Intel's i9-9900K Is Only 12% Faster than AMD's 2700X at Gaming, but 66% Pricier

Whether HT does or doesn't help in games, to me, is less of the issue that it helps in everything else. My 4690k is getting a bit long in the tooth. Even first-time loading up my browsers after a new boot sees 80-90% usage on 4 cores. Loading up specific games pegs it at 99% just trying to read/load files. My next CPU will likely be the Ryzen 2600 (if I can afford it before Zen 2 officially arrives), just to help with daily tasks/general usability.

It also ensures I have cores to spare for programs I May have running in the background. most videos/benchmarks will show you how they run with JUST those games running and barely anything else. What if I have some music going? A browser on the side, Discord up and running. I think it helps to have HT (or SMT, whatever) in general, and if the game can utilise it, great! Better to have it than not in a lot of cases.

That said, I still think for 'gaming only' 4-6 cores (without HT/SMT) are solid enough for 95% of people, but there's definitely a tangible difference overall with the extra threads.
 
A Bugatti is only 17% faster than a Corvette but costs 1066% more.***



***numbers are completely made up but probably close enough to make the point.
 
There are several games out now where even all eight threads of my 4770k will be fully pegged at times and sometimes nearly the whole time even. And when that happens my GPU usage just plummets and in many cases I can't even maintain 60 or even 50 FPS due to being CPU Limited.
 
misterbobby said:
There are several games out now where even all eight threads of my 4770k will be fully pegged at times and sometimes nearly the whole time even. And when that happens my GPU usage just plummets and in many cases I can't even maintain 60 or even 50 FPS due to being CPU Limited.
Click to expand...

what games?
 
I just realized this was an LGA 1151 chip. I had thought it was 2066 all along.
Zarathustra[H] said:
Look,

I agree that it is a large price premium for a small performance benefit, but that's always been the case at the high end. Prices never increase libearly with performance.

If I were in the market today, and I didn't have a soft spot for AMD, this is probably the CPU I would buy.

If you want the best possible per core performance without sacrificing the number of cores, there are no other options. This CPU is it.
Click to expand...


I just realized this is an LGA 1151 chip. I had thought it was 2066 all along.

With only 16 PCIe lanes, I've completely lost interest in this chip.

If this were a 44 lane chip on 2066 it would definitely be on my short list for my next motherboard/cpu upgrade.

I've switched back to thinking AMD now, though I really wish I could get just a little bit more per core performance.


The thing is, even Ryzen's 24 lanes are on the pathetic side too. I'd like at least 40 lanes, like my current i7-3930k. Threadripper is an absolute monster with 64, but I have no need for all those cores.

I wish they would launch a zen+ threadripper with only 8 cores, but ultra-binned for the highest possible clocks, and all those 64 lanes...
 
Last edited:
Zarathustra[H] said:
I wish they would launch a zen+ threadripper with only 8 cores, but ultra-binned for the highest possible clocks, and all those 64 lanes...
Click to expand...

They'd need 16 cores (two dual-CCX die) to get the PCIe lanes though- which is why the 2950X is extremely attractive if the per-core performance is there. If it's not, then Ryzen isn't attractive period, and that's where stuff like the 8700K/9700K/9900K come in.

ReaperX22 said:
It also ensures I have cores to spare for programs I May have running in the background.
Click to expand...

This is what people are missing- HT on the quad-core CPUs (be they a 7700K or a 2600K) helps to keep hard interrupts from hitting gameplay, and that smooths out your maximum frametimes- the 'minimum FPS' you actually feel.


And for those droning on about 'prove to me that HT means anything!', the above is what we're talking about, getting above four threads. Today, if you can get at least six threads (a la 700k/8400+) in some combination of hardware cores and SMT, you have enough CPU resources to keep the system smooth while gaming, and then you're limited by clockspeed and other factors external to the CPU. Anything with four threads or less is going to be a real limitation regardless of clockspeed.
 
IdiotInCharge said:
They'd need 16 cores (two dual-CCX die) to get the PCIe lanes though- which is why the 2950X is extremely attractive if the per-core performance is there. If it's not, then Ryzen isn't attractive period, and that's where stuff like the 8700K/9700K/9900K come in.



This is what people are missing- HT on the quad-core CPUs (be they a 7700K or a 2600K) helps to keep hard interrupts from hitting gameplay, and that smooths out your maximum frametimes- the 'minimum FPS' you actually feel.


And for those droning on about 'prove to me that HT means anything!', the above is what we're talking about, getting above four threads. Today, if you can get at least six threads (a la 700k/8400+) in some combination of hardware cores and SMT, you have enough CPU resources to keep the system smooth while gaming, and then you're limited by clockspeed and other factors external to the CPU. Anything with four threads or less is going to be a real limitation regardless of clockspeed.
Click to expand...
Well really even 4 cores with HT is already a limitation in some games. And even 6 real cores get fully pegged in Assassins Creed Odyssey.

BTW Here is Watch Dogs 2 eating my i7 for lunch. GPU usage drops to 59% here...

Watch-Dogs2-2018-06-26-19-33-49-170.png
img upload
 
IdiotInCharge said:
They'd need 16 cores (two dual-CCX die) to get the PCIe lanes though- which is why the 2950X is extremely attractive if the per-core performance is there. If it's not, then Ryzen isn't attractive period, and that's where stuff like the 8700K/9700K/9900K come in.
Click to expand...
There is the 1900x, (tom's review) which does have the full 64 lanes, if you can stomach the higher cost of the x399 platform, higher power draw (180w) and having your 8 cores split between two dies. A TR2 version would probably make it a much more attractive option.
 
Nobu said:
There is the 1900x, (tom's review) which does have the full 64 lanes, if you can stomach the higher cost of the x399 platform, higher power draw (180w) and having your 8 cores split between two dies. A TR2 version would probably make it a much more attractive option.
Click to expand...

That's just the previous gen though?
 
Nobu said:
There is the 1900x, (tom's review) which does have the full 64 lanes, if you can stomach the higher cost of the x399 platform, higher power draw (180w) and having your 8 cores split between two dies. A TR2 version would probably make it a much more attractive option.
Click to expand...


Yep, it may need two dies for all those PCIe lanes, but that doesn't mean all cores need to be active, just like in the 1900x.

I'd love a second gen version, where they go through all the scrapped dies for one or more cores not being functional disable half of them and then bin the ever living shit out of it, going for max clock on the remaining 4 cores on each die.

Essentially a 2900x, with two dies 4 cores on each and maybe a 4.6 - 4.8Ghz stock clock, or something like that, while still retaining the 64 lanes. Or, heck, I don't care, even if some of them were disabled and I "only" got 48 lanes, I'd still be happy.

I'd buy the shit out of that CPU.
 
It's an interesting idea, though with only quad-channel memory, you're moving more dies away from local memory- you'd really have to want PCIe lanes above all else. Works for storage intensive applications like NAS and perhaps lightweight VMs I'd think but I'm not sure I'd want that for anything too heavy ;).
 
IdiotInCharge said:
It's an interesting idea, though with only quad-channel memory, you're moving more dies away from local memory- you'd really have to want PCIe lanes above all else. Works for storage intensive applications like NAS and perhaps lightweight VMs I'd think but I'm not sure I'd want that for anything too heavy ;).
Click to expand...

I'm not sure how the 1900x did it, but it didn't seem to suffer from the non-local memory problem like the WX chips do. At least I don't recall that from the reviews

edit:

Looks like the [H] preview did mention the potential for infinity fabric latencies.

So compared to a regular Ryzen, on the one hand you have more memory channels, and top 5% binned CPU's, and on the other hand you have infinity fabric latency.

In the end, I wonder how it turns out. I'm going to go looking for Ryzen 1800x vs Threadripper 1900x benchmarks now.
 
Last edited:
Zarathustra[H] said:
In the end, I wonder how it turns out.
Click to expand...

AMD went with CCXs that have a local 'mesh' per two-CCX Ryzen die, so eight cores, and this die has two memory channels. One die is the largest that the consumer parts get, up to the R7, so eight cores on two CCXs with dual-channel memory. The CCXs are connected on-die, and this fabric link is pretty quick; there were latencies that were traced directly to RAM clock, which is how the desirable 3200MHz CAS14 'B-die' became popular for Ryzen, giving a clock boost to the interconnect while limiting latency.

This is also what led to Ryzen builds looking more expensive initially, as Intel platforms didn't benefit from boutique memory nearly as much as Ryzen suffered with more pedestrian RAM like 3000MHz C15.

TR moves up to a quad-die arrangement, with quad-channel memory where two die have two memory channels each, and two die must use the inter-die fabric to access memory. This imparts numerous NUMA levels alongside wildly disparate memory access latencies, depending on the number of active cores on the product (varies from eight for the 1900X to thirty-two for the 2990WX).

So the suitability is really, really dependent on workload, and sensitive to workload adjustments as well as optimizations at every level.
 
Master_shake_ said:
even tech power ups results show Hyperthreading doing nada at 1080p

https://www.techpowerup.com/reviews/Intel/Core_i7_8700K/12.html

https://www.techpowerup.com/reviews/Intel/Core_i7_8700K/13.html

fuck any resolution really. 1080p 720p 1440p and 4k all within margin of error.
Click to expand...

If you say so:
https://www.techpowerup.com/reviews/Performance_Analysis/Final_Fantasy_15/4.html

That was with a 1080ti. We will see even more separation with a 2080ti and future benchmarks.

There is no difference seen between a 6/12 and a 6/6 seen because these are controlled benchmarks with NOTHING else going on as well as gpu limitations.
 
Zarathustra[H] said:
Yep, it may need two dies for all those PCIe lanes, but that doesn't mean all cores need to be active, just like in the 1900x.

I'd love a second gen version, where they go through all the scrapped dies for one or more cores not being functional disable half of them and then bin the ever living shit out of it, going for max clock on the remaining 4 cores on each die.

Essentially a 2900x, with two dies 4 cores on each and maybe a 4.6 - 4.8Ghz stock clock, or something like that, while still retaining the 64 lanes. Or, heck, I don't care, even if some of them were disabled and I "only" got 48 lanes, I'd still be happy.

I'd buy the shit out of that CPU.
Click to expand...


Idea #2.

A dual die Threadripper. Die 1, binned at top 5% of dies. Die 2, some trash low binned die only utilized for its memory controller and PCIe lanes.

CPU essentially operates like a 2950x, but in permanent game mode and with higher clocks due to a combination of binning and the fact that 8C/16T produces a hell of a lot less heat than 16C/32T.
 
You must log in or register to reply here.
Back
Top