this: if there is a 9tf console, most devs will target 1440p with the occasional 1800p just like the PS4Pro

Why you focus on dynamic 4k when XBX runs a lot of games at 4k native, including RDR2. :facepalm

The res target for XBX was/is clearly 4k. It's up to devs to decide if they want dynamic res upscaling for some performance gains for 60fps/other visuals or go with native.

 

[vpance]

Several interesting people follow this account.

[MDIA=twitter]1161211903588413440[/MEDIA]

So Sony knows and they don't care. Or they can't stop the leak, which looks weird on AMD's part.

 

[SonGoku]

Why you focus on dynamic 4k when XBX runs a lot of games at 4k native, including RDR2. :facepalm

Because most graphics intensive games are dynamic, regardless assuming native 4k 11TF+ even more necessary
For a 11TF machine i expect the target resolution to be the same as the X: 4K Dynamic, sometimes static.