i7-8700k & Coffee Lake CPUs

[Jito463]

56 minutes ago, Bananasplit_00 said:

It's one i known of for ages

Oh, I know, that's why I added the "just kidding" below.  

[Bananasplit_00]

13 minutes ago, Jito463 said:

Oh, I know, that's why I added the "just kidding" below.  

I should buy one though just to satisfy my craving for triple cores... 

[VagabondWraith]

2 hours ago, Bananasplit_00 said:

I'm hopeful. I REALLY want a tri core. 

Why not just buy a quad and disable a core? There's your tri-core.

[Bananasplit_00]

Just now, VagabondWraith said:

Why not just buy a quad and disable a core? There's your tri-core.

Because it's not a proper tri core? If I have 4 cores il use them. It's not about having three cores, its about it properly being a tri core with all the others broken. It's about having tri cores for sale, not buying quad-core CPUs and disabling one. 

[Princess Luna]

2 hours ago, done12many2 said:

I sure don't hope so.  While I don't generally use iGPUs, they do come in handy.  

I used to want that back when we were limited to the old 4 cores, making a line that ditched the iGPU in favour for an extra core or something of the sort sounded attractive but now that we finally have 6 cores it would be nice to keep the iGPU indeed.

[Bruno_A]

WTF? 4C/8T Core i3's?!

[Bruno_A]

4 hours ago, deXxterlab97 said:

 

 

WOW, 10GHz? That's some liquid nitrogen shit right there. 

[Dash Lambda]

1 hour ago, done12many2 said:

AMD is very far behind Intel in core strength when you consider that they can only come close up to 4 GHz or stock speeds. After that, they are nowhere in the same ballpark. They are also very far behind in IMC performance, to include RAM support. As an example of this, an overclocked Ryzen single core score in Cinebench R15 averages 155cb to just over 160cb.

An overclocked 7700k and even a Skylake-X chip can do 220+ in the same test. Much stronger cores. Fabric ain't going to fix this. They need to improve IPC.

An AMD chip is pretty much pushed to its limits from the factory. An Intel chip has a pretty fair bit of headroom left in it.

Assuming that they can fix these issues by the next release, we could assume that Intel's 10nm is going to take yet another jump forward.

Right now, Intel can afford to do shit like just add 2 more cores to a 7700k because they know that AMD hasn't actually caught up yet. They are close, but not there yet.  

By my understanding, the clock limitation is a manufacturing process issue which should be easily remedied when they move to a different one.
I don't know how the IMC wouldn't be an easy fix (in the context of changing the architecture) either. It's not even a huge issue now if you get the right RAM (not saying it doesn't need to be improved, it's just no longer the crippling issue it was at launch).

 

Clock for clock, core for core, it looks like Ryzen is about 10-15% behind, dependent on the workload. That's not huge. In fact, up until Intel basically stopped pushing, that's around what we expected from generation-to-generation improvement. I would expect AMD to at least manage the standard generational performance bump and all the biggest bug fixes with Zen2.

-I of course don't think AMD will, at least for long, be the clear performance king across the board at any point in the foreseeable future. Nor do I expect Intel to be.

[dizmo]

I'm glad I waited and didn't immediately jump on Ryzen. I'm going to wait and see how the i7 and i5 8400 perform. If they're in the same ballpark as the Ryzen 5, then it'll be a win for me. The i7 is particularly interesting with it's boost speeds. I imagine none of these numbers are confirmed though so much disappointment could come if I get my hopes too high.

 

5 hours ago, tom_w141 said:

Wasn't the 4c/8t i3 proven fake?

 

If true then go clean yourselves up 7700k owners because you've been FUCKED. i7 -> i3 in 8 months

Except...it's not locked like the i3.

5 hours ago, LinusTechTipsFanFromDarlo said:

If they wanted to add more cores to the i3 line then in my opinion the most logical thing to do is have...

 

Pentiums: 2 cores/ 4 threads (like current line up)

i3's: 4 cores/4 threads

i5's: 4 cores/ 8 threads

i7's: 8 cores/16 threads

i7 8800X: 16 cores/ 32 threads 

 

Does anyone agree with me? 

I don't think that'd make sense. They need something with 6 cores to compete with AMD in the same space.

5 hours ago, tom_w141 said:

Still top to the bottom of the mainstream stack in 8 months thats a colossal dick move, + R.I.P the resale value on a 7700k

They'd have to do it eventually, so I'm not sure why you'd be so upset.

Also the i7 can overclock, the i3 cannot.

5 hours ago, TechyInAZ said:

I'm really curious about the Zen+ core coming soon (i think very late 2018 or 2019), probably not a higher core count but if AMD can get in a better IMC and higher clocks, that'll be really good.

2019 is soon to you? Interesting perception of time.

5 hours ago, SpaceGhostC2C said:

I'd say it is enough for the above stated reasons. To be honest, what wasn't really appealing was Kaby Lake, being just some sort of Skylake redux. Even if Coffee Lake has the same IPC as Kaby Lake, it would still be a reasonable progress from Skylake: marginal IPC improvement and higher core count at the same price (not confirmed, but if they don't do in the mainstream the same they did in HEDT... RIP ). It's really Kaby Lake that starts to look as an unnecessary intermediate step. I wonder what sales are telling Intel, though.

To be fair, they didn't really need to increase performance that much. Their competition was pretty much nil. Why invest in continually increasing performance when you're already the best choice?

What they did do, was increase the mobile chips performance. Much better battery life, etc. They simply shifted focus onto other things, so it's not like they sat around and did nothing.

[tom_w141]

14 minutes ago, dizmo said:

Except...it's not locked like the i3.

i3 has a 4c/8t k sku... go read the OP?

 

14 minutes ago, dizmo said:

They'd have to do it eventually, so I'm not sure why you'd be so upset.

Because Kabylake has been around for 8 months. Knowing this was coming they shouldn't have released this at all. 7700ks will go from $300+ value to <$150 in less than a year since release. Thats madness.

 

14 minutes ago, dizmo said:

Also the i7 can overclock, the i3 cannot.

See above go read the op...

[done12many2]

38 minutes ago, Dash Lambda said:

By my understanding, the clock limitation is a manufacturing process issue which should be easily remedied when they move to a different one.

 

By my understanding, the clock limitation is a limitation of the silicon's speed and design.  It's a little bit more than the "manufacturing process" as you are implying or we'd be seeing subsequent batches increasing in speed.  This does indeed happen with Intel chips and batches.  This isn't happening with AMD chips.

 

Quote

 

I don't know how the IMC wouldn't be an easy fix (in the context of changing the architecture) either. It's not even a huge issue now if you get the right RAM (not saying it doesn't need to be improved, it's just no longer the crippling issue it was at launch).

 

Even with the right RAM, the IMC and memory performance on Ryzen is still pretty bad.  The best most optimized RAM on Ryzen still results in very bad memory latency times.  This is a much bigger deal than bandwidth which both companies provide plenty of. 

 

Quote

 

Clock for clock, core for core, it looks like Ryzen is about 10-15% behind, dependent on the workload. That's not huge. In fact, up until Intel basically stopped pushing, that's around what we expected from generation-to-generation improvement. I would expect AMD to at least manage the standard generational performance bump and all the biggest bug fixes with Zen2.

 

You're right, clock for clock and core for core it's better, but the problem is you are limiting one manufacturer in the equation to make the other sound better.  An Intel chip is not only slightly faster clock for clock and core for core, but it continues to clock up higher from there.  That's the part that makes it bad.   It's also how my 5960x, a chip from 2014 is able to outpace a current day Ryzen CPU.

 

I'm not sure what you base expectations on, but we have no track record to go off of to indicate that AMD can or can't just make things faster and fix the other stuff.  Sure it could happen, but so can the opposite.  

 

Sales aren't what the LTT community is trying to convince people they are.   AMD's profitability hasn't miraculously changed.  All of these things will impact their aggressiveness in future development.  

 

Do not take this wrong.  I'm rooting for them as it only helps everyone.  I'm just not willing to make shit up or hype something that is completely unknown.  I might hype something if there was a trend.  

 

Quote

-I of course don't think AMD will, at least for long, be the clear performance king across the board at any point in the foreseeable future. Nor do I expect Intel to be.

 

By removing a clear leader from both, you've attempted to imply that neither leads in performance, but that is absolutely wrong.

 

In value, AMD is the clear winner.

 

In performance alone, Intel is the clear winner.

 

Most will try to merge those into price groups, but we're back to talking about value, not performance.  

 

 

 

 

 

 

[SpaceGhostC2C]

57 minutes ago, dizmo said:

 

To be fair, they didn't really need to increase performance that much. Their competition was pretty much nil.

That's only partially true: they did compete against themselves, and if they don't have something better than their older stuff, people is not going to upgrade. When you produce a durable good that's always a concern.

 

57 minutes ago, dizmo said:

Why invest in continually increasing performance when you're already the best choice?

What they did do, was increase the mobile chips performance. Much better battery life, etc. They simply shifted focus onto other things, so it's not like they sat around and did nothing.

It doesn't really matter, we are not discussing how active or lazy Intel was, we are discussing whether Coffee Lake is an interesting lineup or not.

[done12many2]

1 minute ago, SpaceGhostC2C said:

It doesn't really matter, we are not discussing how active or lazy Intel was, we are discussing whether Coffee Lake is an interesting lineup or not.

 

...as I run off to hide.  

[Taf the Ghost]

It's still WCCF and we don't have outside confirmation on the i3 parts. I really doubt they're going to drop an unlocked 4c/8t part below 250USD. That would slaughter the sales of the higher SKUs.

 

The 8700k is only going to Benchmark faster because of that single core boost. The funny bit is that it actually will be a better Gaming CPU, but not because it boosts 100 Mhz higher at stock. Those extra cores matter in a number of Open World games already. That's the reason Ryzen is "smoother". The microstutters aren't GPU side but I/O. 

 

Though I do look forward to the "8700k is smoother!" stories. Sadly, I only got maybe 2 stories of "7740X is the best gaming CPU!". 

[Dash Lambda]

1 hour ago, done12many2 said:

By my understanding, the clock limitation is a limitation of the silicon's speed and design. It's a little bit more than the "manufacturing process" as you are implying or we'd be seeing subsequent batches increasing in speed. This does indeed happen with Intel chips and batches. This isn't happening with AMD chips.

I'm not very knowledgeable on the fine details, but from what I've read it's an issue with GloFo's 14nm process, and it should improve when they move to 7nm. I think it makes sense that, even if the manufacturing process can just be tweaked to fix it, we're not seeing the same sort of fine tuning from AMD that we do with Intel -After all, Intel has their own fabs, they have more control.

 

1 hour ago, done12many2 said:

Even with the right RAM, the IMC and memory performance on Ryzen is still pretty bad. The best most optimized RAM on Ryzen still results in very bad memory latency times. This is a much bigger deal than bandwidth which both companies provide plenty of. 

Could you provide a source?

I don't ask that as any sort of argument, I just want to see the numbers you're referring to. I know that, again with the right parts, you should be able to reliably get 3200Mhz CL16, which as far as I know is perfectly fine, but I only started learning more about memory after getting a Ryzen chip.

 

1 hour ago, done12many2 said:

You're right, clock for clock and core for core it's better, but the problem is you are limiting one manufacturer in the equation to make the other sound better. An Intel chip is not only slightly faster clock for clock and core for core, but it continues to clock up higher from there. That's the part that makes it bad. It's also how my 5960X, a chip from 2014 is able to outpace a current day Ryzen CPU.

I isolated IPC after making the point that the clock barrier is likely to be significantly improved with the next process. So I was saying that in the context of the next version possibly being able to match, or at least approach, Intel's clocks -I was not trying to pretend the current generation wasn't clock-limited, I was saying that the clocks by far contribute the most to the performance gap.

 

1 hour ago, done12many2 said:

I'm not sure what you base expectations on, but we have no track record to go off of to that AMD can or can't just make things faster and fix the other stuff. Sure it could happen, but so can the opposite.

Just the fact that they're getting back in the game. There's no certainty, but I find it hard to believe they would just quit after such a huge introduction.

 

1 hour ago, done12many2 said:

By removing a clear leader from both, you've attempted to imply that neither leads in performance, but that is absolutely wrong.

When I say "clear winner," I mean by a significant margin. When the performance between products is competitive, I don't call it a clear win.

Though that's just semantics, so... I can't really argue that my definitions are better than yours.

[done12many2]

7 minutes ago, Dash Lambda said:

I'm not very knowledgeable on the fine details, but from what I've read it's an issue with GloFo's 14nm process, and it should improve when they move to 7nm. I think it makes sense that, even if the manufacturing process can just be tweaked to fix it, we're not seeing the same sort of fine tuning from AMD that we do with Intel -After all, Intel has their own fabs, they have more control.

 

Could you provide a source?

I don't ask that as any sort of argument, I just want to see the numbers you're referring to. I know that, again with the right parts, you should be able to reliably get 3200Mhz CL16, which as far as I know is perfectly fine, but I only started learning more about memory after getting a Ryzen chip.

 

I isolated IPC after making the point that the clock barrier is likely to be significantly improved with the next process. So I was saying that in the context of the next version possibly being able to match, or at least approach, Intel's clocks -I was not trying to pretend the current generation wasn't clock-limited, I was saying that the clocks by far contribute the most to the performance gap.

 

Just the fact that they're getting back in the game. There's no certainty, but I find it hard to believe they would just quit after such a huge introduction.

 

When I say "clear winner," I mean by a significant margin. When the performance between products is competitive, I don't call it a clear win.

Though that's just semantics, so... I can't really argue that my definitions are better than yours.

 

Thanks for taking the time to clarify most of this.  

 

8 minutes ago, Dash Lambda said:

Could you provide a source?

I don't ask that as any sort of argument, I just want to see the numbers you're referring to. I know that, again with the right parts, you should be able to reliably get 3200Mhz CL16, which as far as I know is perfectly fine, but I only started learning more about memory after getting a Ryzen chip.

 

Timings are not what I'm talking.   I'm talking about what those timing result in.  3200 MHz at CL16 will result in a dramatically different outcome with regards to latency performance between a z270 platform and a x370 platform.  With regards to actual memory latency, they are in completely different worlds.  Not sorta close, or getting there.  The dual channel of x370 not only suffer from much higher latency than that of the dual channel from z270, it's also worse than the quad channel from x99 and x299.  

 

If you happen to have Ryzen, I'll gladly demonstrate my point if you want to share some testing?

 

@MageTank and I spend a fair amount of time messing around with memory tweaking.  He's been up and down Ryzen's IMC and memory performance so he's a better source for sharing first-hand experience.

 

Like I said though, if you'd like to compare some memory latency and bandwidth results, I'd be more than glad to demonstrate what I'm talking about.  

 

Diving into this stuff is a lot more fun then sitting around theorizing about it.  

[MageTank]

36 minutes ago, Dash Lambda said:

Could you provide a source?

I don't ask that as any sort of argument, I just want to see the numbers you're referring to. I know that, again with the right parts, you should be able to reliably get 3200Mhz CL16, which as far as I know is perfectly fine, but I only started learning more about memory after getting a Ryzen chip.

I don't know if I am allowed to use myself as a source, but I have experience with the Ryzen 5 1600 and modern Intel platforms when it comes to memory overclocking. Latency differences between the two platforms are extreme. Let's take a look at a result of 3353mhz C14-14-14-32-CR1 against just stock 3200 C14-14-14-34-2 Samsung B-Die XMP:

 

Ryzen @ 3353: 

My 7700k with 3200mhz XMP:

Notice how with even a 5% frequency advantage and nearly identical primary timings (Ryzen having the tighter tRAS and lower command rate, technically giving it yet another advantage), it's latency is still 77% slower than Intel's. Why is that? How can it be that with both a higher frequency overclock AND slightly better primary timings, can it result in a much slower latency? Well, there are a few reasons, such as tertiary timings being completely different, secondary timings (tRFC/tREFI not being tweaked/inaccessible on the platform), low cache clocks (north bridge/uncore) all impacting your memory controllers speed, but the biggest that I've noticed, is the architecture itself. It seems to access ram in an entirely different way. I did not notice it until after I saw the benchmarks for EPYC, but it seems Zen's architecture scales better with more ranks and banks, even if it's frequency is much lower as a result. This is contrary to what AMD has been teaching us in regards to Ryzen's superior scaling with frequency due to the infinity fabric. When it comes to memory performance, you cannot beat rank interleaving, no matter how much raw frequency you try to throw at it.

 

This is especially true when more channels are thrown into the equation, but i'll try to keep it focused on Ryzen. So, Ryzen has a very unique problem. A catch 22-esque problem. You see, the architecture is designed in a way that it scales tremendously with memory frequency (due to the Infinity Fabric Interconnect), but at the same time, it's also designed to take advantage of multi-rank, large bank DIMM's. The biggest problem being: you cannot have both. Due to the lack of multi-rank timings being provided to us, we cannot use multi-rank sticks AND achieve high frequencies at the same time. This is especially troublesome if you attempt to do this with 4 DIMM's populated (2DPC) as it's extra stressful on the signaling for the IMC. The sad part being, you do get an additional bandwidth uplift when more DIMM slots are occupied . This puts you in an awkward situation in which you have to choose lower latency access for memory, or faster access across your CCX's thanks to higher memory frequency. 

 

Ideally, they would give us the option to loosen multi-rank tertiary timings (slightly reducing rank interleaving's benefits, without completely throwing them away) in order to stabilize higher clocked, multi-rank kits. 

 

Even looking at one of the best memory overclocks ever achieved on Ryzen, we still see latency is abysmal:

That's 3700 C14-14-14-34-1, achieving only 61.3ns latency. That is worse than my 2133mhz JEDEC latency on Intel. This is again, only achievable by single rank kits (as seen in the CPU-Z screenshot above) and only occupying 2 slots (as seen by the 16GB size in the CPU-Z screenshot above). 

 

If you'd like something to compare that against, this is what I run 24/7 (on a good motherboard):

 

Now, I am by no means a master memory overclocker, but I certainly do not see myself as an amateur either. The 3700 C14 result on Ryzen is currently a world record. Mine is "just another Tuesday" in the Intel world. It's hard to compare the two when it comes to IMC's, because at this point, Ryzen's IMC is certainly worlds behind. I only hope they can tame that latency with Zen 2.0. 

[Dash Lambda]

48 minutes ago, done12many2 said:

-snip-

I was gonna say I'd be happy to, but it looks like MageTank got there first XP

Either way, I ran the AIDA64 memory test and got 76ns with my G.Skill Trident Z at 3200Mhz 16-16-16-36.

10 minutes ago, MageTank said:

-snip-

That's a bigger gap than I expected.

You said that Ryzen, at least excluding frequency, plays better with more ranks/DIMMs. How much so? I'm especially interested in those numbers because part of my plan was to get a second 2x8GB kit of RAM later on.

 

I wonder if making the infinity fabric independent of RAM speed would improve the latency, or for that matter if it would even be reasonable to do so... 'Cause, from what I've been able to find, it looks like Ryzen's IMC goes through the fabric to the CCXs. Ryzen also has a bit of an issue, though one that's easier to work around, with inter-CCX communication over the fabric, so it seems like it could be more the fabric than the IMC itself causing problems here, especially given that the ring bus Intel usually uses to connect the IMC seems more comparable to a CCX's internal interconnect.
I'm gonna have to read up more on microarchitectures...

[TheCherryKing]

Any word on what's going to happen with Intel's entry-level server/workstation processors for Coffee Lake? Will they still be called "Xeon E3" and end with "v7" or will they change the naming scheme of those too?

[Taf the Ghost]

42 minutes ago, TheCherryKing said:

Any word on what's going to happen with Intel's entry-level server/workstation processors for Coffee Lake? Will they still be called "Xeon E3" and end with "v7" or will they change the naming scheme of those too?

The new SP are Bronze, Silver, Gold & Platinum and just came out. There's no Coffee Lake SP CPUs coming. 

[MageTank]

1 hour ago, Dash Lambda said:

That's a bigger gap than I expected.

You said that Ryzen, at least excluding frequency, plays better with more ranks/DIMMs. How much so? I'm especially interested in those numbers because part of my plan was to get a second 2x8GB kit of RAM later on.

 

I wonder if making the infinity fabric independent of RAM speed would improve the latency, or for that matter if it would even be reasonable to do so... 'Cause, from what I've been able to find, it looks like Ryzen's IMC goes through the fabric to the CCXs. Ryzen also has a bit of an issue, though one that's easier to work around, with inter-CCX communication over the fabric, so it seems like it could be more the fabric than the IMC itself causing problems here, especially given that the ring bus Intel usually uses to connect the IMC seems more comparable to a CCX's internal interconnect.
I'm gonna have to read up more on microarchitectures...

AMD actually released their own data on this (which was very surprising, and coincides with my own findings) in this blog: https://community.amd.com/community/gaming/blog/2017/07/14/memory-oc-showdown-frequency-vs-memory-timings

 

 

With everything being identical (the only difference being dual rank sticks vs single rank), they were able to squeeze out roughly 5-8% more FPS in specific titles, though I wish they would have tested minimum framerates instead of averages, as you are likely to see greater benefits in that regard, but it's an improvement nonetheless. 

 

The more interesting results are the ones right after that, regarding auto-trained tertiary timings vs manually tweaked tertiary timings: 

 

People often mock me for spending copious amounts of time tweaking my ram for "nothing of value", but fail to understand that value does come out of it. It may not seem significant for most titles, but those that do benefit from it, tend to do so by a pretty fair margin. AMD shows a 9% improvement in averages (again, you'd see more in minimums) for ROTR, 15% in Hitman, and 9% in 3DMark SD. 

 

A lot of people don't quite realize how important tertiary timing training really is. It's what actually controls your bandwidth efficiency. You can have a kit that is 3200mhz with a peak theoretical memory bandwidth of 51.2GB/s (dual channel), but with poorly trained tertiary timings, you may only achieve 40GB/s (79% efficiency). This is known as Bandwidth Efficiency, and it's a pretty big deal. By tweaking your tertiary timings, you can improve your Bandwidth Efficiency, allowing you to achieve more of your bandwidth potential. By having more bandwidth, you complete your clock cycles faster, which will also reduce Round Trip Latency as a result. It's a very serious, yet often neglected part of memory speed when RAM is discussed in general. It's not something a board or memory manufacturer can advertise, as the auto training values are determined not only by your motherboard, but also by the CPU's IMC and memory IC's in question. Relying solely on automatic training, eventually costs you performance in the end.

 

As for Ryzen (consumer), you are still better off getting SR sticks, simply because the bandwidth and faster interconnect scales better in games than Rank Interleaving alone. If AMD gave us access to multi-rank timings, this would be an entirely different story, but for the time being, 8GB single rank sticks are the way to go. Now, they did give us access to 2DPC timings (albeit, not that many) so if you can get away with populating all 4 DIMM's, you'd have a bandwidth advantage over 1DPC users even if you suffered a slight frequency disadvantage. 3200 2DPC > 3466 1DPC in that regard. 

 

Threadripper might be an entirely different story, since it's more like EPYC, which scaled tremendously with multi-rank, large bank DIMM's. Here is a result sent to me by a friend using 4x16GB 3200 C14: 

As you can see, latency is almost on par with the guy that used 3700 C14 with single rank kits. 

 

I wish I could afford the platform myself, so that I could test how well it scales with various memory configurations, but I'll see if I can't find someone that has it, and walk them through doing so for the sake of science. Either way, it's looking a little stronger in the IMC department all around compared to smaller Ryzen. This result also needs a little extra work as well. 90% read efficiency, 90% write efficiency, but only 81% copy efficiency, lol. Would prefer to see 95% on both read/write, and 90% on copy if possible. I suppose I am asking for too much out of quad channel though. 

[TheCherryKing]

1 hour ago, Taf the Ghost said:

The new SP are Bronze, Silver, Gold & Platinum and just came out. There's no Coffee Lake SP CPUs coming. 

I'm not talking about the Scalable Xeons. I am talking about the ones that aren't scalable such as Xeon E3.

[Taf the Ghost]

2 hours ago, TheCherryKing said:

I'm not talking about the Scalable Xeons. I am talking about the ones that aren't scalable such as Xeon E3.

https://en.wikichip.org/wiki/intel/xeon_bronze

 

There doesn't seem to be any announcement of a real Xeon D replacement, from what I can find.  There's been Skylake E3s since 2015, and it looks like that's staying for now as the level under "Bronze". At least, that's the information I can find.

 

@MageTank

 

It should be noted that most of those results are DX12 with Nvidia high-end GPUs. Most of the benchmarkers had stopped doing that with Ryzen because there's such a huge performance drop off between DX11 and DX12 in comparative situations. (Nvidia's driver team has come up with a brilliant way of exploiting the Sandy Bridge uArch, which doesn't seem to work as well in DX12. And it works a lot worse with Ryzen.) So part of that uplift might be related to that specifically. (Vega 64 vs 1080 Ti on DX12 on Ryzen are going to be very interesting results.) 

[MageTank]

Just now, Taf the Ghost said:

https://en.wikichip.org/wiki/intel/xeon_bronze

 

There doesn't seem to be any announcement of a real Xeon D replacement, from what I can find.  There's been Skylake E3s since 2015, and it looks like that's staying for now as the level under "Bronze". At least, that's the information I can find.

 

@MageTank

 

It should be noted that most of those results are DX12 with Nvidia high-end GPUs. Most of the benchmarkers had stopped doing that with Ryzen because there's such a huge performance drop off between DX11 and DX12 in comparative situations. (Nvidia's driver team has come up with a brilliant way of exploiting the Sandy Bridge uArch, which doesn't seem to work as well in DX12. And it works a lot worse with Ryzen.) So part of that uplift might be related to that specifically. (Vega 64 vs 1080 Ti on DX12 on Ryzen are going to be very interesting results.) 

It's not related to DX12 though. You can test this on literally any platform with any API, and still see the results. If you have both a single and dual rank kit, you can also test this for yourself. Also, where did you see it was an Nvidia GPU used? 

 

I honestly have no idea what you are talking about here, so I'll need some clarification. 

[Taf the Ghost]

17 minutes ago, MageTank said:

It's not related to DX12 though. You can test this on literally any platform with any API, and still see the results. If you have both a single and dual rank kit, you can also test this for yourself. Also, where did you see it was an Nvidia GPU used? 

 

I honestly have no idea what you are talking about here, so I'll need some clarification. 

It doesn't list the GPU on the post, nor are any of the links to the rest of the work present. However, I'm almost certain that game testing was done with either a 1080 or 1080 Ti. Those numbers are too high for a Fury card.

 

If it's an Nvidia GPU, there is a DX12 gaming problem with Nvidia + Ryzen. There's a larger performance gap between the same Nvidia GPU under DX12 than DX11. So the better trained RAM might simply be working around that issue that's likely on Nvidia's driver more than the performance of the System.

 

That being said, faster RAM in totality helps both Intel & AMD a lot in Gaming performance.