- Mar 3, 2017
- 1,684
- 6,227
- 136
Unless the entire SoC shares the entire bandwidth instead of separating between the CPU and iGPU it'd be quite troubling to feed both componentsOn package RAM isn't a totally new idea. Intel already shipped it in the Kabylake+Vega wacky product they made back in 2018:
So yes, putting dedicated VRAM on package is certainly doable. But I still think that configurable, user replaceable LPCAMM is the way forward- 256 bits of LPDDR5X should be plenty of bandwidth for any APU that fits in a laptop thermal envelope.
Not even 256bits of LPDDR5X-8533 can get near that, they will need to include some cache and maybe RDNA3.5 is a bit more memory efficient than RDNA2 as well.Can 256 bit LPDDR5X feed the potential 40 CU RDNA 3.5 iGPU though? The 6700XT itself has 384 GB/s bandwidth..
RDNA 3 is 33% more memory efficient iirc but they will need even more efficiency or more/better cache technology unless Strix Halo is destined to be always bandwidth starvedNot even 256bits of LPDDR5X-8533 can get near that, they will need to include some cache and maybe RDNA3.5 is a bit more memory efficient than RDNA2 as well.
Indeed. It was a strange kind of cooperation back then. IIRC Apple demanded that and only them & Dell used that product.Wasn't Radeon already an AMD trademark in 2018?
Isn't this how Apple does things or can AMD make something similar for Strix HaloUnless the entire SoC shares the entire bandwidth instead of separating between the CPU and iGPU it'd be quite troubling to feed both components
Sarlak has been confirmed to include 32MB ICNot even 256bits of LPDDR5X-8533 can get near that, they will need to include some cache and maybe RDNA3.5 is a bit more memory efficient than RDNA2 as well.
Yeah, I have been wondering about memory bandwidth for a while...As FP32's TFs are based on clock speed; I believe AMD would clock Sarlak's GPU lower than usual. As calculated here, with 2500MHz, Sarlak's GPU theoretical speed is double of Strix Point which has half of memory bandwidth.....Can 256 bit LPDDR5X feed the potential 40 CU RDNA 3.5 iGPU though? The 6700XT itself has 384 GB/s bandwidth..
If they are aiming for LPDDR5X this late the reason has to be LPCAMM2, and just so happens 2 x LPCAMM2 = 256bits.Yeah, I have been wondering about memory bandwidth for a while...As FP32's TFs are based on clock speed; I believe AMD would clock Sarlak's GPU lower than usual. As calculated here, with 2500MHz, Sarlak's GPU theoretical speed is double of Strix Point which has half of memory bandwidth.....
Of course, the above Sarlak is for 256-bit LPDDR5x version, we should see LPDDR6 version one year later....
how do you know thatOf course, the above Sarlak is for 256-bit LPDDR5x version, we should see LPDDR6 version one year later....
What if they just went with soldered instead of implementing LPCAMM2If they are aiming for LPDDR5X this late the reason has to be LPCAMM2, and just so happens 2 x LPCAMM2 = 256bits.
Radeon branding comes from years before AMD acquired ATi after they transitioned from the earlier 'Rage' 3D accelerator branding.Wasn't Radeon already an AMD trademark in 2018?
LPDDR5 came about 17 months before regular DDR5 did.how do you know that
Ah. So a new REDANO brand is within the realm of possibility!The Qualcomm Adreno GPU brand is an anagram of Radeon in reference to their original low power ISA/µArch acquisition from AMD/ATi.
The thing is LPDDR6 will initially be very expensive and limited in supply, hence being relegated to flagship smartphones that are of course very expensive and sell in low volumes.LPDDR5 came about 17 months before regular DDR5 did.
It's not a stretch to imagine that LPDDR6 will follow a similar cadence and appear significantly before regular DDR6.
Pretty sure he's guessing. MB manufactures can just reuse their current boards and just add 'now supports Zen XXXX CPUs!!!" Or, call them x670E Super Overdrive Extreme or some other marketing gobbilty beloved patriot.Their source is MLID.
Yep, but this round LPDDR6 and DDR6 should launch within the same 2026 year.. Of course, nobody cares about CAMM2 anymore by thenLPDDR5 came about 17 months before regular DDR5 did.
It's not a stretch to imagine that LPDDR6 will follow a similar cadence and appear significantly before regular DDR6.
LPDDR6 are selling directly to OEMs which are much cheaper than DDR6 at retail pricesThe thing is LPDDR6 will initially be very expensive and limited in supply, hence being relegated to flagship smartphones that are of course very expensive and sell in low volumes.
When have laptop makers been so quick to adopt a new LPDDR version?
Probably just better memory overclocking with more expensive traces or extra PCIe 5.0 lanes.Pretty sure he's guessing. MB manufactures can just reuse their current boards and just add 'now supports Zen XXXX CPUs!!!" Or, call them x670E Super Overdrive Extreme or some other marketing gobbilty beloved patriot.
Go calculate memory bandwidth of 192-bit LPDDR6 and think why AMD going to launch Sarlak with 40CU? (instead of 32CU which is double the Strix Point's 16CU)how do you know that
You are just speculating.Go calculate memory bandwidth of 192-bit LPDDR6 and think why AMD going to launch Sarlak with 40CU? (instead of 32CU which is double the Strix Point's 16CU)