Question Zen 6 Speculation Thread

[luro]

AMD inadvertently posted info on embedded Venice on AMD web site, then took it down.

96 cores = 8 x 12 core CCDs

https://twitter.com/x/status/1996157021335507100

Interesting. Does anyone knows what Annapurna is?

 

[Joe NYC]

Interesting. Does anyone knows what Annapurna is?

It said somewhere that it is Fire Range. Which is mobile Zen 5 based on the desktop Zen 5 but in a mobile socket.

 

[Joe NYC]

First signs of mobile Zen 6 - Medusa

https://twitter.com/x/status/1996587920711626879

 

[StefanR5R]

Does anyone knows what Annapurna is?

It said somewhere that it is Fire Range. Which is mobile Zen 5 based on the desktop Zen 5 but in a mobile socket.

The screenshot shows "Embedded Fire Range (EPYC Embedded 2005 Series)" and "EPYC Embedded Annapurna" as separate series. The latter is described in the screenshot as "highly integrated x86 CPU optimized for Network Control Planes ... for entry switches, routers, security appliances, and optical transport". So maybe a monolithic chip? And perhaps Zen 5, or 4 even.

 

[marees]

First signs of mobile Zen 6 - Medusa

https://twitter.com/x/status/1996587920711626879

Medusa point 1 hi — 45w
Medusa point 1 lo — 28w

AMD Zen 6 Medusa Point Surfaces In NBD Manifest With Surprising "High-TDP" And "Low-TDP" Variants

AMD Zen 6-based mobile Medusa Point SKUs are reportedly divided into two categories: 45W and 28W variants as per the leaked NBD shipping logs.

wccftech.com

 

[poke01]

Medusa point 1 hi — 45w
Medusa point 1 lo — 28w

AMD Zen 6 Medusa Point Surfaces In NBD Manifest With Surprising "High-TDP" And "Low-TDP" Variants

AMD Zen 6-based mobile Medusa Point SKUs are reportedly divided into two categories: 45W and 28W variants as per the leaked NBD shipping logs.

wccftech.com

Too bad they are rdna3.5 3nm

 

[LightningZ71]

If 4nm class Gorgon Point is boosting to 3.1Ghz, then 3nm class Medusa should make 3.4Ghz or so. It's certainly not going to be Panther Lake Xe levels of performance, but it should best Kraken point by a good 15+%

 

[desrever]

If 4nm class Gorgon Point is boosting to 3.1Ghz, then 3nm class Medusa should make 3.4Ghz or so. It's certainly not going to be Panther Lake Xe levels of performance, but it should best Kraken point by a good 15+%

The memory and cache are going to matter more than anything else for iGPU performance. The bottleneck has been memory size forever. I really doubt PL will be much different in actual performance.

 

[CouncilorIrissa]

The memory and cache are going to matter more than anything else for iGPU performance. The bottleneck has been memory size forever. I really doubt PL will be much different in actual performance.

PTL has much better memory support (LP5X-9600), so should end up quite a bit better.
edit: nvm, I thought the comparison was against STX

 

[LightningZ71]

KRK isn't very bandwidth limited. Medusa will be incrementaly better there. Same CU count as far as we know. Faster cores and faster GPU should give a decent boost.

 

[poke01]

KRK isn't very bandwidth limited. Medusa will be incrementaly better there. Same CU count as far as we know. Faster cores and faster GPU should give a decent boost.

medusa will be going up against nova-lake-h with C-series GPU. will get totally destroyed, dont know why AMD thought RDNA3.5 was acceptable in late 2026/early 2027

 

[DrMrLordX]

medusa will be going up against nova-lake-h with C-series GPU. will get totally destroyed, dont know why AMD thought RDNA3.5 was acceptable in late 2026/early 2027

The only iGPU AMD cares about is what's going into halo.

 

[marees]

The only iGPU AMD cares about is what's going into halo.

The medusa premium also has a GPU chiplet.

It would definitely be the next gen z3e handheld chip

Microsoft could also unleash (third party) series S based on MDS-P that compete with the Steam Cube (on both price & performance)

 

[Joe NYC]

medusa will be going up against nova-lake-h with C-series GPU. will get totally destroyed, dont know why AMD thought RDNA3.5 was acceptable in late 2026/early 2027

It's all about timing and time to market. Medusa with RDNA5 would cost AMD 6 to 12 months of delay in time to market. And would certainly arrive after Nova Lake.

For this period of time (6 to 12 months), it is more competitive for AMD to sell N3P based Medusa RDNA3.5 than N4P based Strix / Kraken / Gorgon RDNA 3.5.

Keep in mind that Medusa will be cost optimized, with better power efficiency (including LP cores) and will get a performance bump. It would not be a good idea for AMD to leave all of this on the table.

Being cost optimized (reasonable die size, monolithic) has bigger impact on the mass market than iGPU performance. Volume is on lower end of the notebook market, where Intel is pricing itself out of the competition with all of its modern parts.

 

[DrMrLordX]

The medusa premium also has a GPU chiplet.

It would definitely be the next gen z3e handheld chip

Microsoft could also unleash (third party) series S based on MDS-P that compete with the Steam Cube (on both price & performance)

Guess we'll just have to wait and see what actually makes it to market. Some are saying that AMD has killed off the z-series line and that z2e is the last we'll see. AMD actually released six variants on Strix Halo, which kind of tells us that maybe they would rather cut down a Medusa Halo than mess with future z-series based on Medusa hi.

 

[inquiss]

Because GPU configs are getting lighter, not more capable. If you want better graphics you go to a GPU or you get a product with rdna 5 attached. It's better segmentation

medusa will be going up against nova-lake-h with C-series GPU. will get totally destroyed, dont know why AMD thought RDNA3.5 was acceptable in late 2026/early 2027

 

[LightningZ71]

Exactly. Medusa low is targeting a cheaper segment and will be competitive there. Medusa high is targeting the higher end business, non gamer segment of the mass market. Premium is for where you start to care about graphics performance.

Medusa high should offer roughly mobile rtx 2050 performance and be usable for 1080p mid gaming.

 

[511]

Exactly. Medusa low is targeting a cheaper segment and will be competitive there. Medusa high is targeting the higher end business, non gamer segment of the mass market. Premium is for where you start to care about graphics performance.

PTL-U will be there for that and NVL-U as well NVL-U is entirely inhouse.

Medusa high should offer roughly mobile rtx 2050 performance and be usable for 1080p mid gaming.

PTL will be already doing that 2 year before and NVL is a upgrade in GPU over it

 

[LightningZ71]

But how much will that cost?

Keep in mind. PTL-U has a much smaller iGPU than PTL-HX. 4 Xe3 cores of the PTL configuration won't exactly run away with performance vs 4x RDNA 3.5 WGPs on N3P. Will they be better? Yes. But not so much better that it makes a difference at these performance levels. Even regular PTL-H has just 4 Xe3 cores, and is intended for systems with dGPUs. Only the X SKUs get 12 cores, and their pricing is expected to be near Medusa Halo range.

 

[511]

But how much will that cost?

Keep in mind. PTL-U has a much smaller iGPU than PTL-HX. 4 Xe3 cores of the PTL configuration won't exactly run away with performance vs 4x RDNA 3.5 WGPs on N3P. Will they be better? Yes. But not so much better that it makes a difference at these performance levels. Even regular PTL-H has just 4 Xe3 cores, and is intended for systems with dGPUs. Only the X SKUs get 12 cores, and their pricing is expected to be near Medusa Halo range.

Last part is not true Medusa Halo is way more expensive than PTL i would expect same pricing or lower pricing as ARL-H which is nearly Strix point pricing iirc

 

[LightningZ71]

You think that the PTL-HX SKUs are going to be priced on the same neighborhood as ARL-H?!?! Those 12 Xe3 cores are not going to be tiny, and to make it shine, it's going to need very high memory bins. It will not be cheap in the slightest.

 

[511]

You think that the PTL-HX SKUs are going to be priced on the same neighborhood as ARL-H?!?! Those 12 Xe3 cores are not going to be tiny, and to make it shine, it's going to need very high memory bins. It will not be cheap in the slightest.

Well Yes IDK how that's surprising it is literally a ARL Successor the 12Xe3 iGPU is 53mm2 N3E and the rest of compute die is 114mm2 18A 48mm2 N6 Silicon plus the dummy die and advanced packing and Medusa is targeting the same memory speeds.
LNL is 140mm2 N3B for the compute+GFX PTL is like 167mm2(114+53) Compute+GFX

 

[Joe NYC]

Last part is not true Medusa Halo is way more expensive than PTL i would expect same pricing or lower pricing as ARL-H which is nearly Strix point pricing iirc

Medusa Halo is also going to be in entirely different class of performance, both CPU and GPU.

 

[LightningZ71]

The memory and cache are going to matter more than anything else for iGPU performance. The bottleneck has been memory size forever. I really doubt PL will be much different in actual performance.

I really should have specified here. It's not getting to PTL-HX levels of performance, but MDS low/high should easily be able to cover PTL-H and it's 4Xe3 cores. Synthetic benches are already showing the 4 core config having similar performance to 840m. I suspect that it will go a bit higher. MDS l/h will definitely be better.

 

[511]

I really should have specified here. It's not getting to PTL-HX levels of performance, but MDS low/high should easily be able to cover PTL-H and it's 4Xe3 cores. Synthetic benches are already showing the 4 core config having similar performance to 840m. I suspect that it will go a bit higher. MDS l/h will definitely be better.

for the target Market of MDS price matters more and MDS is N3P which had wafer price increased already couple of times PTL is relatively like 90mm2 for the 4+0+4 die and for the Intel 3 iGPU Die is like 20-30mm2 and 48mm2 for N6 IOD plus advanced packing MDS will be monolithic most likely 140-150mm2 N3P