- Mar 3, 2017
- 1,622
- 5,891
- 136
They've even screen dumped the post in the article, so some of us are outed in the news flow now.Now the hype has reached videocardz
Is this the same Kepler whose in our forum?They've even screen dumped the post in the article, so some of of us are outed in the news flow now.
Pretty soon Lisa Su will be citing ATOT performance figuresNow the hype has reached videocardz
AMD Zen5 architecture is allegedly 40% faster core-to-core than Zen4 in SPEC benchmarks - VideoCardz.com
Rumor: AMD Zen5 is 40% faster than Zen4 in SPEC benchmarks According to reputable leaker “Kepler_L2”, AMD’s next-gen CPU architecture is said to provide (up to) 40% increase over Zen4. AMD’s Zen 5 CPU architecture appears to be gearing up for a major improvement over its predecessor, Zen 4...videocardz.com
I think it's fair to say that the pandemic had as much to do with that as the lack of competition. The fact that the 5800x was the only available SKU at the time speaks volumes about how bad its $450 MSRP was. Vermeer was also selling below MSRP well before Alder Lake was out.You also couldn't buy one for many months after launch. The only Vermeer available was the 5800X in that time period.
Are we expecting an AM5 variant of this one too (similar to 8700G for Zen4)? Would be really nice I think.Strix Halo
N3E
16 core Zen 5
40 CU RDNA3+
~50 TOPS XDNA2 NPU
256 bit memory bus + LPDDR5X-8533 = 273 GB/s
32 MB Infinity Cache
125W TDP
No. Only laptop or mini PC. Won't be sold as DIY (And the AM5 platform can't support it to be honest)Are we expecting an AM5 variant of this one too (similar to 8700G for Zen4)? Would be really nice I think.
Is this the same anandtech forums that you and Kepler posts on, or is there other Anandtech forum?
The "CUT-Down" version - it might since it will be 128 bit bus, but realistically how big of a chance for it exists?No. Only laptop or mini PC. Won't be sold as DIY (And the AM5 platform can't support it to be honest)
Good point. Maybe? But I wouldn't count on it to be quite honest.The "CUT-Down" version - it might since it will be 128 bit bus, but realistically how big of a chance for it exists?
By that logic you shouldn't have gotten half that from Zen 1 -> Zen 4 as the ALU count was unchanged. In reality it's considerably over 40% IPCZen 4 has 4 ALU units.
Zen 5 supposedly has 6 ALU units.
Going from 4 to 6 means an increase of +50%.
IPC never scales proportionally to increased resources, and if with +50% more ALU Zen 5 achieves IPC of +20-25%
I don't see the point of that to be honest- on a 128-bit DDR5 bus, that big GPU would be completely bottlenecked.Are we expecting an AM5 variant of this one too (similar to 8700G for Zen4)? Would be really nice I think.
Are we expecting an AM5 variant of this one too (similar to 8700G for Zen4)? Would be really nice I think.
The "CUT-Down" version - it might since it will be 128 bit bus, but realistically how big of a chance for it exists?
They BETTER NOT waste all that bandwidth on anything less than 48GB RAM!It can't be fit on AM5 because it has a 256bit bus.
Novel idea for a new product. Don't anyone patent this under your name, please!I don't see the point of that to be honest- on a 128-bit DDR5 bus, that big GPU would be completely bottlenecked.
They BETTER NOT waste all that bandwidth on anything less than 48GB RAM!
half height PCIe 5.0 daughtercard for Strix Halo that expands the 16GB on-chip RAM
A new AM5+ socket with CAMM2 modules would be quite rad! Could also support Strix point with half the channels.I don't see the point of that to be honest- on a 128-bit DDR5 bus, that big GPU would be completely bottlenecked.
I hope we'll see it in either it's own new socket, or at least available soldered down on Mini-ITX boards.
What does this mean? This proves nothing other than that the use in Zen1 4ALU was far from optimal. Do not get me wrong. I would like Zen 5 to be as good or even better, but I doubt that in the first generation the Zen5 core with 6ALU was expanded so much that it could optimally use 50% more execution units.By that logic you shouldn't have gotten half that from Zen 1 -> Zen 4 as the ALU count was unchanged. In reality it's considerably over 40% IPC
What does this mean? This proves nothing other than that the use in Zen1 4ALU was far from optimal. Do not get me wrong. I would like Zen 5 to be as good or even better, but I doubt that in the first generation the Zen5 core with 6ALU was expanded so much that it could optimally use 50% more execution units.
Of course, this will be verified soon.