Haha I didn't even bother to read the label and it's in small text too. But nar the dev never mentioned 15% in the video.
Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]
- Thread starter BRiT
- Start date
- Status
AbsoluteBeginner
Regular
This is how hearsay starts. People get info on 20K benchmark on synthetic bench. Then they compare this score to best GPUs from ANOTHER vendor (not even AMD) and then say "So since PS5 is on 1080TI level we can expect 10+ TF and better performance then Scarlett".
Few pointers :
1. Firestrike bench can be VERY misleading
2. It is overall system score, not GPU score
3. We dont know how well 8core Zen2 scores, could score considerably higher then 2700 people compared it with
4. Vega56 with Intel i7 8700k scores up to 21600k, 2k more then with 2700
5. AMD cards generally do better compared to Nvidia cards, especially when compared vs scores in game benchmarks
6. 1080 scores better then 2070, yet 2070 is 20% faster card in actual games
7. We dont know how Navi compares to GCN in synthetic benchmarks
What to make of it? We dont know. We WILL know on 7/7 when NDA is dropped and we get Navi scores and Zen2 scores. When that happens, we can run benchmark with 3.2GHZ core clocks on Zen2 matched with Navi. Until then, it can perform anywhere from Vega56 to 1080Ti (not really).
In any case, my opinion is following :
1.Since NaviXT performs ~5% better then 2070 in games (allegedly), and since AMD GENERALLY scores higher then Nvidia, we can assume Navi XT score will be higher then Nvidia 2070 in FS benchmark
2. Navi is 9.75TF card in BEST case scenario (when boosted to 1905MHZ). AMD interestingly put "game clock" of 1755MHZ in slide, which I think is where this card will spend most of its time. On 1755MHZ, Navi XT is ~9TF card.
3. Navi XT with 40CUs is 250mm² card with 225W TDP (I assume under game clocks this will be ~180W). Both still high for console.
Chip is big, if paired with Zen2, some RT cores and 256 bit bus it should be max for console (considering node process is getting ever more expensive).
My conclusion, Navi XT is not 10TF card. Its much more likely to be 9TF card under load. It scores better then 2070 in games, so I assume it will score even better in benches.
AMD is likely concetrating more on rendering performance with new Navi then pure compute, therefore lower TF is still paired with considerably better performance in games.
Having said that, with everything laid out so far (die size, TDP, 7nm costs, Gonzalo) dont expect consoles to have more performance then Navi XT. I assume 36 active CUs with higher clocks and good cooling not to go overboard with die sizes. So all in all around 8TF.
One more intersting coincidence. Zen2 + Navi 10 - 40CUs + ~8% increase in CU size duo to RT and 256 bit bus should be ~316mm² on 7nm node.
This die size matches by mm to leaked PCB dev kit motherboard info that was posted on 20th of May. Since it was posted before Navi die size and CU design was known, he couldnt have known this information.
So that leak was either legit, which I believe it is, or extremely good fake that also got incredibly lucky with predicting how big (in mm accuracy) new RDNA chip with Zen2 would be on 7nm process,before we even had any info on it.
Few pointers :
1. Firestrike bench can be VERY misleading
2. It is overall system score, not GPU score
3. We dont know how well 8core Zen2 scores, could score considerably higher then 2700 people compared it with
4. Vega56 with Intel i7 8700k scores up to 21600k, 2k more then with 2700
5. AMD cards generally do better compared to Nvidia cards, especially when compared vs scores in game benchmarks
6. 1080 scores better then 2070, yet 2070 is 20% faster card in actual games
7. We dont know how Navi compares to GCN in synthetic benchmarks
What to make of it? We dont know. We WILL know on 7/7 when NDA is dropped and we get Navi scores and Zen2 scores. When that happens, we can run benchmark with 3.2GHZ core clocks on Zen2 matched with Navi. Until then, it can perform anywhere from Vega56 to 1080Ti (not really).
In any case, my opinion is following :
1.Since NaviXT performs ~5% better then 2070 in games (allegedly), and since AMD GENERALLY scores higher then Nvidia, we can assume Navi XT score will be higher then Nvidia 2070 in FS benchmark
2. Navi is 9.75TF card in BEST case scenario (when boosted to 1905MHZ). AMD interestingly put "game clock" of 1755MHZ in slide, which I think is where this card will spend most of its time. On 1755MHZ, Navi XT is ~9TF card.
3. Navi XT with 40CUs is 250mm² card with 225W TDP (I assume under game clocks this will be ~180W). Both still high for console.
Chip is big, if paired with Zen2, some RT cores and 256 bit bus it should be max for console (considering node process is getting ever more expensive).
My conclusion, Navi XT is not 10TF card. Its much more likely to be 9TF card under load. It scores better then 2070 in games, so I assume it will score even better in benches.
AMD is likely concetrating more on rendering performance with new Navi then pure compute, therefore lower TF is still paired with considerably better performance in games.
Having said that, with everything laid out so far (die size, TDP, 7nm costs, Gonzalo) dont expect consoles to have more performance then Navi XT. I assume 36 active CUs with higher clocks and good cooling not to go overboard with die sizes. So all in all around 8TF.
One more intersting coincidence. Zen2 + Navi 10 - 40CUs + ~8% increase in CU size duo to RT and 256 bit bus should be ~316mm² on 7nm node.
This die size matches by mm to leaked PCB dev kit motherboard info that was posted on 20th of May. Since it was posted before Navi die size and CU design was known, he couldnt have known this information.
So that leak was either legit, which I believe it is, or extremely good fake that also got incredibly lucky with predicting how big (in mm accuracy) new RDNA chip with Zen2 would be on 7nm process,before we even had any info on it.
AbsoluteBeginner
Regular
Gonzalo leaks are rumor. Apisak benchmark score is a rumor. Both worthy of disscusion, and I actually think Gonzalo is PS5.
Posting crapgamer/wordpress/pastebin talking about that same rumor ,on B3D, is IMO not worthy of discussion.
I think we can already safely predict what PS5/Scarlett will be packing. On 7/7 we will be able to do so with 90% accuracy.
Weird CrapgamerReviews whom favors the Xbox brand would actually put that "15% more powerful than Scarlett," in his video-podcast title. I guess even console warriors from different camps want those "PS5 MoPowitful111!!!" clicks.
Read the thread title again please
, we also have a thread that's more based on technical merit and trust worthy sources too you know.https://forum.beyond3d.com/threads/...ion-with-a-technical-spin-post-e3-2019.61245/
But you're on the right track tho. I also believe Gonzalo is too real not to be in either one of those consoles.
Perhaps not here, but people do, it seems. I think it's an issue with those attached to a brand, and they know they will buy that one over the other, that they want to know their choice is also the best, meaning most powerful (for those caring at this instance). Hearing that the machine you are going to buy is inferior is disappointing. Of course, if you don't care about numbers and only experience, like Nintendo gamers, it's not, but for XB and PS fans, especially considering they'll be playing the same cross-platform games, they want to be assured they are getting the better deal when being loyal.
That'll only apply to those following these rumours. The wider public who don't give a rat's arse live on on blissful ignorance, will buy the platform they like best, and not even know it has 15% inferior image quality.
This is the 'dumb shit' thread. If you don't like the idea of discussing dumb shit, please opt out.
I'm not sure if Firestrike is misleading. It's an empirical test.
Readers are being misled if they don't understand what its testing however.
There are reasons why a lot of amateur data science folks continue to purchase up old 1070s for machine learning as opposed to buying the latest RTX cards. Performance per dollar is significantly better to begin with; but when we look at things like the different types of neural networks we run, some are more compute oriented, others are more bandwidth oriented, some are small enough to fit in large caches, and when somethings like that happens a 1070 will outperform a 2070 for instance (RNN/NLP) (IIRC 1070 > 2070 for some cache sizes)
credit: tim dettmers https://timdettmers.com/2019/04/03/which-gpu-for-deep-learning/
*yellow bar is beast on the pascals*
Understanding the purpose of the benchmarks, and understanding how that may translate to game performance is pretty critical.
There is a lot of talk about TF. But I'd say the silent beast has been cache sizes and bandwidth. And It may not seem a big deal, but that 40% compute difference between X1X and 4Pro is translating to something much larger than 40%, and this is likely because of the increased cache sizes and bandwidth from memory.
Recent news is that Cyberpunk 2077 will be 4K on X1X.
https://www.gamepressure.com/newsroom/cyberpunk-2077-ps4-size-and-xone-x-resolution/zcd71
For a console that was largely believed to not be able to do any titles at native 4K. Many people are mistaken at how far developers have been able to push this machine. TF is not the only metric that will matter here.
Can someone please explain to me, why Sony or Microsoft would allow AMD or TSMC to test their respective undisclosed chips in an off-the-shelf manner, rather than a more secure manner? I mean, wouldn't custom silicon designs testing or even ES/QS chip testing from these type of partnerships be more high-level (secure) towards stopping performance metrics from being inadvertently (or purposely) leaked?
Seriously, if I was Sony or Microsoft I wouldn't have my undisclosed chips being tested in such an open manner where general AMD/TSMC company employees can willy-nilly access this stuff or put two-and-two together.
Seriously, if I was Sony or Microsoft I wouldn't have my undisclosed chips being tested in such an open manner where general AMD/TSMC company employees can willy-nilly access this stuff or put two-and-two together.
Last edited:
We don't know who is doing the testing, that's sort of the issue we have. It could be Sony themselves.
I mean, have they found the Microsoft chips yet? If the answer is no, then we can't necessarily put blame on AMD or TSMC.
AbsoluteBeginner
Regular
Eh, I just thought this was circular rumor getting new spin.This is the 'dumb shit' thread. If you don't like the idea of discussing dumb shit, please opt out.
Apisak > B3D > Resetera > YT Vlog/Pastebin > add spin on it > B3D
But wasn't the first leak of "Gonzalo" from someone supposedly having access to TSMC related data? And if Sony is testing PS5 ES chips in such an open manner, and showing up in comparative benchmark Web results - then they're complete idiots.
As for Scarlett related chips, wasn't there rumors surrounding the other undisclosed chips (forgot the names... art or argus something) from Komachi or APISAK
Everyone knows first version of Navi draws too much power and may not suitable for consoles. Why do some people insist consoles use this version of Navi instead of using other version with better power efficiency?Having said that, with everything laid out so far (die size, TDP, 7nm costs, Gonzalo) dont expect consoles to have more performance then Navi XT. I assume 36 active CUs with higher clocks and good cooling not to go overboard with die sizes. So all in all around 8TF.
We can wait for more info of other versions of Navi GPU (high-end GPU or even 7nm+ RDNA2 based Navi GPU) to estimate console performance precisely.
Forza 7 already did native 4k at launch of X1X.
AbsoluteBeginner
Regular
Because thats what we have? Because people thought MS and Sony will use special, next gen GPUs last two times, but they went with what they got, because AMD would deliver better if they could.
In any case, Navi XT Time Spy.
Big advantage over Vega56 and Vega64 for normal Navi.
You mean:
Ariel == PS5
Arden == Anaconda
Argalus == Lockhart
I'm not sure I followed the information line that linked these back to their respective console vendors, but the only one that had a link were the codes from Gonzalo to Ariel.
Not sure how Ariel is connected to PS5 though.
@anexanhume or @Proelite may recall how this linkage began.
Might be a good time for a Baseless Summary for everyone catching up.
It's likely that "Gonzalo" is rumored "2019 PS5".
Yes Gonzalo may be used in very early devkits to simulate the "2019 PS5". It may end up with 8TF for a console. This is why MS may believe xbox will defeat "2019 PS5".
AbsoluteBeginner
Regular
Maybe. Or it may be that Navi was late, as was pretty much confirmed by AMD and there was no way console will be ready for 2019.
That and fact that Apple literally has monopoly over TSMC 7nm manufacturing capacity till next year.
Navi GPUs have a series of products and current 5700XT may be the less efficient GPU of the whole Navi family.
So using 5700XT you may only predict the baseline performance of next-gen consoles, not final.
Let's avoid these types of assumptions. We have no clue what Microsoft plans were. Other than some internet noise based on hearsay and inept rumors.
- Status
Similar threads
- Locked
