Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Bulldozer modules consist of 2 cores sharing some resources.

If you don't do your research it is not AMD's fault.
 
Dx2-fz-WwAgnaY6.jpg:large


;)

Miles better, than ever before.
 
Miles better, than ever before.
o_O What exactly is so much better? Those characteristics look like that because 2700x has 16MB L3 cache (organized as 2x8MB) , and that Zen 2 has 32MB... What did you expect, that you can fit 16MB of data in 8MB of cache space? Latency across all three levels looks the same, if anything 2700x is better at managing cache (59.06ns at max L3 size vs about 97ns for Zen2 - higher than actual RAM latency! - this has to be some bug, it should be ~75ns@32MB) and Zen 2 is paired with slow RAM. Or has worse memory controller (90ns vs 80ns for RAM). Not that this is going to have any meaningful impact on performance.

What IS interesting and what you missed completely, are the benchmark numbers, Zen2 shows 104 in single core, and IF this is at max turbo of 3.6GHz it would put Zen2 above i9-9900k in IPC. That's what you should be excited about, not L3 cache size. IF they can make 8 core out of it and clock it at 5GHz AMD would finally beat Intel's flagship. But that's two IFs.
 
12 core sample was paired with absolutely atrocious RAM, compared to 2700X ;).

Thats why it looks better ;).

In last paragraph you pinpointed why we should be excited. Zen+ was 3-5% behind Skylake/Coffee Lake/Kaby Lake in IPC, when normalised coreclocks, and RAM were used for both platforms. Zen2 has 13-15% higher IPC than Zen+. Cache latancies are one of the reasons for this ;).

I was expecting Skylake IPC parity. But it appears AMD will have IPC, and core count advantage over Intel, and parity with clock speeds. Everything appears to be tremendously going well for AMD.
 
It appears that 4C/4T Eng sample is in the wild. Cache latency with packages larger than 16 MB are on 55 ns mark. That is VERY good, for AMD CPU. Its the same level L3 cache latency we have seen with Skylake.
 
I recently replaced old X99 for X399, I'm looking to install 3 GPUs so it will be PCIE x16 x8 and x16, so one slot x8 still will be free
 
looks like x16 x8 x16 x8 (last free slot), still better than old X99 when I had x16 x16 using 2 GPUs or x16 x8 x8 using 3 GPUs
According to the AMD slide, X399 supports x16/x16/x16/x8, but I don't know if any motherboard offers that.
 
According to the AMD slide, X399 supports x16/x16/x16/x8, but I don't know if any motherboard offers that.

yeah, but even Zenith Extreme doesn't support that, I still need CPU I'm thinking about 12 cores mid-range solution, I have 2x Titan XP and AMD Pro Duo Polaris from last year
 
yeah, but even Zenith Extreme doesn't support that, I still need CPU I'm thinking about 12 cores mid-range solution, I have 2x Titan XP and AMD Pro Duo Polaris from last year
The 3 slots on the X399M Taichi and X399 Phantom Gaming 6 would be x16 electrically, but with little spacing on the first one.
 
The 3 slots on the X399M Taichi and X399 Phantom Gaming 6 would be x16 electrically, but with little spacing on the first one.

but last 3rd slot it's a mess, too close to connector(s), without riser no way to go, I had similar problem with X99
 
Last edited:
X570 supports Zen+ or newer.

Zen 2 supported via BIOS upgrade on some B350 and X370 or simply B450 and up.

The B450 and X470 may also need a BIOS update.
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.