Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[RnR_au]

Getting to 14.4 will take eons.
10.6 is what you'll be getting if lucky.

14.4 LPDDR6 has already been announced by SK Hynix and will apparently be launched/unveiled at ISSCC in Feb next year.



edit: there is also a presser from Cadence from the middle of this year about tapeout of their IP ; https://www.cadence.com/en_US/home/...stry-first-lpddr65x-144gbps-memory-ip-to.html

 

[adroc_thurston]

14.4 LPDDR6 has already been announced by SK Hynix and will apparently be launched/unveiled at ISSCC in Feb next year.

DRAM vendors lie as easily as they breathe.
It's years away from shipping.

 

[QuickyDuck]

14.4 LPDDR6 has already been announced by SK Hynix and will apparently be launched/unveiled at ISSCC in Feb next year.

View attachment 134558

edit: there is also a presser from Cadence from the middle of this year about tapeout of their IP ; https://www.cadence.com/en_US/home/...stry-first-lpddr65x-144gbps-memory-ip-to.html

The real question is: at what cost?

 

[RnR_au]

The real question is: at what cost?

Yup... don't you have any kidney's left?

 

[Kepler_L2]

Wide bus to fast LPDDR. 384 bits to 14.4 = 614GB/s. That's halfway between 5060 and 5070.

It's more like 448 GB/s (LP6-10667 plus metadata overhead). More than enough for 24 RDNA5 WGPs

 

[adroc_thurston]

More than enough for 24 RDNA5 WGPs

Will get a wee bit tight at desktop frequencies, but overall yes.

 

[LightningZ71]

It's adorable that people think that LPDDR will be available at anywhere near attainable prices anytime soon. Energy usage moved past performance per square foot as a limiting factor a while back and I've seen more than a couple of proposals for using LPDDR in DC spaces.

 

[adroc_thurston]

It's adorable that people think that LPDDR will be available at anywhere near attainable prices anytime soon

It's commodity DRAM so you can get it for normal money.

and I've seen more than a couple of proposals for using LPDDR in DC spaces.

LPDDR lacks both capacity and RAS to be widely used in general DCs.

 

[LightningZ71]

Commodity dram production that wasn't already previously booked is being displaced in every way possible for the in demand DC parts. Multiple vendors are bringing forward CAPEX and bringing online every possible unit of production they can to this end.

 

[soresu]

DRAM vendors lie as easily as they breathe.
It's years away from shipping.

Not so sure about that.

Current top of the line DRAM die densities are 32 Gbits, but this says 16 Gbits only.

Why do a Pinocchio to boost the share price and only go halfway?

 

[Gideon]

Well well, looks like the double cache ccd part is real after all

https://videocardz.com/newz/amd-confirms-ryzen-7-9850x3d-an-even-faster-8-core-cpu-with-3d-v-cachehttps://videocardz.com/newz/amd-confirms-ryzen-7-9850x3d-an-even-faster-8-core-cpu-with-3d-v-cache

 

[eek2121]

Well well, looks like the double cache ccd part is real after all

https://videocardz.com/newz/amd-confirms-ryzen-7-9850x3d-an-even-faster-8-core-cpu-with-3d-v-cachehttps://videocardz.com/newz/amd-confirms-ryzen-7-9850x3d-an-even-faster-8-core-cpu-with-3d-v-cache

Did you read the article?

While AMD has not confirmed the Ryzen 9 9950X3D2 yet

 

[Josh128]

Very interesting that AMD is actually choosing to do this now. Guess they needed more R&D + testing to determine it was safe for the masses. Makes sense as they probably have a load of top tier CCDs to sell and the 16 core just isnt a high enough volume part to use them all.

It is undoubtedly going to be a halo gaming part, but I doubt benchmarks are going to show much gains over 9800X3D, most of which can already boost to 5.4GHz by default / 1 click PBO. We'll be very lucky to see +5% on average in gaming. Im guessing this dude will be priced at $499 as current 9800X3D is still $449.

https://www.amd.com/fr/support/downloads/drivers.html/processors/ryzen/ryzen-9000-series/amd-ryzen-7-9850x3d.html

 

[lopri]

TSMC FTW

 

[LightningZ71]

If it manages to hold a +400Mhz boost advantage over the 9800x3d and at least a 100Mhz nT advantage, it should translate to a bit over 5% better performance on benchmarks and be enough to stay ahead of ARL-R.

 

[Joe NYC]

Very interesting that AMD is actually choosing to do this now. Guess they needed more R&D + testing to determine it was safe for the masses. Makes sense as they probably have a load of top tier CCDs to sell and the 16 core just isnt a high enough volume part to use them all.

It is undoubtedly going to be a halo gaming part, but I doubt benchmarks are going to show much gains over 9800X3D, most of which can already boost to 5.4GHz by default / 1 click PBO. We'll be very lucky to see +5% on average in gaming. Im guessing this dude will be priced at $499 as current 9800X3D is still $449.

https://www.amd.com/fr/support/downloads/drivers.html/processors/ryzen/ryzen-9000-series/amd-ryzen-7-9850x3d.html

Most reviewers test an out-of-the-box experience. Apples to apples, it would be 5.6 GHz / 5.2 GHz = 7.69% clock speed improvement.

But the most important thing is that AMD completely closed the clock speed gap of V-Cache processors. Here are some interesting default boost clocks:
5.2 GHz - 9800x3d
5.5 GHz - 9700x
5.6 GHz - 9850x3d
5.7 GHz - 9950x

 

[Josh128]

If it manages to hold a +400Mhz boost advantage over the 9800x3d and at least a 100Mhz nT advantage, it should translate to a bit over 5% better performance on benchmarks and be enough to stay ahead of ARL-R.

lol Im pretty sure 7800X3D is plenty enough to stay ahead of ARL-R. Barring a miracle from Chipzilla, 9850X3D should be enough to stay ahead of Nova Lake in gaming, even the bBBC version.

 

[DrMrLordX]

it should translate to a bit over 5% better performance on benchmarks and be enough to stay ahead of ARL-R.

Is Arrow Lake refresh even a threat? Er @Josh128 beat me to it lol

 

[LightningZ71]

Supposedly, regular arrow lake with a LOT of tuning can claw out a win or two and close the gap in several other games. I suspect that refresh will add a little more to that. 9850x3d should restore the gap some. I don't expect that it will retain it's pronounced lead over Nova Lake as I am in the camp that Intel will put in a lot of work to address some of ARL's biggest fabric flaws. I also think that the L2 cache sharing between pairs will help more than it hurts in low thread games.

 

[fastandfurious6]

motives for 9850x3d:

1. intermediary market refresh to keep up the lead (big gap until zen6)
2. testing out new efficient design in production for further improvements
3. anything else?

 

[Joe NYC]

motives for 9850x3d:

1. intermediary market refresh to keep up the lead (big gap until zen6)
2. testing out new efficient design in production for further improvements
3. anything else?

It does seem like there is a manufacturing tweak / optimization rather than just better binning.

 

[luro]

motives for 9850x3d:

1. intermediary market refresh to keep up the lead (big gap until zen6)
2. testing out new efficient design in production for further improvements
3. anything else?

Too good yields at TSMC maybe?

 

[CouncilorIrissa]

Probably just binning. 9950X3D CCD0 was already capable of hitting 5.5 GHz. Remember, 9800X3D's CCD silicon quality was just bad for hitting high clocks. Easily the worst 8C CCD outside of TR 9K war crimes.

 

[adroc_thurston]

Supposedly, regular arrow lake with a LOT of tuning can claw out a win or two and close the gap in several other games.

It would also crash after about 3 seconds in y-cruncher.

Is Arrow Lake refresh even a threat?

No.

 

[Josh128]

It does seem like there is a manufacturing tweak / optimization rather than just better binning.

I strongly doubt this. What are you basing this on?