AMD Ryzen - Games in 1080p can improve more than 35% !

[XenosTech]

2 minutes ago, MageTank said:

I want to point out that the average physics score of the 1700x is ranging from 18k-21k depending on clock speeds (and likely various hardware configurations). The average physics score of the 6950x however, ranges drastically from 21k(3.4ghz)-29k(4.6ghz) depending on clock speeds. It's not exactly fair to compare the overclocked 1700X to the stock 6950X and pretend that their relative performance is identical. Especially once you factor in the Intel chips having more overclocking headroom in general. Hwbot lists the average overclock of the 6950x to be "4.2ghz". At 4.2ghz, the Firestrike scores I am seeing, show the 6950x pulling 24k on physics. 

 

Either way, Futuremark's "best CPU list" scores are highly inaccurate for what they can really do. They are not to be used by any means to actually gauge the real performance of a CPU, and certainly shouldn't be seen as the be all, end all when comparing two CPU's. 

I wasn't really look for relative performance of them... was just randomly looking at shit and came across it was was surprized by there those 4 cpus were stacked... I already know what the r7's are good at for what I plan to do with them, which is just VM related so now I don't have to get a small loan to build out a vm server for labs.

[MageTank]

Just now, XenosTech said:

I wasn't really look for relative performance of them... was just randomly looking at shit and came across it was was surprized by there those 4 cpus were stacked... I already know what the r7's are good at for what I plan to do with them, which is just VM related so now I don't have to get a small loan to build out a vm server for labs.

That was my exact plan too. VM farms for cheap. It's certainly an interesting option, compared to my aging 2x 5645 Xeon setup. Sadly, I just can't get over the IMC. If Zen's IMC was good, I honestly wouldn't care at all for it's gaming performance. I just want something that can do heavy lifting for cheap, under my own orders. I need memory to be blazing fast (not so much in terms of raw bandwidth, but mostly latency) and if I cannot fix that, it's not worth having for me.

[XenosTech]

10 minutes ago, MageTank said:

That was my exact plan too. VM farms for cheap. It's certainly an interesting option, compared to my aging 2x 5645 Xeon setup. Sadly, I just can't get over the IMC. If Zen's IMC was good, I honestly wouldn't care at all for it's gaming performance. I just want something that can do heavy lifting for cheap, under my own orders. I need memory to be blazing fast (not so much in terms of raw bandwidth, but mostly latency) and if I cannot fix that, it's not worth having for me.

Well hopefully they can address that with architect optimizations later on... or get the motherbaords to work some snake oil through the bios to help the imc with faster ram kits.... We already know the communication between the ccx's kinda improves with faster ram

[MageTank]

Just now, XenosTech said:

Well hopefully they can address that with architect optimizations later on... or get the motherbaords to work some snake oil through the bios to help the imc with faster ram kits.... We already know the communication between the ccx's kinda improves with faster ram

I'd wager the improved performance is on the latency end of things. As bandwidth improves, latency tends to improve as a result (assuming timings remain the same). Since the kit's I've been seeing recommended are the 3600 C16 kits (ran at 2933 or 3200), it likely means primary timings are remaining unchanged. This means latency should vastly improve, since JEDEC 2133 timings are 15-15-15-36 by default, gaining over 1000mhz on the frequency while only slightly loosening timings to C16, will drastically improve Round Trip Latency.

 

If we had access to the tertiary timings, we could further reduce latency, and likely see whether or not it has a more profound impact on the CCX (or Zen's relative performance in general). Sadly, we don't have this yet. You give me access to tertiary timings, and I assure you, I can make miracles happen. Even if the IMC itself won't agree with me, there are things we can do, corners we can cut to make great improvements at relatively no cost in any other performance metric. Once AMD does that, my faith might be restored. 

[XenosTech]

2 minutes ago, MageTank said:

I'd wager the improved performance is on the latency end of things. As bandwidth improves, latency tends to improve as a result (assuming timings remain the same). Since the kit's I've been seeing recommended are the 3600 C16 kits (ran at 2933 or 3200), it likely means primary timings are remaining unchanged. This means latency should vastly improve, since JEDEC 2133 timings are 15-15-15-36 by default, gaining over 1000mhz on the frequency while only slightly loosening timings to C16, will drastically improve Round Trip Latency.

 

If we had access to the tertiary timings, we could further reduce latency, and likely see whether or not it has a more profound impact on the CCX (or Zen's relative performance in general). Sadly, we don't have this yet. You give me access to tertiary timings, and I assure you, I can make miracles happen. Even if the IMC itself won't agree with me, there are things we can do, corners we can cut to make great improvements at relatively no cost in any other performance metric. Once AMD does that, my faith might be restored. 

I wonder if it's ryzen or just that the motherboard manufacturers haven't had enough time to implement that in the bios. We'll have to wait and see on that one later down as bios revisions come out.

[MageTank]

Just now, XenosTech said:

I wonder if it's ryzen or just that the motherboard manufacturers haven't had enough time to implement that in the bios. We'll have to wait and see on that one later down as bios revisions come out.

ASUS came out and said it was AMD. I believe them entirely, because Asrock doesn't have them either. We are talking about the two biggest manufacturers when it comes to memory overclocking. Every RAM world record, is held on either an OC Formula, or ASUS ROG board (Apex is overpowered). 

 

AMD even came out and said that they will be working in the future to improve DDR4 compatibility, but the sad part is, we can do this ourselves if they'd give us the tools to do it. 

[XenosTech]

Just now, MageTank said:

ASUS came out and said it was AMD. I believe them entirely, because Asrock doesn't have them either. We are talking about the two biggest manufacturers when it comes to memory overclocking. Every RAM world record, is held on either an OC Formula, or ASUS ROG board (Apex is overpowered). 

 

AMD even came out and said that they will be working in the future to improve DDR4 compatibility, but the sad part is, we can do this ourselves if they'd give us the tools to do it. 

Lol probably looking at it as "If we gave you the tools to do it, what would be the point of us ?" If only companies provided the community with tools to do some of their work, they'd fix a lot of shit faster and all they would have to do is just acknowledge  for the help, it's not like all of us want them to give us free shit for helping them fix stuff.