Question AMD Rembrandt/Zen 3+ APU Speculation and Discussion

[izaic3]

Alright, so we've had some leaks so far. I don't know if any of it's been confirmed yet, as it's pretty early, but here is what I've surmised so far (massive grain of salt of course):

• Zen 3 "enhanced"
• DDR5 and LPDDR5 support
• PCIE4
• 6nm
• CVML
• Navi 2/RDNA/RDNA 2
• 12 CU
• Socket FP7/AM5
• Possible USB4
• 2021(unlikely) -2022 (more likely) release date

If if turns out to have RDNA 2 and 12 CU, I could see iGPU performance potentially almost doubling over Cezanne.

If I've made any mistakes or gotten anything wrong, please let me know. I'd also love to hear more knowledgeable people weigh in on their expectations.

 

[andermans]

To complement that, do we have a good source on frequency/power curves for cezanne/renoir? (Note that test app matters for the power usage at a given frequency though)

 

[LightningZ71]

Another variable, does AMD select/use the same set of "dials and levers" for their N7 GPU dies as they use for their APUs and CPUs. In other words, do they use the same track widths, percentages of dark die, cell strategy, etc?

 

[izaic3]

Based off of that tweet from 3 posts ago, I eyeballed a graph and did some work in Excel. It's kind of rough napkin math, but using this, I could see Rembrandt iGPUs having 12 cores clocked at 1475 while drawing 16 Watts. Of course how many watts it pulls is quite a variable. I know AMD's 15 watt chips usually are set to 10-25 watts depending on oem. Does 16 watts iGPU and 9 watts CPU sound reasonable? That's the ratio I was unsure of.

 

[izaic3]

I just realized he ran a new test today and manually lowered his voltage and got new results, so I've just updated my spreadsheet. I think clock speeds of 1650 Mhz are within the realm of possibility (again, assuming 16 watts, which I don't know how realistic that might be).

 

[uzzi38]

I just realized he ran a new test today and manually lowered his voltage and got new results, so I've just updated my spreadsheet. I think clock speeds of 1650 Mhz are within the realm of possibility (again, assuming 16 watts, which I don't know how realistic that might be).View attachment 43328

Btw I've been posting the charts here too: https://forums.anandtech.com/thread...na-architectures-thread.2579999/post-40487064

Also, if you want I can DM the spreadsheet. Just didn't want to post it in case somebody goes and messes with the data.

EDIT: Oh wait, I figured out how to do a read-only link. I think. Here: https://docs.google.com/spreadsheets/d/1R57Xfp0pES5AFwbT-zAf8FXfE_iIRmkY7iKN0-Azbdc/edit?usp=sharing

 

[izaic3]

Btw I've been posting the charts here too: https://forums.anandtech.com/thread...na-architectures-thread.2579999/post-40487064

Also, if you want I can DM the spreadsheet. Just didn't want to post it in case somebody goes and messes with the data.

EDIT: Oh wait, I figured out how to do a read-only link. I think. Here: https://docs.google.com/spreadsheets/d/1R57Xfp0pES5AFwbT-zAf8FXfE_iIRmkY7iKN0-Azbdc/edit?usp=sharing

Thanks man, I appreciate it. I didn't see your posts on here, haha oops! Made a hopefully more accurate prediction now that I've used your spreadsheet. I appreciate all of your testing! If everything goes smoothly, it looks like Rembrandt's iGPU could reach around 1700 MHz.

 

[uzzi38]

Thanks man, I appreciate it. I didn't see your posts on here, haha oops! Made a hopefully more accurate prediction now that I've used your spreadsheet. I appreciate all of your testing! If everything goes smoothly, it looks like Rembrandt's iGPU could reach around 1700 MHz.

View attachment 43372

Only thing I can add is the use of N6 vs N7P for desktop RDNA2 will probably allow an extra +100MHz on top at the same power. Nothing massive for an iGPU every little bit helps.

12CUs at 1.8GHz ain't nothing to sneeze at. To put that into context, my GPU running at 600MHz (across 40CUs) scored 3191pts. Or about 100pts less than a stock RX470. 12CUs at 1.8GHz is ever so slightly less compute, but with the front-end running faster I suspect the difference will be made up.

As for how an RX 470 compares vs current gen APUs, its still over a doubling of performance.

 

[movdx]

The real question about rembrandt is, is it going to be available before H2 / 2022? I can guess it will be announced at CES 2022 but availability?

 

[dr1337]

The real question about rembrandt is, is it going to be available before H2 / 2022? I can guess it will be announced at CES 2022 but availability?

I thought I read a rumor somewhere that rembrandt (or maybe it was van gogh?) taped out only a few months after cezanne. Even if I dreamt that rumor, the driver leaks are at a steady enough rate that there no way it comes out any later than CES22. But if its true that it taped shortly after Cezanne then I'd fully expect a computex announcement and summer release/availability.

 

[izaic3]

I thought I read a rumor somewhere that rembrandt (or maybe it was van gogh?) taped out only a few months after cezanne. Even if I dreamt that rumor, the driver leaks are at a steady enough rate that there no way it comes out any later than CES22. But if its true that it taped shortly after Cezanne then I'd fully expect a computex announcement and summer release/availability.

I've heard that rumor too, but to me it makes more sense that Van Gogh would release first. It could act as a pipecleaner for mobile RDNA2. It also seems so weird to me to release it so soon after Cezanne, which still isn't being sold in volume yet.

Unless that's the point, taping it out early so that AMD can deliver in volume upon release. Though that seems like it might be expensive to store all of those cpus, but idk.

 

[Mopetar]

I hadn't been following the rumors, but was that something AMD had intended or more of an accident, say development of one running a little late and the other finishing sooner than anticipated?

From a certain perspective it doesn't really matter if both compete for the same wafers. Phasing out one product for another a bit earlier than usual doesn't really matter.

Of course the rumors could always be wrong.

 

[uzzi38]

The main customer for Van Gogh isn't using it any more. At this point who knows what will happen to it.

 

[scineram]

The main customer for Van Gogh isn't using it any more. At this point who knows what will happen to it.

And who would have that been?

 

[uzzi38]

And who would have that been?

Let me remind you that _rogame previously talked about how it was targeted at "premium tablet devices"

 

[moinmoin]

It also seems so weird to me to release it so soon after Cezanne, which still isn't being sold in volume yet.

Cezanne is most certainly sold in volume already, in Germany it's readily available (right now 14 configurations from 3 brands).

The main customer for Van Gogh isn't using it any more. At this point who knows what will happen to it.

In the leaked roadmap Van Gogh started a new rail and got a designated successor (assumed to be Dragon Crest, though that name is unlike all the others) which would be odd for a semi custom chip not intended for the open market.

 

[andermans]

I suspect AMD hoped the main customer might have been the trailblazer for the market segment? At that point it wouldn't be semi-custom but there might not be many other customers lined up at this point and the sustainability of the market segment might be in peril.

 

[Shivansps]

Let me remind you that _rogame previously talked about how it was targeted at "premium tablet devices"

I really want to belive they didnt do the same "Kabini/Temash" mistake again, VG is way too big for that market.

 

[soresu]

In the leaked roadmap Van Gogh started a new rail and got a designated successor (assumed to be Dragon Crest, though that name is unlike all the others) which would be odd for a semi custom chip not intended for the open market.

I suspect the painter/artist names will carry on for certain APU's, and others will have this new mythical bird/creature codename format.

Zen4/RDNA3 Phoenix is rumoured to succeed Rembrandt while Raphael keeps the painter lineage going.

I think the painter's will be chiplet designs (with/without GPU), while mythical creatures will be for monolithic APU dies.

 

[uzzi38]

In the leaked roadmap Van Gogh started a new rail and got a designated successor (assumed to be Dragon Crest, though that name is unlike all the others) which would be odd for a semi custom chip not intended for the open market.

Dragon Crest is the correct name but I'm not sure it's a new product. So far the only rumour surrounding it is that it's a refresh.

 

[moinmoin]

Dragon Crest is the correct name but I'm not sure it's a new product. So far the only rumour surrounding it is that it's a refresh.

So if we won't see Van Gogh after all we'll at least have the chance to see its tweaked offspring Dragon Crest? Fine with me.

 

[Shivansps]

I really dont see whats the big deal with VG is, unless it has some crazy tech on it (like a 256bit LPDDR5 bus), it can serve as the mid range APU for notebooks petty well... and that may allow for more RMB dies to desktop were there is a severe APU supply issue. What it matters if it was a semi-custom chip?

Now if it is a semi custom and has something crazy on it, i could see the problem, but i i think in that case it would have been cancelled already.

 

[uzzi38]

I really dont see whats the big deal with VG is, unless it has some crazy tech on it (like a 256bit LPDDR5 bus), it can serve as the mid range APU for notebooks petty well... and that may allow for more RMB dies to desktop were there is a severe APU supply issue. What it matters if it was a semi-custom chip?

Now if it is a semi custom and has something crazy on it, i could see the problem, but i i think in that case it would have been cancelled already.

I don't think the 256b bus thing was accurate. I think 64b is far more likely.

There's nothing really all that special about VGH, it's just the first RDNA based APU we knew of, and people are still wondering where on earth it'll be used.

I say forget about it for now and focus on Rembrandt instead.

 

[TESKATLIPOKA]

Van Gogh has 8CU RDNA2 IGP, so only 64bit lpdd5 is not enough unless It is 6.4GHz.

 

[uzzi38]

Van Gogh has 8CU RDNA2 IGP, so only 64bit lpdd5 is not enough unless It is 6.4GHz.

You're forgetting that it's a 7.5-18W TDP chip and you're comparing it's bandwidth requirements to a 12W-45W chip.

 

[TESKATLIPOKA]

You're forgetting that it's a 7.5-18W TDP chip and you're comparing it's bandwidth requirements to a 12W-45W chip.

I don't think the standard 9W Van Gogh will have that much lower average clockspeed for IGP than 15W Cezanne.
I consider 128bit 3200Mhz as good enough for 15W Cezanne.
6.4GHz lDDR5(LPPDR5) with 64bit memory controller will be at best comparable to 3.2GHz DDR4 with 128bit memory controller, but most likely worse thanks to higher latencies.
If you say 128bit LPPDR5 is an overkill for 8CU RDNA2 I agree, but for example 64bit paired with anything other than 6.4GHz LPDDR5 doesn't look so great to me, but maybe I am too optimistic about Van Gogh's clocks.