Ars Rips on intel with KL review

[cytg111]

Yep and the results are more or less like the old days meaning less than impressive results with ht
https://www.techpowerup.com/forums/...rks-core-i7-6700k-hyperthreading-test.219417/...

Nice thread, thanks for that. So it seems like, in games at least, that it is the extra cache that does the difference.

But I googled around a bit and the samples i could dig up sort of agreed with your assessment.
Here is an 4790k with HT disabled, going towards 100% cpu .

https://www.youtube.com/watch?v=RCkQH9K4ri4

and

https://www.reddit.com/r/battlefiel...psa_hyper_threading_improves_cpu_performance/

But hardly clinical evidence.. Still though. And watchdogs2 .. Yet it may still just be crappy code and not a sign of the divine threading descending on us from the heavens..

 

[jpiniero]

Apple qualcom and samsung kind of prefer their own cpu core...so it was doomed to be a failure and i said so 4 years ago in this forum also remarking perhaps x86 wasnt the best position to start from. From that i took so much flak i still have to play bf to get rid of some of it. It turned out the atom soc for mobile was bad so it went from impossible to worse. It was played to the tune of 4b a year.

Yeah I think they underestimated how much better suited ARM is for mobile, both from a cost and power efficiency standpoint.

 

[Rifter]

8 cores is a bit much. I would be happy with six cores. In fact the way I see it six cores is the perfect upgrade:

-fewer cores helps overclocking

-a 50% multithreaded boost on paper makes the inner nerd in me feel like "I finally got an upgrade from Sandy"

In fact the 5280k is almost perfect for meeting these requirements so I have last one to add:

I want all that working on a modern $100 mobo.

That is what I hope AMD delivers.

I agree, 6 cores for a gaming build is where its at,4 cores are not going to be doing so well in the years to come. And i agree about mobo pricing.

 

[LTC8K6]

So much misunderstanding of ipc
"
Sandy Bridge to Ivy Bridge: Average ~5.8% Up
Ivy Bridge to Haswell: Average ~11.2% Up
Haswell to Broadwell: Average ~3.3% Up
Broadwell to Skylake (DDR3): Average ~2.4% Up
Broadwell to Skylake (DDR4): Average ~2.7% Up
"
http://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9

The progress is always exagerated. But having multiple generation tested paint the right picture.

Is that the test with 2133 DDR4? At the time, iirc, it was already known that SL shined much better with faster ram.

 

[majord]

As a gamer, IPC and frequency is what I'm looking for. Its not that I don't want more cores, its that I don't want to pay the price of lower real world gaming performance given the actual alternatives. The Intel HEDT platform delivers an older architecture and lower clock speeds. Its not a match for mainstream SL/KL.

are you sure about that?

https://www.computerbase.de/2017-01...re-i7-7700k-i5-7600k/3/#abschnitt_spiele_720p

the 'old' 5960x equal to a 6700K in CB's roundup. What happens when you OC both it? (orKBL for that matter, since it only affords 1-200mhz more headroom at most)

and more examples

http://gamegpu.com/images/stories/Test_GPU/Action/Homefront_The_Revolution/test/new2/h_proz.png
http://media.gamersnexus.net/images/media/2016/game-bench/battlefield/cpu/bf1-cpu-benchmark-dx11.png

I certainly agree, at least for HEDT, the value factor is not there, but that may change soon. The main point is though, you don't actually compromise gaming performance at all by choosing the 8 core BW-E .

 

[scannall]

Apple qualcom and samsung kind of prefer their own cpu core...so it was doomed to be a failure and i said so 4 years ago in this forum also remarking perhaps x86 wasnt the best position to start from. From that i took so much flak i still have to play bf to get rid of some of it. It turned out the atom soc for mobile was bad so it went from impossible to worse. It was played to the tune of 4b a year.

I had been saying that for a long time as well. Intel needed to go where the market was. Not try and drag the market to them. I'm sure Intel with their great engineers could have made a really great ARM product for mobile, but then they would have to directly compete. But no OEM is going to want to be locked down to Intel, as past history has shown that's a bad idea.

 

[KompuKare]

I had been saying that for a long time as well. Intel needed to go where the market was. Not try and drag the market to them. I'm sure Intel with their great engineers could have made a really great ARM product for mobile, but then they would have to directly compete. But no OEM is going to want to be locked down to Intel, as past history has shown that's a bad idea.

For the big players (Apple, Samsung etc.) it's the lack of flexibility which probably doomed Intel there. They keep talking about custom and foundry work, but so far we've never seen the kind of flexibility the console guys got from AMD. And mobile needs even more of that flexibility where (almost) nobody can do a complex phone SoC without licensing stuff from others. Also didn't help that Intel's iGPU isn't that great in terms of transistors/perf.
Yes, AMD were almost bankrupt at the time and desperate but I think their general approach is far more suitable than Intel's for this kind of thing. At least AMD didn't have $4 billion to blow on getting into phones/tablets.
Nvidia also blew ~$1billion on Tegra although they are trying to salvage that with automotive stuff.

 

[NTMBK]

Nvidia also blew ~$1billion on Tegra although they are trying to salvage that with automotive stuff.

They also managed to (finally) get a console win with Nintendo. If the Switch is a 3DS scale modest success, that will help them.

 

[Ranulf]

They also managed to (finally) get a console win with Nintendo. If the Switch is a 3DS scale modest success, that will help them.

Heh, that is a big IF. Given Nintendo's treatment of the Wii U, no way in hell I'm putting any money into the Switch.

 

[krumme]

Nice thread, thanks for that. So it seems like, in games at least, that it is the extra cache that does the difference.

But I googled around a bit and the samples i could dig up sort of agreed with your assessment.
Here is an 4790k with HT disabled, going towards 100% cpu .

https://www.youtube.com/watch?v=RCkQH9K4ri4

and

https://www.reddit.com/r/battlefiel...psa_hyper_threading_improves_cpu_performance/

But hardly clinical evidence.. Still though. And watchdogs2 .. Yet it may still just be crappy code and not a sign of the divine threading descending on us from the heavens..

Thanx for the video ! - man its nice to see someone having the same challenge. Its exactly the map ST quentin's scar 64 player where i have most problems.
Shame there is no with HT comparison as i can see?
I will have to make a vid but here is some observation. When i am playing ST quentin's scar 64 player, its in campaign mode, so we battle for sectors (i love that btw). And its damn important difference because it means 80-90% of the players goes towards a certain point at the same time. It often means 50 man and 2 tanks trying to take eg the church (i think it is - its sector 3/4 flag B). Think of that. 50 man and 2 tanks on 200m2 while a bomber drops for some extra destruction ! - its a massive load on the cpu and we might have all kinds of weak links here, drawcalls what not.
Compared to the start of the vid where we go in the 60 fps, and i notice dip below, i wouldnt be surprised if the same setup did go to 15 fps in that situation at the church. Its a huge difference, and i would be surprised if HT did make a huge difference here. Obviously its about good coding, and there must be a lot of different factors in play besides just threads, cache... hmm.

 

[krumme]

Is that the test with 2133 DDR4? At the time, iirc, it was already known that SL shined much better with faster ram.

Yes it is, but it answers the question about IPC as there is a comparison using the same ram type.
And ddr4 2400 wouldnt have changed the results radical. Notice i wrote 3% and not the 2.7% that was the result for the 2133 ddr4 - to give a more realistic impression of the difference. The HSW wasnt running on 2133 ddr3 either, because the cpu should be run within their official specs.
The point is the last big ipc uplift was with HSW and it was in the order of aprox 10% - as it was here they went from 3-4 wide core thats hardly a surprise, but the general conception that each step leads to either a 5-8% uplift in tick and 10-15% in tock simply doesnt ad upp over the years. Its a perception that have no grounds in the numbers.
Go look at IPC for SB vs SKL. Its a aprox 24% uplift for 4 gens meaning an average of 6% for each gen.

I dont think its bad, because widening the cores more have next to no effect but it will have great cost for efficiency and diesize and these cpu go into servers and mobile where efficiency is king. Its just we cant pretend everything is like the progress was for the first 3 gens of core.

 

[LTC8K6]

Yes it is, but it answers the question about IPC as there is a comparison using the same ram type.
And ddr4 2400 wouldnt have changed the results radical. Notice i wrote 3% and not the 2.7% that was the result for the 2133 ddr4 - to give a more realistic impression of the difference. The HSW wasnt running on 2133 ddr3 either, because the cpu should be run within their official specs.
The point is the last big ipc uplift was with HSW and it was in the order of aprox 10% - as it was here they went from 3-4 wide core thats hardly a surprise, but the general conception that each step leads to either a 5-8% uplift in tick and 10-15% in tock simply doesnt ad upp over the years. Its a perception that have no grounds in the numbers.
Go look at IPC for SB vs SKL. Its a aprox 24% uplift for 4 gens meaning an average of 6% for each gen.

I dont think its bad, because widening the cores more have next to no effect but it will have great cost for efficiency and diesize and these cpu go into servers and mobile where efficiency is king. Its just we cant pretend everything is like the progress was for the first 3 gens of core.

You shouldn't really count Broadwell, though. As a regular desktop chip it was rare, and it had edram to help it out, and there were only two socketed desktop variants. A single i7 and a single i5.

The last ipc increase should be calculated from Haswell to Skylake, imo.

Desktop Broadwell is an outlier, imo.

 

[pantsaregood]

With any luck, Zen will be a close competitor and light a fire under Intel. If anyone recalls, this kind of complacency in Intel is what let AMD catch up from 1999-2002 and outperform them from 2003-2006.

 

[Insomniator]

You can't massively improve performance forever. Perhaps the only thing they can do at this point is add cores, but there is no reason to with AMD being useless for the last 7 years, plus the fact that hardly anything most people do shows serious improvement with the extra cores anyway.

If you are a so called 'enthusiast' that DEMANDS (lol) 6 cores, well the 5820 has been out for nearly 3 years at this point. Hell, its only $320 at Microcenter - I considered it myself for a while but I don't need to rip movies 5% faster.

I'm also guessing that Zen isn't gonna light crap under Intel. I can't wait to see benchmarks blow us away... but if its a turd it would be business as usual.

 

[cytg111]

You can't massively improve performance forever. Perhaps the only thing they can do at this point is add cores, but there is no reason to with AMD being useless for the last 7 years, plus the fact that hardly anything most people do shows serious improvement with the extra cores anyway.

If you are a so called 'enthusiast' that DEMANDS (lol) 6 cores, well the 5820 has been out for nearly 3 years at this point. Hell, its only $320 at Microcenter - I considered it myself for a while but I don't need to rip movies 5% faster.

I'm also guessing that Zen isn't gonna light crap under Intel. I can't wait to see benchmarks blow us away... but if its a turd it would be business as usual.

I agree with this POV.
If Zen turns out to be competetive then worst case scenario is that Intel just plays along with the pricing game. Best case scenario is they begin cooking again and we get a history repeat with Core and thus shiny new toys once more.
Baseline is, if Zen is "just" competetive, we get the same performance as we have today(and had yesterday, day before, day before that...), just at a lower price point. Nothing really new here. I am hoping for the long game here, be it Itanium III++ or something else.

 

[LTC8K6]

With any luck, Zen will be a close competitor and light a fire under Intel. If anyone recalls, this kind of complacency in Intel is what let AMD catch up from 1999-2002 and outperform them from 2003-2006.

Unfortunately I don't see Zen lighting a fire under Intel. Intel can respond to Zen with pricing and core adjustments.

Zen does not appear to be a leap in ipc for the consumer.

 

[raghu78]

Unfortunately I don't see Zen lighting a fire under Intel. Intel can respond to Zen with pricing and core adjustments.

Zen does not appear to be a leap in ipc for the consumer.

If AMD Zen forces Intel to cut prices on their HEDT and core i7 line don't you think that in itself is a huge victory for us consumers. btw at the very high IPC levels of Skylake its damn hard to come up with significant jumps in IPC. How do you think AMD can overcome the problems Intel has been facing for quite some time. I would be happy if AMD can provide Broadwell IPC and 4.5 Ghz max clocks on air for 24x7 usage. The pricing is key. AMD needs to target the USD 200 - USD 600 range.

 

[scannall]

If AMD Zen forces Intel to cut prices on their HEDT and core i7 line don't you think that in itself is a huge victory for us consumers. btw at the very high IPC levels of Skylake its damn hard to come up with significant jumps in IPC. How do you think AMD can overcome the problems Intel has been facing for quite some time. I would be happy if AMD can provide Broadwell IPC and 4.5 Ghz max clocks on air for 24x7 usage. The pricing is key. AMD needs to target the USD 200 - USD 600 range.

If Zen is an outstanding chip, and AMD triples their market share it won't matter to Intel. It comes down to the numbers. A price war would cost them more bottom line dollars than just ignoring AMD would. And well, that's the bottom line.

 

[LTC8K6]

If AMD Zen forces Intel to cut prices on their HEDT and core i7 line don't you think that in itself is a huge victory for us consumers. btw at the very high IPC levels of Skylake its damn hard to come up with significant jumps in IPC. How do you think AMD can overcome the problems Intel has been facing for quite some time. I would be happy if AMD can provide Broadwell IPC and 4.5 Ghz max clocks on air for 24x7 usage. The pricing is key. AMD needs to target the USD 200 - USD 600 range.

Who said anything about pricing victories?

 

[WhoBeDaPlaya]

Zen does not appear to be a leap in ipc for the consumer.

It'll probably be a nice leap over FX.
Now I like the FX well enough given its price (have a bunch of 8320s still in use), but it sucked that IPC regressed from Phenom II, which was itself just slightly better then Core2 / trailing Nehalem.

 

[Atari2600]

Unfortunately I don't see Zen lighting a fire under Intel. Intel can respond to Zen with pricing and core adjustments.

Possibly.

I do think AMD's CPU designs have traditionally been more risky/innovative than Intel's. I'd see AMD as more likely to move beyond a glass ceiling than Intel.

- Hammer was diametrically the opposite of where Intel were going with P4, yet its success is still fondly remembered.
- Barcelona was badly hobbled by the 65nm process. If they'd been able to launch on a process equivalent of Penryn with Thuban (instead of 65nm), they'd likely have continued their K8 dominance. What Barcelona vs Clovertown/Harpertown made clear was - the memory controller had to be integrated.
- Bulldozer was a mis-step. Risky, innovative, but ultimately wrong.
- Zen is TBD.


We could take a few guesses as to the next big thing in CPUs, 3D designs (and not merely stacking) is most likely the avenue for silicon based chips to take.

 

[krumme]

You shouldn't really count Broadwell, though. As a regular desktop chip it was rare, and it had edram to help it out, and there were only two socketed desktop variants. A single i7 and a single i5.

The last ipc increase should be calculated from Haswell to Skylake, imo.

Desktop Broadwell is an outlier, imo.

Its fine for me to compare hsw to say skl. Personally i would also rate the ipc difference different than what AT did, and give IB and BW a bit less and hsw and skl a bit more, but at the end of the day if you compare ipc of sb to what you buy today meaning kbl its the 24% over like 6 years. And since hsw its more like 7% over the last 3 years. Its clearly stagnating. And since hsw freq is also comming to a halt for the desktop side of things. Within a few years we are also at the wall for the cores number too for normal usage. 8c /16t is extreme and its also hitting a hard wall.

I also bet on some itanium#3 solution. Its the only way to progress from here. And Intel better take the lead as ARM instruction set will put them under serious pressure on desktop, laptops and servers within a handfull of years.

 

[NTMBK]

I also bet on some itanium#3 solution. Its the only way to progress from here. And Intel better take the lead as ARM instruction set will put them under serious pressure on desktop, laptops and servers within a handfull of years.

No way, Itanium was a terrible idea.

 

[PingSpike]

Weird in those threads with the denial of the HT pointlessness. People started saying its important on i3s...but don't i3s have extra cache and clockspeed over pentium dual cores making it the same Apples to Oranges comparison?

 

[PingSpike]

No way, Itanium was a terrible idea.

I remember not knowing much about Itanium, but a lot of people were excited about it back then. I went back and read about a few weeks ago. It was a company wide mass delusion, it was never going to work.