Is using "Infinity Cache" for Navi 2X a good move?

T

TheRookie

Limp Gawd
Joined
May 21, 2019
Messages
213
AMD's previous flirting with exotic memory has been a disaster thanks to high prices and limited availability (think HBM2 in Vega and HBM in Fiji).

With Navi 2X using plain old GDDR6, AMD can get it from anywhere and inexpensively.

NVIDIA is stuck with getting relative expensive GDDR6X memory from Micron.

On the other hand, Infinity Cache has likely ballooned the size of the die and TSMC's 7nm is capacity constraint.
 
Dude.. You are trying to get people to lose their minds. 'Capacity constraints'?

Watch out for everybody's favorite show 'NerdStorm Return of Rage 2:- especially if the bots win
 
Nah .. Amd said they can deliver. Lets see if they can. I do not think the hype for these cards are going to be as big as Nvidia because "Amd" lol

This is a tightly packed forum community and our sense of environment does not reflect the outside world. This is a true echo chamber here.

People going to Microcenter etc... gonna be like oh look Radeon blah .... Hey sales dude wheres the 3080s?
 
tangoseal said:
Nah .. Amd said they can deliver. Lets see if they can. I do not think the hype for these cards are going to be as big as Nvidia because "Amd" lol

This is a tightly packed forum community and our sense of environment does not reflect the outside world. This is a true echo chamber here.

People going to Microcenter etc... gonna be like oh look Radeon blah .... Hey sales dude wheres the 3080s?
Click to expand...
100%, I think the 6000 series is a nice line up and priced appropriately. Non-techie consumers are only going to know Nvidia just released some fire and now they did it again. They probably haven't read one review and are just following the news hype.
 
I'm actually amazed the AMD sourced both 2GB chip for all their cards, while also providing 16Gbps speeds.

I'm pretty sure NVIDIA is intentionally waiting until they can offer 2GB GDDR6X chips on the 3080 before they double the 3070- memory to 16GB. But it makes no fucking sense why they wouldn't at least ship the 3070 with the same memory they have shipped with the 2080 Super for well-over a year now?

We won't know if Infinity Cache is a Good Idea™ until we see independent tests - so far, th AMD-supplied test require an Unobtaium Zen 3 chip running with some unknown memory transfer optimization.
 
Last edited:
TheRookie said:
AMD's previous flirting with exotic memory has been a disaster thanks to high prices and limited availability (think HBM2 in Vega and HBM in Fiji).

With Navi 2X using plain old GDDR6, AMD can get it from anywhere and inexpensively.

NVIDIA is stuck with getting relative expensive GDDR6X memory from Micron.

On the other hand, Infinity Cache has likely ballooned the size of the die and TSMC's 7nm is capacity constraint.
Click to expand...

Honestly, we need to wait for reviews and actual user testing before we can conclude anything. However, Infinity Cache reminds me of the crap NVIDIA tried back when it stuck the 9800GX2 with a 256bit bus. It resulted in a Quad-SLI with two 9800GX2's being slower than 3x 8800GTX's in 3-Way SLi. Whenever you shove an amount of data normally requiring a 384bit bus or whatever down a 256bit pipe, it doesn't work out so well. The problem is that we don't know anything about how Infinity Cache really works. Without more information, its tough to say. Though I am concerned given that caching usually requires that cache to be filled before it can be of use.

That being said, AMD's been pretty honest about what we can expect from them over the last few years. So, I will remain cautiously optimistic until I have a reason not to be.
 
I need to read more about this, but I think both AMD and Nvidia are just now trotting out the new architecture made possible by MS - and neither they nor the software side know yet what will happen.

I mean it's the first real pipeline change we've seen in a long time - both cards now capable of loading directly to the VRAM and bypassing the CPU? CPU using the VRAM directly? Until we can see the software implementation we won't really know how much things have changed
 
tangoseal said:
Nah .. Amd said they can deliver. Lets see if they can. I do not think the hype for these cards are going to be as big as Nvidia because "Amd" lol

This is a tightly packed forum community and our sense of environment does not reflect the outside world. This is a true echo chamber here.

People going to Microcenter etc... gonna be like oh look Radeon blah .... Hey sales dude wheres the 3080s?
Click to expand...

Sounds like the 3XXX series got a big shipment in just in time for the 3070 launch. Nvidia's way of raining on the parade I'd guess.

Honestly, I think Nvidia's echo chamber is louder judging by how many times I've heard "BUT CYBERPUNK 2077!!!!!!" in response to a lack of RTX games available.

I think everyone should just wait for reviews on 6XXX at this point. CP is delayed until December so everyone can take a look at the landscape before deciding.
 
tangoseal said:
Nah .. Amd said they can deliver. Lets see if they can. I do not think the hype for these cards are going to be as big as Nvidia because "Amd" lol

This is a tightly packed forum community and our sense of environment does not reflect the outside world. This is a true echo chamber here.

People going to Microcenter etc... gonna be like oh look Radeon blah .... Hey sales dude wheres the 3080s?
Click to expand...

You're wrong and you know you're wrong. This place used to be pro-intel when Intel was the best. Now it's heavily AMD, as it should be.

If AMD comes out with a better card, at a competitive price, people will buy it. You can see countless comments in the new card thread stating exactly that. People that have been on Nvidia for a decade considering switching if the news is correct.

You always push this "no one will use them because AMD lol"... Crazy little world you live in.
 
TheRookie said:
AMD's previous flirting with exotic memory has been a disaster thanks to high prices and limited availability (think HBM2 in Vega and HBM in Fiji).

With Navi 2X using plain old GDDR6, AMD can get it from anywhere and inexpensively.

NVIDIA is stuck with getting relative expensive GDDR6X memory from Micron.

On the other hand, Infinity Cache has likely ballooned the size of the die and TSMC's 7nm is capacity constraint.
Click to expand...

Repeat after me: "I donno how this woks, but I like to pretend I do!" :ROFLMAO:
 
STEM said:
Repeat after me: "I donno how this woks, but I like to pretend I do!" :ROFLMAO:
Click to expand...
That is true.

I have never used a wok before, but if someone asks me to use one for cooking, I'll pretend like I know how to use it.
 
TheRookie said:
That is true.

I have never used a wok before, but if someone asks me to use one for cooking, I'll pretend like I know how to use it.
Click to expand...
First thing you do is look up how to use it.
Then maybe take advice from people who DO know how to use it.
Then use it.
You can fuck up a good cast iron wok pretty damn easily if you don't know what you're doing.
Pretending to use it doesn't help anyone and is actually far more embarrassing than simply admitting that you don't know how it works.

Also, woks are great. Stir fry is awesome.
---------
I don't know how infinity cache works either, but the extra huge cache on my 3900X didn't break anything so I have hopes that this will work well.
 
TheSlySyl said:
First thing you do is look up how to use it.
Then maybe take advice from people who DO know how to use it.
Then use it.
You can fuck up a good cast iron wok pretty damn easily if you don't know what you're doing.
Pretending to use it doesn't help anyone and is actually far more embarrassing than simply admitting that you don't know how it works.

Also, woks are great. Stir fry is awesome.
---------
I don't know how infinity cache works either, but the extra huge cache on my 3900X didn't break anything so I have hopes that this will work well.
Click to expand...

If you are making Egg Fried Rice, make sure you get Uncle Roger's commentary on who does it well.
 
whateverer said:
I'm actually amazed the AMD sourced both 2GB chip for all their cards, while also providing 16Gbps speeds.

I'm pretty sure NVIDIA is intentionally waiting until they can offer 2GB GDDR6X chips on the 3080 before they double the 3070- memory to 16GB. But it makes no fucking sense why they wouldn't at least ship the 3070 with the same memory they have shipped with the 2080 Super for well-over a year now?

We won't know if Infinity Cache is a Good Idea™ until we see independent tests - so far, th AMD-supplied test require an Unobtaium Zen 3 chip running with some unknown memory transfer optimization.
Click to expand...

Huh? No! that is an added feature. 6800xt does just fine without SAM. Did you even see those benches. It was doing just fine without smart memory access against 3080.
 
At this point my next build will likely be all AMD.

While the CPU is me going bang for buck (C'mon Intel, get your crap together), GPU-wise its simply based upon likely availability. I can't even find a ghost of an Nvidia 30xx card right now even if I wanted to ... and I've tried. Just no luck. Nvidia lost out on my $1500+ and AMD will likely get my $1000 instead.
 
TheRookie said:
AMD's previous flirting with exotic memory has been a disaster thanks to high prices and limited availability (think HBM2 in Vega and HBM in Fiji).

With Navi 2X using plain old GDDR6, AMD can get it from anywhere and inexpensively.

NVIDIA is stuck with getting relative expensive GDDR6X memory from Micron.

On the other hand, Infinity Cache has likely ballooned the size of the die and TSMC's 7nm is capacity constraint.
Click to expand...
I haven't got a clue of what I am talking about, and it'll be sharing it with increased cpu production, but I seem to recall reports from the beginning of this year that amd had reserved a doubling of 7nm production capacity at tsmc for the second half of this year.
 
Some people are stuck in the past, where AMD GPUs suck and their CPUs will "never" catch up to Intel.
 
§kynet said:
Some people are stuck in the past, where AMD GPUs suck and their CPUs will "never" catch up to Intel.
Click to expand...
Maybe more cult like, where some worship a company and the cult leader. Blinders are instantly put on when facts go against the assumed beliefs while blindly attacking the truth.

As for Infinity Cache, it looks very promising, both Sony and Microsoft will be using that tech, promising AMD published results. Agree that we should also see independent reviews before purchasing.
 
Honestly I am impressed with AMD making such a turn around the past 3 years. Lisa Sue need CEO of the century award. Most CEOs that come into a sinking ship normally just sink the board and run away richer then ever before with the only life raft.
 
Last edited:
Dan_D said:
Honestly, we need to wait for reviews and actual user testing before we can conclude anything. However, Infinity Cache reminds me of the crap NVIDIA tried back when it stuck the 9800GX2 with a 256bit bus. It resulted in a Quad-SLI with two 9800GX2's being slower than 3x 8800GTX's in 3-Way SLi. Whenever you shove an amount of data normally requiring a 384bit bus or whatever down a 256bit pipe, it doesn't work out so well. The problem is that we don't know anything about how Infinity Cache really works. Without more information, its tough to say. Though I am concerned given that caching usually requires that cache to be filled before it can be of use.

That being said, AMD's been pretty honest about what we can expect from them over the last few years. So, I will remain cautiously optimistic until I have a reason not to be.
Click to expand...
According to AMD:

Measurement calculated by AMD engineering, on a Radeon RX 6000 series card with 128 MB AMD Infinity Cache and 256-bit GDDR6. Measuring 4k gaming average AMD Infinity Cache hit rates of 58% across top gaming titles, multiplied by theoretical peak bandwidth from the 16 64B AMD Infinity Fabric channels connecting the Cache to the Graphics Engine at boost frequency of up to 1.94 GHz. RX-547
Click to expand...

The cache has a hit rate of 58%.

Consequently, 42% of the time, it has to access the GDDR6 memory.

That means that it has the effective bandwidth of 256b/0.42 = 609.52b effective bandwidth

Now 609.52b/256b = 2.38X effective bandwidth

According to AMD, the effective bandwidth is 2.17X

I have no idea how AMD came up with that number.

RX6800_09.jpg
 
Last edited:
You must log in or register to reply here.
Back
Top