AMD: Navi Speculation, Rumours and Discussion [2019-2020]

[Rootax]

Vega already presented a big change in cache organization with enormous 45MB of SRAM on die. Adding even more with Navi seems highly unlikely....

This number was a bit misleading no ? Like, it was the whole memory on the chip added up, a number we can't really compare to anything AFAIK... The L2 was 4mb.

I'm curious about Navi, but about why they didn't show more too... Maybe focusing on Zen 2 was the goal.

 

[Bondrewd]

but about why they didn't show more too

Because E3.
They have a E3 keynote.

 

[Rootax]

Because E3.
They have a E3 keynote.

Yeah, in a middle of all the others announces from the E3 participants. Yesterday was their day. And they didn't push navi much. But you're right, they have to keep some for E3.

 

[Ike Turner]

Yeah, in a middle of all the others announces from the E3 participants. Yesterday was their day. And they didn't push navi much. But you're right, they have to keep some for E3.

Besides Microsoft, nobody else is expected to talk about new HW so they won't really be drowned in there, especially given their partnership with MS, Sony & Google.

 

[Bondrewd]

Yeah, in a middle of all the others announces from the E3 participants. Yesterday was their day. And they didn't push navi much. But you're right, they have to keep some for E3.

Navi is for gaming.
E3 is a gaming event.
Kinda fits, doesn't it?

 

[Deleted member 13524]

They might also be contractually obligated to keep Navi details under wraps until one or two console makers spill the beans on their next-gen's specs, during or around E3.

Just like they didn't disclose Vega 12's details until apple announced their new macbooks with Vega Pro 16/20, and didn't disclose the TrueAudio DSP in March 2013's Bonaire until the PS4 was announced in November of the same year.

 

[iroboto]

Yeah, in a middle of all the others announces from the E3 participants. Yesterday was their day. And they didn't push navi much. But you're right, they have to keep some for E3.

It's June 10. E3 Day 1 I believe. They are more or less first out of the gate. Only MS and a few others doing a Day 0 event.

 

[Ryan Smith]

For all the times she's gone on stage, you'd think we'd just have the physical measurements for Lisa's hand for easier size analyses.

The problem is that if I have Ian ask for Lisa's hand, she's going to get the wrong idea. So will Ian's wife, for that matter.

 

[BRiT]

The problem is that if I have Ian ask for Lisa's hand, she's going to get the wrong idea. So will Ian's wife, for that matter.

Route the request through AMD's PR and HR departments then.

 

[Love_In_Rio]

Is anyone trying to measure that Navi's size?
Seems to be a direct replacement for Polaris 10, something between 200mm^2 and 250mm^2.





I'm actually betting on Vega 14nm parts as comparison.

If they were comparing to Radeon VII, then the +50% efficiency comparison would be much closer to the +25% IPC.

My guess is the architectural improvements are coming at very low power cost, so they're getting:
- x1.25 more performance out of new cache hierarchy and CU adjustments
- x1.2 higher clocks out of 14nm -> 7nm transition (which is the clock difference between a 1.45GHz Vega 64 and a 1.75GHz Radeon VII)

1.25 x 1.2 = 1.5, hence the 50% higher efficiency, or higher performance at ISO power.

There's also an update at Anandtech quoting Lisa Su who said the new efficiency comes partly from new process technologies.

I think it's a 40 CU / 2560sp part running at 1.75GHz, with a power budget close to 200W.
It should get Vega 64 performance at 300W / 1.5 = 200W.

Is Vega 64 295 watts at boost clock(1,5 Ghz)?.

 

[Deleted member 13524]

Is Vega 64 295 watts at boost clock(1,5 Ghz)?.

When set at Balanced Mode (which is the driver default AFAIK), Vega 64 consumes around 290W and clocks at ~1.45GHz average:

https://www.tomshardware.com/reviews/amd-radeon-rx-vega-64-undervolting,5234-2.html


I was picking typical power and average clocks, not boost values.

 

[Love_In_Rio]

When set at Balanced Mode (which is the driver default AFAIK), Vega 64 consumes around 290W and clocks at ~1.45GHz average:

https://www.tomshardware.com/reviews/amd-radeon-rx-vega-64-undervolting,5234-2.html


I was picking typical power and average clocks, not boost values.

Then a 64 CU Navi at 1,45/1,25 = 1160 Ghz would perform as a Vega 64 at 290 Watts but at 290/1,5 = 193 Watts right?.

If so, can´t see where to fit the rumors regarding the Gonzalo 1,8 Ghz clocks.

 

[Bludd]

The problem is that if I have Ian ask for Lisa's hand, she's going to get the wrong idea. So will Ian's wife, for that matter.

Can't you use a picture of her holding a previously released product which has known measurements to work out the size of her hands?

 

[Deleted member 13524]

Then a 64 CU Navi at 1,45/1,25 = 1160 Ghz would perform as a Vega 64 at 290 Watts but at 290/1,5 = 193 Watts right?.

I don't know why you would assume a new graphics chip made on 7nm for desktop would clock at 1.16GHz default. Lisa Su clearly said the 50% efficiency uplift comes from 25% higher IPC and higher clocks than Vega.
The die size is also a clear indicator that it has less than Vega 20's 64 CUs.

 

[Love_In_Rio]

I don't know why you would assume a new graphics chip made on 7nm for desktop would clock at 1.16GHz default. Lisa Su clearly said the 50% efficiency uplift comes from 25% higher IPC and higher clocks than Vega.
The die size is also a clear indicator that it has less than Vega 20's 64 CUs.

Because a NAVI CU at 1,8Ghz would be the equivalent in performance to a (1,8*1,25) 2,25 Ghz Vega CU. The chip to get the RTX 2070 performance would be way smaller than the one shown (no very far from a Radeon 7 in size).

 

[Deleted member 13524]

Because a NAVI CU at 1,8Ghz would be the equivalent in performance to a (1,8*1,25) 2,25 Ghz Vega CU.

The point of increasing IPC is to avoid hitting clock efficiency walls, yes.

The chip to get the RTX 2070 performance would be way smaller than the one shown (no very far from a Radeon 7 in size).

How much larger is RDNA's front-end? How much larger are RDNA's CUs with revamped cache sizes / hierarchy?

AMD found that Vega 20 would boost to 1.75GHz and average at ~1.69GHz. Why would AMD's second 7nm GPU (3rd 7nm chip) clock lower than their pipe-cleaner, even moreso with a smaller chip?


Navi wasn't developed to show off big TFLOPs numbers or achieve high compute throughput per area unit. They have Vega for that.
We shouldn't expect Navi RX 5700 to reach Vega 10's compute throughput. We should expect the opposite of that.

 

[Bondrewd]

Why would AMD's second 7nm GPU (3rd 7nm chip) clock lower than their pipe-cleaner?

New uArch can throw a wrench into just about everything.

 

[Love_In_Rio]

How much larger is RDNA's front-end? How much larger are RDNA's CUs with revamped cache sizes / hierarchy?

They seem to have gone the Nvidia way, more transistors throwed to no ALUs logic. They will get with 40 CUs the same they got before with 64. They are still 25%-40% behind Nvidia in efficiency anyway ( 2070 is 175 watts at 12nm).

 

[Ike Turner]

They seem to have gone the Nvidia way. They will get with 40 CUs the same they got before with 64. They are still 25%-40% behind Nvidia in efficiency anyway ( 2070 is 175 watts at 12nm).

They have been doing this for years but people seem hell bent on thinking that AMD has been sitting on GCN without doing anything. For example on compute workloads a Radeon VII (60CUs) is 2x-x2.5 (100%-150%) faster than a Fury X (64CUs) which is huge. And this obviously can't be attribute only to node shrinks (28nm-7nm). AMD's big problem is their horrendous execution (time to market & broken HW functionalities).

 

[Deleted member 13524]

They seem to have gone the Nvidia way. They will get with 40 CUs the same they got before with 64.

Not the same. The same gaming performance. They're sacrificing something here, and it's compute throughput.

They are still 25%-40% behind Nvidia in efficiency anyway ( 2070 is 175 watts at 12nm).

How can "faster than 2070" at 200W be 40% behind nvidia in efficiency, if there's a 12.5% difference in power consumption?

Are you perhaps comparing real 7nm GPUs from AMD with imaginary 7nm GPUs from nvidia?