AMD Big Navi would get Infinity Cache

#5839117

Ryu5uzaku

2020-10-06 11:47

Interesting

#5839127

Mpampis

2020-10-06 11:58

Tried to find how much more expensive faster GDDR6X memory is than the one AMD is expected to use. Didn't have luck with that. I wonder if it's just a price reduction technology, or a game changer.

#5839134

Kaarme

2020-10-06 12:10

Cplifj:

Fancy new name for what essentially is the HBCC tech from Vega ? I bet it is. Small details may differ but that is irrelevant.

Isn't HBCC just to help with too little VRAM? The problem with the 6000 series, if the rumours (or leaks, like some like to call them) are true, the GPUs might be memory bandwidth starved, not memory amount starved, like some say the 3080, for example, is. But sure, the lesser models of the AMD 6000 ought to have less VRAM, so it might be relevant if it works. However, if this Infinity Cache is to help with the bandwidth issue and is real cache, I wonder where it's physically located.

#5839136

cucaulay malkin

2020-10-06 12:16

they have to compete with 16gbps 256-bit against 19gbps 320-bit I bet this will be hit&miss,can't make up for 1.53x bandwidth advantage for competing with a 3070 this may show,at least to some extent at 4K.unless this is hbcc rebranded.

Mpampis:

Tried to find how much more expensive faster GDDR6X memory is than the one AMD is expected to use. Didn't have luck with that. I wonder if it's just a price reduction technology, or a game changer.

gddr6x is nvidia+micron collaboration,amd will never get this.

#5839145

cryohellinc

2020-10-06 12:36

New challenge - Install Doom onto GPU Cache.

#5839147

Silva

2020-10-06 12:37

cucaulay malkin:

gddr6x is nvidia+micron collaboration,amd will never get this.

Didn't know about the collab part, but it must be expensive too (and hot). AMD has to work with what its available. 256-bit is to make it more affordable and competitive, I hope it's enough to not starve the card.

#5839149

Undying

2020-10-06 12:38

So the rumors are true. We'll see how much cache will actually be there.

#5839153

JiveTurkey

2020-10-06 12:47

I assumed this Infinity cache is AMD answer to RTX IO.

#5839154

Fediuld

2020-10-06 12:53

JiveTurkey:

I assumed this Infinity cache is AMD answer to RTX IO.

Nope.

Cplifj:

Fancy new name for what essentially is the HBCC tech from Vega ? I bet it is. Small details may differ but that is irrelevant.

How much you bet to send you an invoice through PP to pay me.

#5839156

rl66

2020-10-06 12:57

cucaulay malkin:

they have to compete with 16gbps 256-bit against 19gbps 320-bit I bet this will be hit&miss,can't make up for 1.53x bandwidth advantage

3 exemples that is not an exact logic: AMD: Vega VII vs traditional other "high end" AMD... NVidia: the GTX 960/950 with low gbps and bit, compensate by compression that perform better better than what expected... General computer more AMD but it work with Intel too: it's better to have low CL ram than faster frequency in most of the case A GPU depend more on the whole pack (drivers + GPU + board + technologic used) than focus on spec. It's a bit like race car, i have been in front of many horses (running or prancing) despite having lot less HP and top speed... (but i was way lot lighter, and with smaller tire 14" or 15") So, as the RTX30*0 before, we have to see it in real world before to say if AMD have made good choise or not.

#5839166

Kaarme

2020-10-06 13:17

cucaulay malkin:

gddr6x is nvidia+micron collaboration,amd will never get this.

HBM was developed by AMD and Hynix, yet Nvidia also uses it in some professional products. Wouldn't GDDR6X be the same?

#5839175

Fediuld

2020-10-06 13:34

Kosmoz:

Infinity Cache is like an L3 mega cache (128mb) which is for making a 256bit bus act like a much bigger bus, thus giving enough bandwidth to compete with RTX 3000. This solution is also cheaper. AMD said they consider GDDR6X too power hungry, expensive and in low volume at this time. As you can see that's exactly the case (not only that) with the RTX 3090/3090 issues and underwhelming performance. Ampere is an AI and professional architecture primarily, turned gamer architecture for convenience. Navi 2x is a pure gaming architecture, that has nothing from the old one, as in not a hybrid like Navi 1x was. This time AMD is not joking around. They will be at least as good if not better than nvidia class per class, while being less power hungry, with more Vram and slightly better prices. Performance wise I have no worries from AMD, what I need to see is software side: drivers and DLSS and ray tracing performance, and there is the only place I can see nvidia still being ahead, but not on pure perf/watt/$.

Yep. AMD has CDNA now for the compute cards as it split out compute and gaming RDNA. Arcturus (CDNA) is a big evolution over Vega (GCN), while RDNA2 is new architecture, remotely related to RDNA1 in name. And next year RDNA3 is even more drastic ground up designed architecture.

#5839190

cucaulay malkin

2020-10-06 14:12

rl66:

3 exemples that is not an exact logic: AMD: Vega VII vs traditional other "high end" AMD... NVidia: the GTX 960/950 with low gbps and bit, compensate by compression that perform better better than what expected... General computer more AMD but it work with Intel too: it's better to have low CL ram than faster frequency in most of the case A GPU depend more on the whole pack (drivers + GPU + board + technologic used) than focus on spec. It's a bit like race car, i have been in front of many horses (running or prancing) despite having lot less HP and top speed... (but i was way lot lighter, and with smaller tire 14" or 15") So, as the RTX30*0 before, we have to see it in real world before to say if AMD have made good choise or not.

yes,except your expamples don't reflect the scope of this here situation,amd absolutley need that bandwidth,that's why they're introducing this cache - they're lacking bandwidth.

#5839194

asturur

2020-10-06 14:13

i wonder how 3080 has underwhelming performance.

#5839196

cucaulay malkin

2020-10-06 14:14

Kosmoz:

AMD said they consider GDDR6X too power hungry

it's lower power than gddr6 regular,almost hbm2 level https://images.anandtech.com/doci/15978/MicronPower2.png

#5839198

Truder

2020-10-06 14:17

JiveTurkey:

I assumed this Infinity cache is AMD answer to RTX IO.

RTX IO is just nvidia branding for their implementation of DirectStorage

#5839210

kanenas

2020-10-06 14:36

cucaulay malkin:

they have to compete with 16gbps 256-bit against 19gbps 320-bit I bet this will be hit&miss,can't make up for 1.53x bandwidth advantage for competing with a 3070 this may show,at least to some extent at 4K.unless this is hbcc rebranded. gddr6x is nvidia+micron collaboration,amd will never get this.

cooperation does not mean exclusivity

#5839212

Alessio1989

2020-10-06 14:36

Not another hardware-closed API please... Waiting for DirectStorage, although with 32-64GB RAM system it will not be such huge improvement and I bet in 10 years those APIs will be completely abandoned...

#5839214

Undying

2020-10-06 14:38

Hbm2e really needs to find its way to gaming gpus. Amd make that happen already.

#5839224

schmidtbag

2020-10-06 15:00

Considering it's "infinity", makes me wonder if this is actually supposed to be a shared cache for APUs. Right now, memory bandwidth is the #1 issue for APUs and the caches they have are just simply not sufficient. dGPUs don't really need a fancy cache, since you can just widen the memory bus for a major bandwidth increase.