Speculation: AMD's response to Intel's 8-core i9-9900K

[epsilon84]

This made my day: people paying a considerable premium for the latest and greatest product from Intel, yet saving a few bucks on memory. The acrobatics are amazing, the savings are real!

It actually makes sense to do this on the lower end mainstream chips, or when you're using a non Z370/390 mobo since they are limited to DDR4 2666 anyway.

On a 9900K though? I agree that it seems silly to penny pinch on RAM with a $450 CPU, especially as logic dictates a higher core count CPU also needs more memory bandwidth to keep those cores fed. These chips should be paired with high end B Die kits running at DDR4 4000 or higher speeds

 

[arandomguy]

I feel there is going to be some regional and situational considerations involved in that. At least here you have exponentially rising costs for memory past 3200CL16.

At least when I was pricing potential new builds last year while the cost difference for anything below up to 3200CL16 was very small (something like a $10?$15 USD difference?) to get past that for the kits required to hit 3200 on Ryzen was an $50-$60 extra. Anything faster than that you're looking at $100-$150. Even with a $350-$450 CPU cost that is relatively pricey to move up and I wouldn't consider it combined with diminishing returns.

Also with resell the ownership cost for CPUs to me calculates out differently. This is why I regret buying a 2500k way back instead of a 2600k. If we factor in resale used 2600k still carry a similar premium over 2500k like they did when new and have a larger resale market due to being the end of the upgrade. It's doubtful that applies to higher speed DDR4 when that time comes.

 

[epsilon84]

I feel there is going to be some regional and situational considerations involved in that. At least here you have exponentially rising costs for memory past 3200CL16.

At least when I was pricing potential new builds last year while the cost difference for anything below up to 3200CL16 was very small (something like a $10?$15 USD difference?) to get past that for the kits required to hit 3200 on Ryzen was an $50-$60 extra. Anything faster than that you're looking at $100-$150. Even with a $350-$450 CPU cost that is relatively pricey to move up and I wouldn't consider it combined with diminishing returns.

With Intel you can get away with higher latency kits compared to Ryzen, but RAM speed is still king. You'll definitely want to get the RAM speed as high as possible especially on a 8/16 CPU.

But I see your point about diminishing returns, it's the reason why I reused my old DDR4 2666 kit when I upgraded to a 8700K. I do run it at 3400 speeds though as I get the best performance that way, even with looser timings.

If I were to get a 9900K (not likely) I would probably upgrade the memory though.

 

[coercitiv]

Even with a $350-$450 CPU cost that is relatively pricey to move up and I wouldn't consider it combined with diminishing returns.

Are you saying you're considering buying the 9900K and combine it with a 3200 C16 memory kit?

 

[arandomguy]

Are you saying you're considering buying the 9900K and combine it with a 3200 C16 memory kit?

At most the entry Samsung B-die kits but I would consider 3200C16. Like I said the problem is exponentially cost scaling. Anything past that you're looking at $300 or more for the memory versus around $150 for 3200C16.

Also as I said CPU price differences at least going as far back as Sandybridge seem pretty much recoupable upon resale (up front cost isn't a consideration for me). I didn't really save any money going with a 2500k vs. 2600k, they basically still resell for nearly the same price difference comparing to buying new back 2011. I doubt high speed DDR4 is going to resell for the same premium vs 3200C16. Also there is more of a market for used peak platform CPUs, people with 2500k look to upgrade to the 2600k, how many people are looking to still put in 2500k's in that platform? Same situation will likely apply for Coffeelake CPUs down the line.

Another issue is before platform EOL for me I might need to do a capacity upgrade anyways for memory (16GB to 32GB) which causes further complications to consider.

I'll be looking at options around the 8700, 8700k, 9700k, 9900k, and 2700x again towards the end of this year. But I'll have to judge based upon the market situation by then, memory situation could be completely different from what I know and what usage I'm looking to get out of them.

 

[ChiefBigFeather]

Well, I too think they will wait it out.

What I really don‘t get is why L4 cache models to improve gaming performance is not a thing. Maybe gaming really isn‘t an interesting marked segment?

Ryzen has another advantage:
If you want a fast m.2 on Intel, you have to cut GPU pci-e bandwidth in half. M.2 express SSDs are a thing for years now, are they sleeping? Or is this anti Samsung?

 

[TheGiant]

This made my day: people paying a considerable premium for the latest and greatest product from Intel, yet saving a few bucks on memory. The acrobatics are amazing, the savings are real!

Well I am not going to buy DDR4 2400 here. I am talking about 50bucks down from buying a decent latency/higher freq memory for intel cpus, as they don't get as much from latency than ryzen chip does (where buying top latency DD4 3466 really makes a difference).
That makes the price difference like 100 EUR between those 2 CPUs and that is perfectly reasonable actually.

If I consider total system cost of a 1080TI, 16+GB RAM, top board, good cooling and then CPU, it doesn't make sense in this price category and use to buy a ryzen CPU. Even now the 8700K is a little better but the 9900K with 5GHz overlockability will destroy anything AMD offers here. At least for now.

 

[coercitiv]

I am talking about 50bucks down from buying a decent latency/higher freq memory for intel cpus, as they don't get as much from latency than ryzen chip does (where buying top latency DD4 3466 really makes a difference).

If low latency RAM would improve minimal framerates in games on Intel chips by 20% or more, would you buy it?

 

[ChiefBigFeather]

Even now the 8700K is a little better but the 9900K with 5GHz overlockability will destroy anything AMD offers here. At least for now.

Average FPS: Certainly!

Frametime experience with well tweaked ram: Are you sure?
I haven't seen good benchmarks comparing a 2700x to a 8700k when you tweak both chips to their full potential. Intel wins at average FPS but when comparing a tweaked 2700x to a tweaked 7700k and look at 1% lows, they perform about the same (see mindblank tech on youtube).

 

[TheGiant]

Average FPS: Certainly!

Frametime experience with well tweaked ram: Are you sure?
I haven't seen good benchmarks comparing a 2700x to a 8700k when you tweak both chips to their full potential. Intel wins at average FPS but when comparing a tweaked 2700x to a tweaked 7700k and look at 1% lows, they perform about the same (see mindblank tech on youtube).

as a former progamer, I am looking for min fps or low percentile fps

and every testing out there shows the 8700K is still better at this, I agree it was a problem with 7700K but not now

looking for what intel 8C can do

 

[TheGiant]

If low latency RAM would improve minimal framerates in games on Intel chips by 20% or more, would you buy it?

do you have any frametimes/etc testing with low latency on intel out there? if it does improve for 20%, then for sure yes I will buy top RAM
I care only for min/low percentile fps, because it defines gameplay for me, couldn't care less for average or median fps

 

[JoeRambo]

If you want a fast m.2 on Intel, you have to cut GPU pci-e bandwidth in half. M.2 express SSDs are a thing for years now, are they sleeping? Or is this anti Samsung?

Not sure what is wrong just running 1-2 M.2 drives off chipset. For desktop uses one would be HARD pressed to find limitations.
And if you do, neither of desktop systems is relevant - you need to go to Skylake-X or TR ( and even then, be prepared to run custom bleeding edge Linux to max out IOPS ).

 

[ChiefBigFeather]

Not sure what is wrong just running 1-2 M.2 drives off chipset. For desktop uses one would be HARD pressed to find limitations.
And if you do, neither of desktop systems is relevant - you need to go to Skylake-X or TR ( and even then, be prepared to run custom bleeding edge Linux to max out IOPS ).

Last time I checked, it adds a lot of latency, bringing it close to sata ssd performance.

 

[coercitiv]

do you have any frametimes/etc testing with low latency on intel out there? if it does improve for 20%, then for sure yes I will buy top RAM
I care only for min/low percentile fps, because it defines gameplay for me, couldn't care less for average or median fps

Video review here contains data for a 5Ghz 7700K running both 3466 CL16 and optimized 3466 CL14. I hope people can ignore the Ryzen data for the sake of this discussion

Relevant screenshots bellow, out of the games tested only AOTS did not show any gains on the Intel platform, the rest varied from 15% to as much as 40% improved minimums.

Spoiler: Screenshots

One may argue the 6c/12t and 8c/16t variants based on this architecture will exhibit a somewhat different response to fast (low latency) RAM due to the larger number resources (both execution and cache), but reality is games become ever more demanding and sooner or later the memory subsystem will face the stress 7700K is seeing now, especially considering clocks are creeping up past 5Ghz as we speak.

 

[TheGiant]

Video review here contains data for a 5Ghz 7700K running both 3466 CL16 and optimized 3466 CL14. I hope people can ignore the Ryzen data for the sake of this discussion

Relevant screenshots bellow, out of the games tested only AOTS did not show any gains on the Intel platform, the rest varied from 15% to as much as 40% improved minimums.

Spoiler: Screenshots

One may argue the 6c/12t and 8c/16t variants based on this architecture will exhibit a somewhat different response to fast (low latency) RAM due to the larger number resources (both execution and cache), but reality is games become ever more demanding and sooner or later the memory subsystem will face the stress 7700K is seeing now, especially considering clocks are creeping up past 5Ghz as we speak.

well I suspected it that major sites testing reviews are overlooking what I felt that the intel test are just pure another averages...

thanks

so by those results it looks like intel CPUs benefit from ULL RAM at the most crucial fps=low 0,1/1% the same as AMD ryzen, but Intel with ringbus is latency wise fast enough for average fps to be "almost untouched"

thanks again

 

[ChiefBigFeather]

as a former progamer, I am looking for min fps or low percentile fps

and every testing out there shows the 8700K is still better at this, I agree it was a problem with 7700K but not now

looking for what intel 8C can do

I'm not aware of any 8700k reviews comparable to the one at Mindblank tech. The 8700k doing better is plausible because of the L3 cache, but I would still be interested in seeing data. Do you have a link?

I'm always careful judging 0.1% lows. While they are important too, they are very susceptible to random variance in the benchmarks.

 

[Ranulf]

This made my day: people paying a considerable premium for the latest and greatest product from Intel, yet saving a few bucks on memory. The acrobatics are amazing, the savings are real!

Especially when there have been many posts/threads talking about ram speed and game performance in this forum for several years now.

 

[epsilon84]

Average FPS: Certainly!

Frametime experience with well tweaked ram: Are you sure?
I haven't seen good benchmarks comparing a 2700x to a 8700k when you tweak both chips to their full potential. Intel wins at average FPS but when comparing a tweaked 2700x to a tweaked 7700k and look at 1% lows, they perform about the same (see mindblank tech on youtube).

Here is the thing - 'tweaked RAM' is different for both platforms. For AMD, that would mean DDR4-3466 @ aggressive timings. For Intel, that would equate to DDR-4000 (or higher) at more relaxed timings. There is actually testing done already testing both the 8700K and 2700X (albeit overclocked) with tuned 3466 memory https://www.techspot.com/review/1655-core-i7-8700k-vs-ryzen-7-2700x/

I would argue this is actually a 'best case' scenario for the 2700X, as I said before 3466 LL is the AMD 'sweet spot' whereas Intel can take advantage of RAM speeds far higher than 3466: https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page3.html

This was done with first gen Skylake chips, fwiw. I would imagine a modern CFL chip with up to twice the cores/threads would respond even better to higher bandwidth memory than shown above.

 

[epsilon84]

Video review here contains data for a 5Ghz 7700K running both 3466 CL16 and optimized 3466 CL14. I hope people can ignore the Ryzen data for the sake of this discussion

Relevant screenshots bellow, out of the games tested only AOTS did not show any gains on the Intel platform, the rest varied from 15% to as much as 40% improved minimums.

Spoiler: Screenshots

One may argue the 6c/12t and 8c/16t variants based on this architecture will exhibit a somewhat different response to fast (low latency) RAM due to the larger number resources (both execution and cache), but reality is games become ever more demanding and sooner or later the memory subsystem will face the stress 7700K is seeing now, especially considering clocks are creeping up past 5Ghz as we speak.

I've done testing with my 8700K - 3400 CL15 beats 2666 CL12 in every game I ran so far, in terms of both mins and avg fps. Not by a whole lot, we're talking a few % here at most, but its definitely faster, at least measurably, if not perceptively. I can only imagine (though wish I had B Die modules to actually verify) this trend would continue on to say, DDR4 4000 CL17 vs DDR4 3466 CL14, for example.

I'll probably take a better look at min fps when I have time, my testing is by no means exhaustive, it was more of a curiosity thing for me to see how to best tweak my memory (Hyper X 2666, Hynix based) for the best gaming performance, but it suggested that a 8700K responded better to higher bandwidth at the expense of lower latency, at least at the RAM speeds that I tested with.

I can't imagine upgrading my RAM from 3400 CL15 to say, CL13 on a B Die kit would net a 20% increase in min fps though. As another poster above me said, 0.1% lows can be somewhat misleading as there tends to be far more variability than say, the 1% lows.

 

[JoeRambo]

Last time I checked, it adds a lot of latency, bringing it close to sata ssd performance.

It adds some latency, but "close" to SATA performance is overstatement of the decade, easy. I would not call my 970 PRO on Z370 close to SATA perf in any way. Sure i run with Melt/Spectre fixed disabled that might influence things, but performance is epic compared to sata 850 pro

 

[ub4ty]

It adds some latency, but "close" to SATA performance is overstatement of the decade, easy. I would not call my 970 PRO on Z370 close to SATA perf in any way. Sure i run with Melt/Spectre fixed disabled that might influence things, but performance is epic compared to sata 850 pro

https://www.techspot.com/review/1646-storage-performance-intel-z370-vs-amd-x470/

TL;DR : You get a performance decrease but nothing substantial enough for day to day performance impact

 

[ChiefBigFeather]

Here is the thing - 'tweaked RAM' is different for both platforms. For AMD, that would mean DDR4-3466 @ aggressive timings. For Intel, that would equate to DDR-4000 (or higher) at more relaxed timings. There is actually testing done already testing both the 8700K and 2700X (albeit overclocked) with tuned 3466 memory https://www.techspot.com/review/1655-core-i7-8700k-vs-ryzen-7-2700x/

Thanks for the review! Interesting read. Seems like the 8700k got quite a bit faster on many newer titles. Too bad they only made summaries for average fps :/

https://www.techspot.com/review/1646-storage-performance-intel-z370-vs-amd-x470/

TL;DR : You get a performance decrease but nothing substantial enough for day to day performance impact

Interesting! 4k random qd 1 used to be quite bad on chipsets if I remember right. Seems like they improved a lot! Sorry for not updating my knowledge and posting misinformation.

 

[epsilon84]

Thanks for the review! Interesting read. Seems like the 8700k got quite a bit faster on many newer titles. Too bad they only made summaries for average fps :/

Did the 8700K really improve with newer games? I suspect it may be more due to the fact that the 8700K has a lot more overclocking headroom than the 2700X, and these tests were done at overclocked speeds.

At stock speeds, using tuned 3466, they are actually quite closely matched (8700K is about 5% faster) but the fact that the 8700K is at 5GHz from 4.3GHz (16% OC) compared to the 2700X at 4.2GHz from 4.0GHz (5% OC) probably contributes to the impression that the 8700K 'improved'.

Also agreed that a min fps summary would have been useful too, as a gamer I pay more attention to min fps rather than avg fps as well.

 

[beginner99]

https://www.techspot.com/review/1646-storage-performance-intel-z370-vs-amd-x470/

TL;DR : You get a performance decrease but nothing substantial enough for day to day performance impact

Of course if the m2 is the only thing running on the chip set, then it has no impact. But as soon as you have other devices connected it will be a limit. Albeit I must admit that if you aren't copying from m2 to m2 I don't know when you would ever need that much bandwidth.

 

[PeterScott]

https://www.techspot.com/review/1646-storage-performance-intel-z370-vs-amd-x470/

TL;DR : You get a performance decrease but nothing substantial enough for day to day performance impact

Ironically this really only seemed to be a problem for AMD which has more PCIe lanes from the CPU.

Pretty much the opposite of the original claim surrounding Intels lack of PCIe lanes from the CPU.

Intels chipset based lanes actually ran just as fast as AMDs CPU lanes. It's only AMDs chipset lanes that slow down significantly.