i know this is just a cutscene but this is the most insane thing ive seen this gen.
it does matter which is precisely why we are saying 10 Navi Tflops are equivalent to 14 polaris tflops.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Next-gen PS5 and next Xbox speculation launch thread |OT5| - It's in RDNA
- Thread starter Mecha Meister
- Start date
- Status
it does matter which is precisely why we are saying 10 Navi Tflops are equivalent to 14 polaris tflops.
Yes, but comparing FLOP count isn't apples to apples.
The PS3 for example had a 230 GFLOP GPU.
The PS4 had a 1.8 TFLOP GPU.
Each architecture has its own "equivalent" to another. So numerically that was an 8x fold increase. So, would a 14.2 TFLOP Polaris GPU still be an 8x over a GCN card?
Last edited:
it would seem they did do a redesign.
there were rumors of a 2019 launch back in 2018. MS was certain they were going with an 8 tflops GPU. Panello was at MS back then and was sure the PS5 was 8 tflops.
Then something changed. Console seemed to have been delayed to 2020. Now Sony somehow has the more powerful console. 7nm EUV should give a 10-15% reduction in die size allowing them to push clockspeeds higher. its a likely explanation for this sudden shift in tflops wars.
Or MS was always using 7nm EUV as well since they were always targetting holiday 2020 and thats why Phil was so sure of xbox being more powerful. Then Sony saw that and said might as well go with 7nm EUV as well.
Lastly, RDNA2 cards are on 7nm+. We know RDNA2 cards will have RT. We know consoles have RT. Makes sense consoles will be on 7nm+ just like RDNA2 cards which were always designed with EUV/7nm+ in mind.
Devil's advocate:
Chiplet Zen 2 on 7nm with GPU + IO on 7nm EUV.
FWIW, I think consoles will be monolithic 7nm DUV 2nd gen.
If we use the 1.8TF of the base ps4 and if 14 TF is indeed the range, that whould be around 7x.
230 gflop has cell, rsx (ps3 gpu) has 192gflops with very old architecture (before unified shader), so the jump to 1.84tf modern gpu was huge.
Ok, thanks. "TheThreadsThatBindUs" was confusing me a little.
Polaris is GCN.
I dont know how efficient polaris was compared to the base PS4 GCN GPU which is why i gave you the comparison to the Pro and X1X which were both based on Polaris GPU.
To reiterate, Navi tflops offer 40% better performance compared to Polaris. Since Pro and X1X are polaris, we can assume a 10 tflops Navi GPU in a next gen console would offer 14 Polaris tflops performance. Thats 2.5x x1x and 3.5x over the Pro.
it would seem they did do a redesign.
there were rumors of a 2019 launch back in 2018. MS was certain they were going with an 8 tflops GPU. Panello was at MS back then and was sure the PS5 was 8 tflops.
Then something changed. Console seemed to have been delayed to 2020. Now Sony somehow has the more powerful console. 7nm EUV should give a 10-15% reduction in die size allowing them to push clockspeeds higher. its a likely explanation for this sudden shift in tflops wars.
Or MS was always using 7nm EUV as well since they were always targetting holiday 2020 and thats why Phil was so sure of xbox being more powerful. Then Sony saw that and said might as well go with 7nm EUV as well.
Lastly, RDNA2 cards are on 7nm+. We know RDNA2 cards will have RT. We know consoles have RT. Makes sense consoles will be on 7nm+ just like RDNA2 cards which were always designed with EUV/7nm+ in mind.
Unless/until one of ZhugeEX, Matt or Jason say otherwise I can't believe there has been any redesign.
Sorry, I was being an idiot. I should've done my research.
That sounds good still. I was worried that with a slowdown in GPU progress we would be in trouble, but it looks to be pretty steady.
Damn. I remember hoping for 12-14Tf PS5 on the old forum. Crazy to think we might actually be getting that.
Tangential: LG is doubling their large panel OLED capacity. Gimme that sub $2K 65" panel MSRP.
The PS4 and XB1 GPU's were GCN based. The next gen consoles are part of a new architecture known as RDNA. So from what I understand the RDNA arch has a 1.25-1.4x performance multiplier over GCN. An example would be a 10 RDNA TFLOP card having equal performance to a 12.5-14 TFLOP GCN card. Not that it matters much because I've recently learnt that TFLOPS is a pretty useless way of comparing performance.
Last edited:
Nope, it could mean that sony initially set out to launch in 2019 and looking at the system they would be building realized that it would end up costing them an estimated $600+ (like maybe $620 using their most optimistic estimates if launching the system in 2019) but if they waited till 2020 the prices on a number of the components they are using would have dropped (fab maturity, increased supply...etc) and that would make it possible for them to make the same system for like $530.
I believe that there are multiple design lines for possible systems during the design phase. And as time goes on resources and time are focused into the one that makes the most sense or is decided upon. So some time 2018 their 2019 plans were shelved and then they moved onto the 2020 plan. Not that they couldn't release it in 209, but that if they did it would cost them too much.
More like 30% (max) if we compare with same number of CUs. We'll know more when they release RDNA GPU with more CUs.
I was talking about Microsoft only, not Sony, I do think Sony is using a one big 1TB SSD. Microsoft can't use that patent so anything regarding the patent is irrelevant.The price of a HDD is 35 $ incompressible. The only difference is storage size growing during the time. One day it can be 3 or 4 TB for the same price.
All mechanical and eltecro mechanical part + optical drive and HDDD of PS4 were 100 dollars of PS4 cost at launch and because of mechanical part there is an incompressible price as much as the SOC... I am sure using a custom SSD 256 GB like in patent + a 2TB HDD against a 1TB SSD is probably a little bit more expensive at launch and the cost difference grows in time...
According to the table you've brought in this post, the One had in 2013 a 500GB HDD with 8MB cache for 37$. In 2016 the One S had a 1TB HDD with 32MB cache for 32$ and the 2TB HDD with 8MB was 55$. Today, a 256GB NVMe SSD cost ~1/4 the price of a 1TB NVMe SSD. The total BOM at launch should be more or less the same for a 256GB SSD cache drive + 2TB HDD VS 1TB SSD.
Cost after launch is irrelevant. Consoles go through revisions all the time, they switch HDD models all the time. If your system knows who to work with a cache SSD + HDD, it also knows how to work with a pure SSD storage solution. If after two years having a single 2TB SSD is cheaper than using 256GB SSD + 2TB HDD, they will just use a pure SSD solution at that time-frame. There is no need for future proofing your machine storage wise as long as you upgrade your storage and never downgrade. I'm pretty sure that a pure SSD solution is an upgrade, so no problem there.
Regarding repair and warranty costs, the HDD doesn't have to be included in the warranty. A lot of appliances cover only certain parts of the warranty. If you have an easily replaceable HDD, anyone can go to the store and buy a new one for 40$ or just use an external HDD so no real problem there. If you want to worry about anything, worry about the SSD soldered to the motherboard. If anything happens to it, they have to replace the whole thing.
If I remember correctly, and chris 1515 probably read the patent much more thoroughly than me, in the AMD RT patent they want to take the TMUs and change them into something called Texture Processor which will also have a BVH intersection check unit. There are 4 TMUs inside each CU so if the PS5 will have 40CUs, it will have 160 Texture Processors (which also does what Turing is doing in order to accelerat RT) with each additional CU adding 4 more Texture Processors.
This is an RDNA CU, the lower part is grayed out because that's the second CU that share its' resources. The yellow part on the right are the four TMUs which should be replaced with something larger called TP (Texture Processor) that does both the TMU's job and RT BVH intersection checks:
Again, if I got something wrong please correct me because I read this patent weeks ago without diving too deep.
Looky here what i found on the other forum.
a 5700 overclocked to 1.8 ghz (36 CU at 1.8 = 8.29 tflops)
and a 6 core/12 thread Ryzen 5 at 3.6 ghz. should be slightly less powerful than a 3.2 ghz zen 2.
Regardless, graphics score for an 8 tflops 36 CU GPU is 24.3k which is roughly on par with rtx 2070's graphics score of 24.1k paired with a ryzen 5.
Albert Penello explained this in one of the previous threads. A lot of what you say here isn't how it works.
Albert Penello said:
It's a great score but I just wish we would hit that 10 TF Navi .hehe give me that double digits Navi TFLooky here what i found on the other forum.
a 5700 overclocked to 1.8 ghz (36 CU at 1.8 = 8.29 tflops)
and a 6 core/12 thread Ryzen 5 at 3.6 ghz. should be slightly less powerful than a 3.2 ghz zen 2.
Regardless, graphics score for an 8 tflops 36 CU GPU is 24.3k which is roughly on par with rtx 2070's graphics score of 24.1k paired with a ryzen 5.
boost of this ryzen is 4.2ghz but not bad representation of 3.2ghz 8c zen2Looky here what i found on the other forum.
a 5700 overclocked to 1.8 ghz (36 CU at 1.8 = 8.29 tflops)
and a 6 core/12 thread Ryzen 5 at 3.6 ghz. should be slightly less powerful than a 3.2 ghz zen 2.
Regardless, graphics score for an 8 tflops 36 CU GPU is 24.3k which is roughly on par with rtx 2070's graphics score of 24.1k paired with a ryzen 5.
So if 8 RDNA TFLOPS = 2070 performanceLooky here what i found on the other forum.
a 5700 overclocked to 1.8 ghz (36 CU at 1.8 = 8.29 tflops)
and a 6 core/12 thread Ryzen 5 at 3.6 ghz. should be slightly less powerful than a 3.2 ghz zen 2.
Regardless, graphics score for an 8 tflops 36 CU GPU is 24.3k which is roughly on par with rtx 2070's graphics score of 24.1k paired with a ryzen 5.
10 RDNA TFLOPS = 2080?
There is no way to know what Sony plan was since we have no idea when the delay happen ( not even certain if you can call it a delay )
We hear about it mid 2018 for all we know they decide not to in 2017 .
It really make no sense to think about what if .
We hear about it mid 2018 for all we know they decide not to in 2017 .
It really make no sense to think about what if .
Yes. See below.
So thats a 40 CU 1.93 ghz 9.88 tflops GPU scoring a 26.3k score.
My RTX 2080 and i7-8700 3.2 ghz gave me a 26.5k score. (core clock was 1.5 ghz but i saw it stay in the 1.9 ghz range pretty much the entire going up to 2.05 ghz at times.
I will be ok with a 10 tflops RDNA gpu in this case.
Wow, that would be amazing! Your prediction for 12.9 TFLOPS would probably be like a 2080 TI.
There is absolutely 0 chance either company releases a console on par with a 2080ti. Zero, nada, zilch.
I can't really see any difference between having some internal SSD+HDD solution compared to some internal SSD + external HDD one.
Both would require some kind of tiering technology, that would move game data from the slower disk to the faster one, and unused blocks from the faster disk to the slower one.
Such technology has heavily been used in NAS/SAN storage arrays for years, so Sony and MS could perfectly implement it in their consoles.
And just like you can start playing a game although it's not fully downloaded on PS4, you don't need the whole game to be moved to the SSD at once. Rest of the tiering operation can happen in the background.
If they both release with a 2080 equivalent GPU (my prediction) than maybe with optimization we could get close.
While i dont think we will get close to that on a 500 dlrs machine ( or a value near that), i hope you end up being right just because it would be awesome for us as consumers.
But honestly, thats to high of a performance to achieve. Maybe with the redesigns in 2023.
How much more powerful is the 2080 TI compared to the 2080?
I think the best we could hope for is somewhere in between GTX 1080 and 1080Ti performance.
With raw specs like the 1080 ( 8.8 TF ) and equivalent performance of 1080Ti thanks to RDNA.
I don't want to compare with RTX/Turing GPUs because I don't believe AMD will have comparable raytracing HW, even for Scarlett.
With raw specs like the 1080 ( 8.8 TF ) and equivalent performance of 1080Ti thanks to RDNA.
I don't want to compare with RTX/Turing GPUs because I don't believe AMD will have comparable raytracing HW, even for Scarlett.
I am not contradicting what he said.
I am basically saying there are "multiple step 1s". Based on multiple potential configurations early on in the design process. Kinda like how MShad an anaconda and Lockhart and then decided to shelve one and focus on the other.
I am saying that rather than Sony/MS saying in 2015 "ok, we are NOT going to make anything more than a $400 box" it was more like "we aren't going to retail for any higher than $499." And there are just some things that back in 205/16 they just would not have known to be able to make any hardened decision on. They wouldn't have known if 7nm would, by all means, be ready in 2019/2020. They Wouldnt know how aggressively SSD prices or RAM prices may drop or go up.
I think this is just common sense, to assume so many things you have no control over will ALL go exactly according to plan would be flat out stupid hence why there would no doubt be multiple design directions.
When people mention RT capabilities I always think of this post: https://www.reddit.com/r/PS5/comments/bjne11/for_the_people_who_dont_believe_the_next/
How do we know Sony or MS couldn't implement a chip like this?
Im not sure on exact numbers, but i believe its in the realm of 30%+ performance wise. I might be wrong though, havent checked benchmarks in a while.
Thanks! Btw I was referring to 2080 performance without ray tracing included. I don't know if the RTX cards would technically be more powerful without RT enabled.
I think we have our next-gen console there. Especially if Gonzalo is the PS5.
A downclocked 8C/16t CPU would probably perform just a tad better or on par with a 6C/12T [email protected]. And the Gonzalo info claimed a combined score of over 20k.
Imagine this level of destruction in an open world!
The last few pages have been rough. 16 TF, 2080 Ti performance, 7nm EUV, 220w and up power draw. Might be worth having one of those Bingo cards at this point because speculation in here has officially gone off the rails.
This video could almost be used as a next gen tech showcase:
Although it is exclusively UE4.
Although it is exclusively UE4.
- Status