Support NeoGAF

[DonJuanSchlong]

wats a green meltdown?

Probably just checking the site for RTX 30XX, and realizing it's out of stock, since it's beat it's competitor for like the 8th year in a row... And people just can't get the best performance till they come back in stock.

 

[mansoor1980]

Probably just checking the site for RTX 30XX, and realizing it's out of stock, since it's beat it's competitor for like the 8th year in a row... And people just can't get the best performance till they come back in stock.

i am having a green meltdown?

 

[DonJuanSchlong]

i am having a green meltdown?

If you have to get a Red card to suffice instead of the best of the best, then yes lol

 

[Knitted Knight]

Very interesting. Makes the wait for Oct. 28 presentation the more bothersome. At the end of the day, what matters is performance, in real time applications (including games). Theoretical maxes are great for marketing but that's about it.

 

[mansoor1980]

If you have to get a Red card to suffice instead of the best of the best, then yes lol

what the hell?a good product is good regardless of the brand.........oh well im new to neogaf

 

[Knitted Knight]

what the hell?a good product is good regardless of the brand.........oh well im new to neogaf

Don't make judgements on NeoGaf's quality of posters and discussion based on DonJuanSchlong . Not a great ambassador.

Welcome to NeoGaf.

 

[DonJuanSchlong]

what the hell?a good product is good regardless of the brand.........oh well im new to neogaf

I'm not saying they aren't good. But if rumors are true, they won't have an answer to DLSS or raytracing performance of Nvidia. They will be cheaper for sure, but how long into the gen till you need more performance?

Don't make judgements on NeoGaf's quality of posters and discussion based on DonJuanSchlong . Not a great ambassador.

Welcome to NeoGaf.

Interesting for you to say, with the history of a console warrior....

 

[Knitted Knight]

Interesting for you to say, with the history of a console warrior....

Projection is a helluva drug.

 

[DonJuanSchlong]

Projection is a helluva drug.

Am I wrong? You kinda did mention me, and try and call me out on top of it... I'm just telling the obvious truth... You can look through my history and compare it to yours. I don't console war, you do... No need to mention me in a bad light, and think you are free of ridicule. Anyone can see our posts to confirm....

 

[mansoor1980]

can we plz get back to the super sony sauce

 

[DonJuanSchlong]

can we plz get back to the super sony sauce

Its hot sauce ofc

 

[Knitted Knight]

Am I wrong? You kinda did mention me, and try and call me out on top of it... I'm just telling the obvious truth... You can look through my history and compare it to yours. I don't console war, you do... No need to mention me in a bad light, and think you are free of ridicule. Anyone can see our posts to confirm....

An obvious newcommer noticed elitism with a slight that led to unproductive discussion other than warring about red vs. green. It's not subtle. Obviously first impressions are everything - he definitely shouldn't take that comment and your posting habits as representative of Gaf commentary.

You can deny that you don't part take in part time console warring for Xbox, while being a full time pretend PCMR but that won't make it any less true. So if we're strictly using terms, not only do you console war for a console but also for PC.... what would you call that? Full-spectrum Platform wars? Lmao...

 

[mansoor1980]

An obvious newcommer noticed elitism with a slight that led to unproductive discussion other than warring about red vs. green. It's not subtle. Obviously first impressions are everything - he definitely shouldn't take that comment and your posting habits as representative of Gaf commentary.

You can deny that you don't part take in part time console warring for Xbox, while being a full time pretend PCMR but that won't make it any less true. So if we're strictly using terms, not only do you console war for a console but also for PC.... what would you call that? Platform wars? Lmao...

man neogaf is LIFE

 

[DonJuanSchlong]

An obvious newcommer noticed elitism with a slight that led to unproductive discussion other than warring about red vs. green. It's not subtle. Obviously first impressions are everything - he definitely shouldn't take that comment and your posting habits as representative of Gaf commentary.

You can deny that you don't part take in part time console warring for Xbox, while being a full time pretend PCMR but that won't make it any less true. So if we're strictly using terms, not only do you console war for a console but also for PC.... what would you call that? Platform wars? Lmao...

I'm a performance kinda guy. I just go for what's best for me. I don't have a history of console warring like you, I just like open discussions. Forgive me if I'm skeptical about AMD gpu's, as they haven't been the best performers, whether it be on consoles or PC. If you would like to prove me wrong or have a discussion, feel free to. But don't try and paint a bad picture when our post history it's accessible to all of GAF. Thank you.

 

[Knitted Knight]

man neogaf is LIFE

It's. It's console war season so debates are more heated than usual but there is some gold to be found amist the chaos. Enjoy your time.

 

[mansoor1980]

It's. It's console war season so debates are more heated than usual but there is some gold to be found amist the chaos. Enjoy your time.

i know been a lurker for more than a decade.............shit can get toxic real quick

 

[vpance]

PS5 12TF confirmed

Now John’s comments make sense.

 

[CuNi]

22 % IPC would make a 23 to 24 TF card punch close to a 30 TF card, its not all about TF. Certainly interesting and cant wait. This generation will be about speed and caches from AMD at least.

Are your 22% IPC before you say it's a 23/24TF card or are those IPC gains already added when you say it's a 23/24TF card?
Because a 23TF card, no matter the tech, will always be worse than a 30TF card.
I'm not talking about gaming, I'm talking about the use case where you would care for TF.
If you care for TP, then more TF is always better than less TF.

This is your daily reminder that we should move away from using TF as a metric for game performance because TF is not a metric by which you can measure game performance at all.

 

[danielJackson]

I'm not saying they aren't good. But if rumors are true, they won't have an answer to DLSS or raytracing performance of Nvidia. They will be cheaper for sure, but how long into the gen till you need more performance?


Interesting for you to say, with the history of a console warrior....

Many PC gamers(like humans in general) are blind morons whom lost their point .

Why? Because they have lost the point: joy of gaming.

Instead of just playing, they stress from the idiotic things like is fps 80 or 90, is fps 140 or 170, does game feel laggy once in an hour?

Fastest card isnt the best card(technically it is, but not in reality), unless they are super rich.

Like spending 700€ vs 1500€ to gain 5-20% performance is just plain stupid. Getting better desk/chair/audio etc would benefit them more. Or just save the money.

All PC parts are "slow" eventually, so it is much smarter to buy 400-500€ card now, and upgrade it after 3-4 years than buy one 1000€ card now, because it will be slower than 500€ card 4 years in the future. Nothing is Future proof so buying expensive parts isnt smartest move. Those last 20% of performance cost way too much.

IMO pc gaming went nuts around 2010, people are special snowflakes whom cry if fps is under 60.

If AMD can sell cards for half the price but 80% performance, they are much better investiment.

It is just sad that people stare those fps counters + pixels instead of just gaming.

Anyone whom says "30fps is unplayable" isnt gamer, they are benchmarkers.

Almost any modern cpu with +6 cores and any gpu above 1060 is good enough for gaming, so it is collective madness to think that cards like 3080 are really that good or needed.

Like yeah, you can buy ferrari for grocery shopping trips but it is horrible overkill

Good that amd continues their business, otherwise nvidia would ask 2000€ for entry level stuff

 

[DonJuanSchlong]

Many PC gamers(like humans in general) are blind morons whom lost their point .

Why? Because they have lost the point: joy of gaming.

Instead of just playing, they stress from the idiotic things like is fps 80 or 90, is fps 140 or 170, does game feel laggy once in an hour?

Fastest card isnt the best card(technically it is, but not in reality), unless they are super rich.

Like spending 700€ vs 1500€ to gain 5-20% performance is just plain stupid. Getting better desk/chair/audio etc would benefit them more. Or just save the money.

All PC parts are "slow" eventually, so it is much smarter to buy 400-500€ card now, and upgrade it after 3-4 years than buy one 1000€ card now, because it will be slower than 500€ card 4 years in the future. Nothing is Future proof so buying expensive parts isnt smartest move. Those last 20% of performance cost way too much.

IMO pc gaming went nuts around 2010, people are special snowflakes whom cry if fps is under 60.

If AMD can sell cards for half the price but 80% performance, they are much better investiment.

It is just sad that people stare those fps counters + pixels instead of just gaming.

Anyone whom says "30fps is unplayable" isnt gamer, they are benchmarkers.

Almost any modern cpu with +6 cores and any gpu above 1060 is good enough for gaming, so it is collective madness to think that cards like 3080 are really that good or needed.

Like yeah, you can buy ferrari for grocery shopping trips but it is horrible overkill

Good that amd continues their business, otherwise nvidia would ask 2000€ for entry level stuff

PC gamers are morons.... The smell of butthurt


You don't have to get the best of the best, that's the beauty of PC gaming. You can pick whatever hardware, from cpu/gpu, monitor, ram speeds, etc. You can go balls to the wall or you just want to play next gen games.


Not everyone is an enthusiast. You can get a prebuilt or console for that.


Everything else doesn't really make sense, as it's up to user preference. You get what you pay for, you can get a ford pinto or you can get a Ferrari.


Why hate on us who want the best performance? We don't hate on you guys who get consoles. Again, you get what you pay for. You can't expect all games 4K 60fps with raytracing with a $500 budget, as we've seen before next gen even started.


This is off topic now. Feel free to pm me or @me in an appropriate thread.

 

[alabtrosMyster]

im gonna call it .....silent sony sauce

MS is like that girl who is so loud in bed but just fake it all the way.

Sony is that shy girl who does amazing stuff with her tongue .

 

[MadAnon]

Remember Rapid packed math, Asynchronous compute etc.?

 

[alabtrosMyster]

All PC parts are "slow" eventually, so it is much smarter to buy 400-500€ card now, and upgrade it after 3-4 years than buy one 1000€ card now, because it will be slower than 500€ card 4 years in the future. Nothing is Future proof so buying expensive parts isnt smartest move. Those last 20% of performance cost way too much.

While that seems reasonable, you could replace 500with whatever you think is the sweet spot is for your situation, many play on iGPUs (AMD is pretty hot on that front now... Almost making products like the 1050ti useless).

That's it though, if you really don't care too much a 1060 or a Radeon 580 still does the trick (assuming your monitor is 1080p/60hz and you realized that the ultra setting is a waste of resources).

Those high end cards are definitely worth it if you have a 4k+ monitor and you really don't like compromise (also, for many people 1000$ is not that much).

 

[longdi]

Is it possible that Cerny and AMD worked closley on this together, and it's a feature the PS5 has (but not XSX) and that's why Sony have been so shy on details, because of a possible NDA?

Nope i dont expect so. Any special Cerny sauce, will probably be something simple in tune of easier development but won't deliver diffentiating performance boost.

I been following Richard DF next gen videos, and he has been very dismissive about secret sauces. Im sure he has more industry contacts than the usual pre-launch fanboy dreams.

You can literally feel at times Richard wants to shout out on camera against SS FUD going around.

 

[alabtrosMyster]

You can literally feel at times Richard wants to shout out on camera against SS FUD going around.

 

[llien]

for every year

Let me pretend there are realistic scenarios of you buying from Huang's competitors or that I even remotely give a fuck about it, shall I.
Yeah, for every year AMD was spanking Intel, doubling+ its mcap, and disrupting NV's rollout of new cards to a clusterfuck levels.
5700 was also "so bad", something something, drivers, something something black screen.

 

[Mahavastu]

AMD's market cap is the same as Sony's, so I'm going with that haha.

Even before the big bull run, hard to buy something with 1/3rd the market cap as you even. Even looking at Microsofts biggest acquisitions, that were what, 22 billion or something, 100B is a different beast.

For a long time AMD was worth only 4-5 Billion, but back then Sony was not worth that much either...

I remember seriously considering getting a bunch of them for 2 dollars a share and just leaving it for decades...Stupid, stupid...

I sold AMD for less then $2, not that much better either

 

[DonJuanSchlong]

Let me pretend there are realistic scenarios of you buying from Huang's competitors or that I even remotely give a fuck about it, shall I.
Yeah, for every year AMD was spanking Intel, doubling+ its mcap, and disrupting NV's rollout of new cards to a clusterfuck levels.
5700 was also "so bad", something something, drivers, something something black screen.

You keep missing the point. I'd get one if they will be any good. As in if it has an answer to DLSS & RTX and can perform the same or better than some of Nvidia high end cards. No need to get triggered over people that prefer performance or that keep picking green team for that exact reason.

 

[Mahavastu]

VERY interesting video, very interesting

My mind is currently circling... Somehow it MUST be more expensive to access the L1 cache of another core instead of your own, and that this intercore communication has to be a bottleneck a lot of the time. I assume this is when the private mode (do everything as until now) kicks in.
But then he also said that until now the L2 bandwith is a bottleneck a lot of the time, and with a shared L1 cache the pressure on this should be dramaticly reduced, which should cause a speed up. Especially when the L2 cache grows (some rumors) the pressure on the L2 cache should grow too.

The other confusing aspect of MS is rhe XSX chip is dual purpose as a server, and running 4 instances it may be arranged to have L1 private as no point in sharing L1 cache among 4 different games.

The L1 cache would automaticly only be shared between the cores for your current instance, because only the cores in your virtual machine accesses the same part of the memory.
But for the L2 cache it will be a "the loudest gets more" contest

 

[geordiemp]

VERY interesting video, very interesting

My mind is currently circling... Somehow it MUST be more expensive to access the L1 cache of another core instead of your own, and that this intercore communication has to be a bottleneck a lot of the time. I assume this is when the private mode (do everything as until now) kicks in.
But then he also said that until now the L2 bandwith is a bottleneck a lot of the time, and with a shared L1 cache the pressure on this should be dramaticly reduced, which should cause a speed up. Especially when the L2 cache grows (some rumors) the pressure on the L2 cache should grow too.


The L1 cache would automaticly only be shared between the cores for your current instance, because only the cores in your virtual machine accesses the same part of the memory.
But for the L2 cache it will be a "the loudest gets more" contest

The paper lists which work loads benefit from sharing, and which are best private. The L1 infinity cache can swicth between the 2, the cost of that switching is private mode looses 4 % due to controlling that step, they sample the work loads.

But the shared mode is massive gains some above 50 %, so the average is 22 % net imporvement (depends on mix of work load). BVH was big as well I noted.

But accessing another L1 core, even through a mesh interconnect, Remember L1 on 5700 was almost 4 TB/s, even if you slow it down when cross mesh, its 10 X faster than ps5 memory bandwidth in practical terms. 1000 % sounds more lol.

I cant comment on how MS is running the XSX in a server application, I have not looked or read about it.

 

[raul3d]

Ya'll really trying to make this about PS5 and secret sauce now? How well that work out last time? What was it last time...FP32?

It was FP16 and that is now standard in every AMD and NVIDIA GPU. So I would say it turned out pretty well, don't you think?

 

[Mahavastu]

The paper lists which work loads benefit from sharing, and which are best private. The L1 infinity cache can swicth between the 2, the cost of that switching is private mode looses 4 % due to controlling that step, they sample the work loads.

But the shared mode is massive gains some above 50 %, so the average is 22 % net imporvement (depends on mix of work load). BVH was big as well I noted.

But accessing another L1 core, even through a mesh interconnect, Remember L1 on 5700 was almost 4 TB/s, even if you slow it down when cross mesh, its 10 X faster than ps5 memory bandwidth in practical terms. 1000 % sounds more lol.

I cant comment on how MS is running the XSX in a server application, I have not looked or read about it.

Thank you!

 

[thicc_girls_are_teh_best]

Good write-up geordiemp, and nice to give some info on the size of the trace links for implementing Infinity Cache. It'll help with my spec-up on mid-gen and next-next gen systems .

Personally I don't think either system has Infinity Cache or, if they do, it'll probably be a more scaled-down, rudimentary implementation of it. But if they do, I'll be pleasantly surprised.

I'm curious what type of mesh network topology they would be using for it, though. Conceptually Infinity Cache sounds analogous to certain switch fabric networks with multiple processors sharing a block of memory. Just treat the memory as cache and the processors as CUs in a GPU and bam, Infinity Cache. Oversimplifying, but that's kind of like it.

 

[Krappadizzle]

It was FP16 and that is now standard in every AMD and NVIDIA GPU. So I would say it turned out pretty well, don't you think?

The fact that it's a standard in both GPU's kinda takes away from it being a "secret sauce" weapon don't ya think?

 

[raul3d]

The fact that it's a standard in both GPU's kinda takes away from it being a "secret sauce" weapon don't ya think?

What is "secret sauce" even supposed to mean? There is nothing "secret" about having more advanced technology than your competitor, which FP16 in the PS4 Pro certainly is. It is the same as the Xbox 360 having unified shaders, the PS3 having 7 cores or the iPhone 12 having a 5nm chip. These technologies help their products perform better against the competition where it is not available. Call it whatever you like, but it is an advantage nonetheless.

 

[geordiemp]

Are your 22% IPC before you say it's a 23/24TF card or are those IPC gains already added when you say it's a 23/24TF card?
Because a 23TF card, no matter the tech, will always be worse than a 30TF card.
I'm not talking about gaming, I'm talking about the use case where you would care for TF.
If you care for TP, then more TF is always better than less TF.

This is your daily reminder that we should move away from using TF as a metric for game performance because TF is not a metric by which you can measure game performance at all.

TF is a metric of maximum calculations if every CU has data and workload every cycle. It is fixed.

But 100 % use of CUs that never happens in gaming, CU are always waiting for data, and its something like if you use your CU in a frame at 40 % that is good efficient code. What if its 30 % what if you could boost CU utilisation and efficiency to 55 % ?

So its 1.22 x performance of RDNA2 and if you want to think of it as a 23 TF card then thats up to you but 23 does not define its gaming power. It will be compared to a 30 TF amperer 3080 and efficiency of Ampere running that game and engine.

The metric is FPS and resolution a game runs, adding up rasterisation, fill rate, shading and all the effects such as shadows and lighting. How quickly the AMD card can feed the CU will be as important as the number of them or their paper specs in an ideal world.

Out of interest, the amd card will have a 256 bus and half the memory bandwidth of a 3080. We will learn things.

Is it not entertaining ?

And yes, the AMD card will likely be poorer at mining, as sharing cache data for BVH would not help.

 

[CuNi]

TF is a metric of maximum calculations if every CU has data and workload every cycle. It is fixed.

But 100 % use of CUs that never happens in gaming, CU are always waiting for data, and its something like if you use your CU in a frame at 40 % that is good efficient code. What if its 30 % what if you could boost CU utilisation and efficiency to 55 % ?

So its 1.22 x performance of RDNA2 and if you want to think of it as a 23 TF card then thats up to you but 23 does not define its gaming power. It will be compared to a 30 TF amperer 3080 and efficiency of Ampere running that game and engine.

The metric is FPS and resolution a game runs, adding up rasterisation, fill rate, shading and all the effects such as shadows and lighting. How quickly the AMD card can feed the CU will be as important as the number of them or their paper specs in an ideal world.

Out of interest, the amd card will have a 256 bus and half the memory bandwidth of a 3080. We will learn things.

Is it not entertaining ?

And yes, the AMD card will likely be poorer at mining, as sharing cache data for BVH would not help.

You perfectly explained why TF are not a metric that can be used in any way to compare gaming performance, bravo.
So can we finally stop using this number at all?

 

[geordiemp]

You perfectly explained why TF are not a metric that can be used in any way to compare gaming performance, bravo.
So can we finally stop using this number at all?

Can you kindly stop coming into threads to console or card war, I am talking about infinity cache and its meaning to RDNA2 performance and how it is used in architecture.

You clearly have got nothing intelligent to add to this, or likely any other conversation and your input was just poor..

I ignore trolls and idiots now. Goodbye.

 

[CuNi]

Can you kindly stop coming into threads to console or card war, I am talking about infinity cache and its meaning to RDNA2 performance and how it is used in architecture.

You clearly have got nothing intelligent to add to this, or likely any other conversation and your input was just poor..

I ignore trolls and idiots now. Goodbye.

I'm not "card warring" at all.
When I see someone say things like "RDNA2 TF are like 2x Ampere TF!!" I have the right to correct this person.
That's like saying "1 German Meter is like 2 French Meters!!" which is equally bullshit.
TF has a meaning and is a measurement in performance in Floating Point Operations and is made to be comparable. A 20TF card will deliver a better floating point performance than a 15TF card. There is no "but". 20TF is more than 15TF. Period.

The issue is that people still cling to this metric and compere completely different architectures to each other that hold no meaning.
Gaming is a completely different use case than floating point operations alone and this is exactly the reason why a theoretical 20TF card could beat a theoretical 30TF card, because games don't exclusively need TF but much much more.
You ignoring and dismissing this valid point just proves that you are not interested in a debate and resort to calling someone a "troll" and probably ignoring him than to admit that you were wrong or to admit that cards cannot be compared by TF alone.

If you are not interested in any kind of meaningful and intellectual conversation, you should stay away from debates about tech.