The next logical step for AMD is on-package HBM for its APUs, not hex-core CPU.
Right, before DDR5 arrives adding HBM is the only way out of the memory bottleneck 2400G is already reaching. The question then becomes is an MCM feasible on AM4? The Raven Ridge die imo is too low end to warrant comparably costly packaged memory. A higher end APU die that is only usable when combined with HBM is likely too costly to reach a big enough audience to warrant a separate die design. The only option left (and technically not an APU, but CPU + dGPU like Kaby Lake G) is a Zeppelin (or successor) die combined with Vega M and HBM, but intuitively such an MCM would already be too big for an AM4 package, wouldn't it?Well I thinking of something with 4GB of on package memory, but I forgot to mention it.
Here is a 1600X@3.6GHz version 4.0.4. It is 200MHz higher base clock, but I would say the single and multi-threaded scores are within "ball park" numbers.GB4 4.0.3 is rather old and probably has problems identifying Ryzen cache hierarchy and the score is probably not comparable to newer versions.
Here is a 1600X@3.6GHz version 4.0.4. It is 200MHz higher base clock, but I would say the single and multi-threaded scores are within "ball park" numbers.
https://browser.geekbench.com/v4/cpu/2253116
Well I'm thinking of using a single die with a 256-bit bus on package. Yes it wouldn't be HBM/HBM2 but I don't think that much bandwidth is needed for iGPUs anyway.Right, before DDR5 arrives adding HBM is the only way out of the memory bottleneck 2400G is already reaching. The question then becomes is an MCM feasible on AM4? The Raven Ridge die imo is too low end to warrant comparably costly packaged memory. A higher end APU die that is only usable when combined with HBM is likely too costly to reach a big enough audience to warrant a separate die design. The only option left (and technically not an APU, but CPU + dGPU like Kaby Lake G) is a Zeppelin (or successor) die combined with Vega M and HBM, but intuitively such an MCM would already be too big for an AM4 package, wouldn't it?
People have made predictions of 200-400MHz, but that remains to be seen. The 2400G despite delidding, and liquid metal cooling was not able to achieve higher clocks.Hmmm, I wonder then if the 2600x variant will then be base clocked at 3.8Ghz or higher or will be able to turbo boost past that 4.0Ghz barrier?
There are a lot of inconsistencies with benchmarks on that website. Here is a link and you can see what I'm seeing. https://browser.geekbench.com/v4/cpu/search?dir=asc&q=1600x&sort=created_atThat's misdetected clock. That score for 4.0.4 is too high for stock 1600X. It's most likely overclocked to 4GHz.
People have made predictions of 200-400MHz, but that remains to be seen. The 2400G despite delidding, and liquid metal cooling was not able to achieve higher clocks.
We can call it 12nm, but that is mostly a marketing term. Since there is a move from the 9T library to 7.5T library for the transistors, there is a logic density change by virtue of smaller cells. But it's unlikely given the information so far that 12nm, originally labeled 14nm+, will offer much more performance over 14nm. Skylake to Kabeylake consisted mostly of higher base clock speeds. Until release that's all that we have seen increase so far. AMD never came out and said the performance or power advantage came from shrinking their own 14nm process. They instead compared it to other "industry" processes.Raven Ridge isn't on the GF12 node though.
I really hope ths mem-latency part is true. That alone should make ryzen 2 a better gaming CPU. Even more so, when you add Precision Boost 2.0 to the mix, Meaning higher relative clocks in all but the most well threaded games
Very unlikely considering the baseline I used was the 1600X using 2933MHz RAM scoring 5196@83.3ns.Yeah, if true, that difference would be huge for games.
But most of that could also be provided simply by using different ram. I don't think we can really conclude anything from so little information.
Noticeable improvements in memory latency:
1600X Memory Latency 5196 83.3 ns
2600 Memory Latency 4285 9.90 Moperations/sec
A difference of 17.5%. which amount to 68.7ns.
Very unlikely considering the baseline I used was the 1600X using 2933MHz RAM scoring 5196@83.3ns.
Keep in mind that latency advantage all but goes away in the multi-threaded test. Of course, it's also worth mentioning that the 2600X is probably running 2600MHz RAM.
Noticeable improvements in memory latency:
1600X Memory Latency 5196 83.3 ns
2600 Memory Latency 4285 9.90 Moperations/sec
A difference of 17.5%. which amount to 68.7ns.
2933 CAS 14-16 vs. ? = 17% latency improvement. Going from CAS 16 to CAS 14 at 2933 will not gain you a 17% internal latency improvement. The RAM would have to be significantly faster, or we can choose the more reasonable option that CPU is designed with improved latency.2933MHz ram but at what timings? Those matter quite a lot for latency.
I'm wrong and you are right. Higher score relates to lower latency when compared to other Intel processors. https://browser.geekbench.com/v4/cpu/7123767I think it's the otherway around because the result is reported as score where higher is better. 2600 has ~101ns latency. Also 'memory copy' and 'memory bandwidth' scores are lower on 2600. Still 2600 performs better overall so maybe slightly lower memory performance doesn't matter or geekbench isn't very accurate test or 2600 is even better when equipped with faster memory.
What blank die?So could some one tell what the blank die does please??