AMD to Make “Big Splash” with Overclocking on Ryzen 7000 Series Zen 4 Processors

[Tsing]

The overclocking potential of AMD's next-generation platform and family of processors should be pretty exciting, if the latest comments from executives are to be believed.

Go to post

 

[Dan_D]

AMD has a history of overpromising and underdelivering on this type of thing. The fact is, AMD needs to bin its CPU's pretty close to the edge of what the silicon can do in order to be competitive. It's the same thing Intel's been doing. Competition is too fierce right now. The only way Zen 4 can be an overclocking monster is if the architecture has massively improved IPC to the point where it can compete at lower frequencies than current CPU's can while still maintaining enough ability to scale up the clock speeds.

But that theoretical situation begs the question: Why not simply clock the CPU's higher in the first place and totally dominate? Likely, power consumption and heat would be the reason. Both of those things are hinderances to overclocking. I expect Zen 4 to be good, but I'm not expecting a "big splash" in terms of overclocking. Overclocking is a niche and I'm not entirely sure AMD understands that part of the market a whole lot better than Intel does.

 

[Uvilla]

I would bet they mean their PBO can go fairly higher as a percent than now. Not that you will get secret ghz on top of pbo.

 

[THUMPer]

They need to focus on DDR5 optimizations out of the box. Plug it in and go.

 

[Dan_D]

They need to focus on DDR5 optimizations out of the box. Plug it in and go.

Yeah.....that's not going to happen. AMD can make a memory controller, but it isn't good at writing AGESA or reference BIOS code to make that happen. Plus, the QVL standards vary across different manufacturers of the motherboards. Intel has tighter reigns on things like this and DDR5 is still a bit of a **** show on Z690. As was DDR4 when it first came out on X99.

 

[LazyGamer]

They need to focus on DDR5 optimizations out of the box. Plug it in and go.

To echo Dan's response: DDR5 is... special. There's half a dozen separate voltages that affect memory speed potential, hundreds of separate timing settings with... millions+ of potential combinations. Performance is not consistent with the same settings between different ICs. Memory manufacturers are not paying attention to cooling, so the best kits out of the box are either Corsair's top kit for running stock or bare sticks from the likes of Dell for adding cooling to. DDR5 comes in at a JEDEC speed of 4800MHz for one DIMM per channel, and you're likely only going to get up to 5600MHz or perhaps 6000MHz before you need active cooling to keep the memory truly stable.

And then there's the silent killer - ECC. We've seen this on GPUs already, where VRAM pushed too hard will error out just enough to reduce performance without actually causing an application fault. DDR5 does this too, but it is not at all understood how to systematically detect it without extremely isolated testing.

I've learned more about memory in the last couple of months than I have ever known. Cost and effort was never really worth it, but this time around I committed to really understanding DDR5, particularly after I couldn't get my 6400 C32 kit to run stably anywhere near 6400 C32 - after I had three separate kits outright die on a board (all now returned, the kits and the board).

---

Now, for the other side of the coin: JEDEC has DDR5 speeds up to DDR5-8000 on the books. And while latency has gone out the door for the moment, similar to every previous DRAM UDIMM release, those pushing the edge of what DDR5 can do are returning results that are very close to the latencies seen on the best DDR4 setups, while also approaching the bandwidth available on quad-channel DDR4 setups.

And at those speeds, the performance downsides of DDR5 in comparison to DDR4 cease to exist.

So imagine a 5950X, but with +30% IPC uplift and twice the memory bandwidth. Are y'all ready to rock?