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

[LTC8K6]

The additional Spectre patches probably aren't helping things on the Intel side.

Or on the AMD side.

 

[epsilon84]

Well Gaming is about 10-12% when both are overclocked..which having looked at a large bench suite of games and removing the outliers like CS:GO is very good for ryzen.
https://www.techspot.com/amp/review/1655-core-i7-8700k-vs-ryzen-7-2700x/page8.html
When not overclocked that pulls the results closer to 2700x, so yes it is not as bigger deal as I thought.

I remember reading a couple of reviews early on that must have shown 2700x in a bad light..as I thought it was closer to 20% average.
Still my point still stands that next year pricing will be close and likely under 400$ IF icelake appears to fight zen2...if not you could see 12 core castle peak 3800x at 500$ just like 1800x.

There is a simple explanation for this - 99% of reviewers will have tested the 2700X using stock XMP timings with the RAM - this is a perfectly legitimate way of testing as XMP profiles are what the memory is designed to be run at. Techspot / Hardware Unboxed instead uses highly optimised RAM timing profiles provided by our resident CPU guru 'The Stilt'. These very aggressive timings tend to benefit Ryzen a lot more, as can be shown here:

https://www.computerbase.de/2018-04..._benchmarks_mit_ddr43466_und_scharfen_timings

As you can see, Ryzen 2700X gains about 7% from the aggressive timings, whereas the 8700K only gains about 1.5%

We are therefore looking at a 'best case' scenario here where an experienced overclocker who also knows how to correctly apply Stilt's RAM profiles (and also has access to high grade 'B Die' memory) can tune a 2700X to be within ~10% of a 8700K.

That being said, the 8700K can also be tuned further - Intel CPUs can generally handle much faster memory speeds, far higher than the ~DDR4-3466 ceiling of most 2nd gen Ryzen CPUs. The B-Die kits that can run 3466 at tight timings can usually handle 4000+ with looser timings, and this will actually benefit Intel more than tightening the timings at 3466 memory speeds. I can (sort of) confirm this with my own testing on my 8700K - my memory is only capable of DDR4-3400 @ CL15, but gaming performance is a few % higher than when running at DDR4-2666 @ CL12, as a point of reference.

Techspots own testing shows that when running at DDR4-4000 speeds, in certain CPU bound gaming scenarios, the gains can be substantial compared to lower memory speeds.
https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page3.html

Of course, I bet no reviewer would be game enough to actually review a 8700K running DDR4-4000 versus a 2700X with 'only' DDR4-3466, even if its highly tuned 3466 - imagine all the AMD fanboys screaming 'SHILL! YOU AREN'T RUNNING THE SAME MEMORY SPEEDS FOR BOTH AMD AND INTEL'

Personally, as a 8700K owner, if I actually had a high end B-Die memory kit, I most certainly would max out the memory frequency - my motherboard even claims it is capable of DDR4-4133+ speeds, though I can't actually verify this as I don't have memory remotely fast enough to test out this claim.

 

[french toast]

There is a simple explanation for this - 99% of reviewers will have tested the 2700X using stock XMP timings with the RAM - this is a perfectly legitimate way of testing as XMP profiles are what the memory is designed to be run at. Techspot / Hardware Unboxed instead uses highly optimised RAM timing profiles provided by our resident CPU guru 'The Stilt'. These very aggressive timings tend to benefit Ryzen a lot more, as can be shown here:

https://www.computerbase.de/2018-04..._benchmarks_mit_ddr43466_und_scharfen_timings

As you can see, Ryzen 2700X gains about 7% from the aggressive timings, whereas the 8700K only gains about 1.5%

We are therefore looking at a 'best case' scenario here where an experienced overclocker who also knows how to correctly apply Stilt's RAM profiles (and also has access to high grade 'B Die' memory) can tune a 2700X to be within ~10% of a 8700K.

That being said, the 8700K can also be tuned further - Intel CPUs can generally handle much faster memory speeds, far higher than the ~DDR4-3466 ceiling of most 2nd gen Ryzen CPUs. The B-Die kits that can run 3466 at tight timings can usually handle 4000+ with looser timings, and this will actually benefit Intel more than tightening the timings at 3466 memory speeds. I can (sort of) confirm this with my own testing on my 8700K - my memory is only capable of DDR4-3400 @ CL15, but gaming performance is a few % higher than when running at DDR4-2666 @ CL12, as a point of reference.

Techspots own testing shows that when running at DDR4-4000 speeds, in certain CPU bound gaming scenarios, the gains can be substantial compared to lower memory speeds.
https://www.techspot.com/article/1171-ddr4-4000-mhz-performance/page3.html

Of course, I bet no reviewer would be game enough to actually review a 8700K running DDR4-4000 versus a 2700X with 'only' DDR4-3466, even if its highly tuned 3466 - imagine all the AMD fanboys screaming 'SHILL! YOU AREN'T RUNNING THE SAME MEMORY SPEEDS FOR BOTH AMD AND INTEL'

Personally, as a 8700K owner, if I actually had a high end B-Die memory kit, I most certainly would max out the memory frequency - my motherboard even claims it is capable of DDR4-4133+ speeds, though I can't actually verify this as I don't have memory remotely fast enough to test out this claim.

Yea this kind of thing had occurred to me, cost of new memory is also a factor, but I think 3200 CL16 is a fair balance to compare.

 

[epsilon84]

Yea this kind of thing had occurred to me, cost of new memory is also a factor, but I think 3200 CL16 is a fair balance to compare.

The performance hit from using 3200 CL16 would be quite substantial for Ryzen unfortunately, as it exacerbates one of the main reasons Ryzen is slower at gaming to begin with - higher memory latency.

Basically if we use 3200 CL14 XMP as a baseline, a 2700X is about 10% off a 8700K in games (both at stock speeds) With tuned 3466 we are talking 5% difference, almost nothing. However, I suspect with CL16 3200 Ryzen would revert to being 15% behind because it is far more sensitive to memory timings than Intel.

This is both a good and bad thing - if you get some high end kits and know how to tweak memory timings, you can get some great gains with Ryzen. However, due to high memory prices RAM is often skimped on, for example in Australia I often see stores bundle Ryzen CPUs and mobos with budget DDR4 2400 to keep prices low which just makes me shake my head...

 

[french toast]

I would go with 3200 CL14 XMP.
That is not too expensive for customers buying these CPUs, it is not exactly cheap either when benchmarking the cheaper mainstream SKUs, but it does more accurately represent ryzen performance with being somewhat affordable.

 

[TheELF]

Well Gaming is about 10-12% when both are overclocked..which having looked at a large bench suite of games and removing the outliers like CS:GO is very good for ryzen.
https://www.techspot.com/amp/review/1655-core-i7-8700k-vs-ryzen-7-2700x/page8.html
When not overclocked that pulls the results closer to 2700x, so yes it is not as bigger deal as I thought.

I remember reading a couple of reviews early on that must have shown 2700x in a bad light..as I thought it was closer to 20% average.
Still my point still stands that next year pricing will be close and likely under 400$ IF icelake appears to fight zen2...if not you could see 12 core castle peak 3800x at 500$ just like 1800x.

Single threaded is not about averages...you get ~25-30% better performance in quite a bunch of these games and even up to ~50% better in fringe cases,average is besides the point if you're talking about single threaded or IPC.
It doesn't matter how many games there are that will bring even the highest GPU to it's knees even at 720p.
It matters if you are searching for best perf/$ or whatever else but not for single threaded.

 

[Batboy88]

What do you think Lol...Go out and get at least like the 7820x.
They should work on more clocks and if they think got something and a Revision in mind.

 

[Markfw]

What do you think Lol...Go out and get at least like the 7820x.
They should work on more clocks and if they think got something and a Revision in mind.

You have to be kidding. One of the worst CPUs Intel has ever made. and its still a lot more than the 2700x and gets beat by it. $470 vs $319 ?

 

[LTC8K6]

The 7820X is $399 at Micro Center, but the 9900K or whatever it's called will likely trounce it in almost every way.
The advantages the 7820X will have are quad channel ram, PCIE lanes, and AVX512.

 

[Markfw]

The 7820X is $399 at Micro Center, but the 9900K or whatever it's called will likely trounce it in almost every way.
The advantages the 7820X will have are quad channel ram, PCIE lanes, and AVX512.

For the 99% of the people in the world NOT near a microcenter, its still $469.

And yes, the new 8 core will most assuredly trounce it in performance. I still say its one of the worst Intel mas ever come out with.

 

[PeterScott]

For the 99% of the people in the world NOT near a microcenter, its still $469.

And yes, the new 8 core will most assuredly trounce it in performance. I still say its one of the worst Intel mas ever come out with.

In the reviews against the $500 Ryzen 1800X, I remember 7820x beat it at nearly everything. I haven't seen reviews comparing it to the 2700X, but time marches on, and older processers do get surpassed.

So Skylake X -7820x doesn't seem that bad. Certainly not the worse Intel has come out with. Kaby Lake X was a joke, and a much worse effort from Intel.

 

[Markfw]

In the reviews against the $500 Ryzen 1800X, I remember 7820x beat it at nearly everything. I haven't seen reviews comparing it to the 2700X, but time marches on, and older processers do get surpassed.

So Skylake X -7820x doesn't seem that bad. Certainly not the worse Intel has come out with. Kaby Lake X was a joke, and a much worse effort from Intel.

At the time the 7820x was $599 and the platform was also more expensive, and yes it won then.

But I am talking todays prices and performance..

 

[coercitiv]

So Skylake X -7820x doesn't seem that bad.

Skylake X did not deliver, everything that recommends it is platform specific, not chip/arch specific. Great expectations (including from me personally), few encouraging results.

Think of it in the following terms: at 8-10 cores Broadwell-E was likely a better arch had we normalized for power consumption and 14nm maturity. Hopefully this experiment will ensure better results & scaling with the new cache structure in future iterations.

 

[PeterScott]

Skylake X did not deliver, everything that recommends it is platform specific, not chip/arch specific. Great expectations (including from me personally), few encouraging results.

Think of it in the following terms: at 8-10 cores Broadwell-E was likely a better arch had we normalized for power consumption and 14nm maturity. Hopefully this experiment will ensure better results & scaling with the new cache structure in future iterations.

Based on what? 7900X is kicking 6950x to the curb in just about every multi-threaded test:
https://www.anandtech.com/show/1155...core-i9-7900x-i7-7820x-and-i7-7800x-tested/11

From the Conclusion to the Anand Review:
The most poignant of which should be when we pit this generation 10-core over last generations 10-core. The Core i9-7900X has a frequency advantage, an IPC advantage, and a significant price advantage, which should make for an easy steamrolling.
In the end, this is what we get: aside from some tests that are L3 memory sensitive such as DigiCortex, WinRAR, and some of the PCMark8 tests, the Core i9-7900X wins every CPU test. For anyone who was unsure about getting the 10-core on the last generation on a compute basis, this new one seems to be the one to get.

We are talking about $1700 part with monstrous 25MB cache, getting largely steamrolled by a $1000 part with 13.75MB cache.

It looks like a very clear win to me.

 

[Batboy88]

7820x is only yeah what $469 now...and it clocks too....Just not how it was we can actually start seeing more Limitation's of the Quad Core and even no matter how fast it is. Why I said no doubt in my mind next Future buy will be an 8 core 16 thread at least.

Realistically for now yeah probably not exactly needed for most People...If you were I don't know creating Content, Editing, Crazy Streaming stuff or something Else sort of Heavy, then yeah.

 

[StinkyPinky]

7820x and 9900k are two years apart from each other, seems unfair to compare the two.

 

[Markfw]

7820x and 9900k are two years apart from each other, seems unfair to compare the two.

Batboy88 seems to have an obsession with the 7820x, every post I have seen by him is about it, and how great it is.

 

[Batboy88]

Batboy88 seems to have an obsession with the 7820x, every post I have seen by him is about it, and how great it is.[/QUO

Batboy88 seems to have an obsession with the 7820x, every post I have seen by him is about it, and how great it is.

It is Awesome...Everyone Else says it is and Agrees....
I just think it was a little disappointment that they only got the 2700x after waiting that long up to only 4.3-4.4ghz or whatever....That said it also appears to be a good chip.

 

[Batboy88]

Like i Mentioned still don't have a huge urge to get anything else right now...the 7700k still runs really well.

 

[LTC8K6]

7820x and 9900k are two years apart from each other, seems unfair to compare the two.

Yeah, Cascade Lake X will be along soon for comparison.

https://en.wikichip.org/wiki/intel/microarchitectures/cascade_lake

 

[epsilon84]

Like i Mentioned still don't have a huge urge to get anything else right now...the 7700k still runs really well.

As it should, unless you have a pressing need for more cores / threads.

A 7820X though? Going to be made totally redundant (if it isn't already) once the 9900K hits. To be honest even today it makes little sense to get one over say, a 2700X unless you absolutely needed the AVX2 performance.

 

[Batboy88]

Because AMD Was right...and been right about More cores for a long time...I told see they are not Stupid and can't make cpus or a gpu lol....And now we see a lot more and new getting into the Limitation's with Less...

Trust me I like AMD...Done ok and pretty well with Ryzen for what it is...Other than a few Minor things and buggy stuff it all got fixed.

 

[Campy]

As bad as the 7820X may look today, the 7800X was even worse considering the 8700K came out only a few months after it.

On topic; AMD won't do anything except maybe adjust pricing. TR2 is coming soon too, so they have that launch to tide them over until Ryzen 3000 series.

 

[coercitiv]

We are talking about $1700 part with monstrous 25MB cache, getting largely steamrolled by a $1000 part with 13.75MB cache.

It looks like a very clear win to me.

It doesn't look clear to me at all: the smaller L3 cache introduced IPC penalties in many workloads, which SKL-X compensated for by raising clocks. That's why I said normalizing for power and process advantage would reveal . Here's what Tom's Hardware had to say on the matter:

Pushing all of the Core i9-7900X’s cores with Prime95 or LuxRender propels power consumption to incredible heights. You do get 48 percent more rendering performance in LuxRender, but at the expense of 58 percent-higher power use. This approach has the elegance of a sledgehammer. Then again, if you need speed at any cost, Core i9-7900X is top-notch.

The transition to the new mesh topology was a bit painful, though necessary. Hopefully when the L3 cache size gets a size bump and a latency cut we'll witness those clear wins the doubled L2 cache was supposed to bring.

 

[PeterScott]

It doesn't look clear to me at all: the smaller L3 cache introduced IPC penalties in many workloads, w

That is Skylake X/S comparison on IPC and wouldn't be cache size, that would bus topology penalty.

That Toms review you mention actually has IPC comparison vs the 6950x:

https://www.tomshardware.com/reviews/intel-core-i9-7900x-skylake-x,5092-3.html

Guess what, 7900x beats the 6950x clock for clock with only half the cache.

So again 7900X simply beats the 6950x everywhere, every way, with half the cache for $700 cheaper.

It's a total win over 6950x.