Kunter

About all those "intel is in trouble" threads...

Finnish tech news site io-tech.fi has got some weird results testing the new Core i7-7700K against Intels previous processors. They tested Kaby Lake against Skylake clock-to-clock like many other news sites, but included also Sandybridge, Ivy Bridge and Haswell in the testing. All processors were run at 4.0 GHz and with 2400 MHz memory speed (except Sandybridge which only supports 2133 MHz).

In synthetic benchmarks and rendering the results were as expected: Kaby Lake and Skylake scored even but otherwise performance had improved slightly by every generation. Kaby Lake beat Sandybridge by around 25-30% depending on the test.

When tested in games with GTX 1080 at 1080p the results don't seem to make any sense however. The frame rates were measured with OCAT. In Battlefield 1 (DX12) Ivy Bridge beat all the others by 12 fps despite several re-runs. With Kaby Lake and Skylake the test hitched in couple of places resulting in noticeable drop in fps. Ivy Bridge suffered from the same behavior, but not in the same level.
 
In Gears of war: Warhammer (DX12) and Civilization VI AI Benchmark (DX12) Kaby Lake and Skylake beat Ivy Bridge but Haswell rose to the top.
 
In GTA V (DX11) Haswell and Ivy Bridge were on top of Kaby Lake and Skylake.
 
Take this how you will. At least it seems that the IPC improvements don't really transfer to games. A small increase in clock speed would lift any one of the processors to the top.
 
Tell me if you liked me to translate more of the article to English. The testing methology in more detail for example.
 
Source and all the benchmarks for you to see: https://www.io-tech.fi/artikkelit/intel-core-i7-7700k-core-i5-7600k-kaby-lake/

That's a pretty big jump in logic, I think.
 
Skylake hit 7025 MHz in a similar attempt here: http://www.pcgamer.com/the-world-record-skylake-overclock-is-an-insane-7ghz/
And 5 GHz is practically unheard of on Skylake, let alone 6 GHz.
 
Edit: It's been done with Ivy Bridge and Haswell, too.
https://www.pcper.com/news/Processors/Overclocker-Pushes-Intel-Haswell-Core-i7-4770K-CPU-Beyond-7GHz
https://www.techpowerup.com/forums/threads/ivy-bridge-7-ghz-rips-through-pi.163114/

AsRock, the last bastion of acceptable-looking Intel motherboards have fallen.
 
Time of death 04:45 on the 21st of december.
 
Good thing I'm going AMD next.

Do you really need to make multiple threads about how the 480 beats the 1060 in the same type of benchmarks (to be more precise, DX12 games)?
 
Also, that title is really biased.
The 1060 is getting 42FPS minimum, and 61 FPS average.
The RX 480 is getting 35FPS minimum, and 65 FPS average.
(1080p numbers)
 
The 1440p results are more in favor of the RX480.
The 1060 is getting 22FPS minimum, and 44 FPS average.
The RX480 is getting 26FPS minimum, and 47FPS average.
 
 
So at 1080p I'd say the 1060 wins. I would gladly trade 4 FPS (which are still above 60) for 7 extra minimum FPS.
The RX480 clearly wins at 1440p.
 
Both cards are unplayable at 4K so why even bother including that (that's by the way where the clickbate "16% faster" number comes from)?
Woho! The AMD card runs the game at 26 FPS instead of 23! Suck on that Nvidia! Oh, your minimum frame rate is 20% higher at 1080p? Well that doesn't count!
Something you also have to remember is that Nvidia cards sometimes performs worse in DX12 mode, so you need the DX11 numbers to get the whole picture.

yeah this is bs if you read the article they used a thermalright 6 cooler before delidding then they used a kraken x62 after delidding

Naaa, 2x math means 2 16bit floating point operations at once, this is why they have 25TF. And what 2 GPU solution should run below 300W? This is big Vega.

i dont think colorful cares, this is their ridiculous cards, for their ridiculous customers
if you have that mobo, you arent gonna get something from colorful
 
mind you that colorful is the manufacturer that gave the world this GTX680:

 
and this completely passively cooled GTX680:

 
colorful doesnt think practical, they think sheer awesomeness.

Because they would have been stuck on 28nm until at lest 2025. Its not that easy to just jump to a way lower node

I was right.

I want it,  but i can't pay for it

i just wanna mention this: that website has so many ads you have to search for the article itself..

"yeah, like play on my GameBoy/GBA/DS/PSP! Or watch some cartoons! Life was so different back than!"
Give a kid a mobile phone to play some Angry birds or watch some YouTube, same difference.
Or he was using his parent's phone? 
 

I personally find that a bit unreasonable. If Intel's performance could so easily be matched; IBM, Sun, Oracle, and ARM would have done it ages ago and swept up the profits for doing it.
 
Intel is dumping billions into this. No, Intel is making massive improvements and hoping software starts catching up. IPC has improved by 8x since the Pentium III days, but real world performance is limited by execution constraints. It's not limited by data or instruction parallelism. That's a testament to bad programming, not bad hardware design.

And I bet we're gonna keep them for quite some time.

I can answer that: Nvidia has more cache, and no matter what IC you optimize your software for, you focus on cache access first. That determines data layout, optimal loop unroll factors, and more. AMD can have the theoretical advantage on paper, but if you waste cycles constantly pulling new data from VRAM in a way not hidden by prefetches and look-ahead buffers, you lose real world performance.
 
It's literally rule #1 in performance tuning. And that's why even for medium N bubble sort will beat quicksort. It's only when you start having to stream a ton of data in and out of cache that quicksort's asymptotic runtime advantage matters.

let's see how much more can we back on this junk: https://www.techpowerup.com/reviews/CoolerMaster/Masterliquid_Maker_92/6.html
 
test system
CPU: Intel Core i7-6700K @ 4.2 GHz & 4.6 GHz OC (Skylake) mobo: MSI Z170A Gaming M7
 
oh my god! even the 1/2 decade old Hyper 212 EVO stayed in the game  what the shit, CM
 
---
 
TPU staff:
 

I've already commented on Youtube, but I think this is a better place for honest feedback.
 
Sicereley, honestly, positively and kindly:
Having available one of the most interesting products in recent times, you have managed to make the laziest and less interesting video of all times. Usually I love your videos, but I can't this one. Really disappointing, and truly surprised, especially by the laziness/rush/apathy shown.
"Quadro M6000 video: check. NEXT!
You put way more effort on the giant pad for god sake, and I liked that video, at least was really fun to watch.

God I hate RGB so much...

And Intel said its Xeon Phi outperformed Nvidia's Maxwell on AI deep learning only to have Nvidia come out and prove they used an outdated (though stable) version. AlexNet is so buggy though I would never use it.
 
AMD is in all likelihood lying.

Pretty sure it had 2, too. Maybe the high core count Xeons are picky?

no, Patricks "order of bias" goes like this
Intel > NvIdIa > IBM/ARM > VIA/Matrox > AMD
 

In the UK, it's £260 for a Sapphire RX 480.
Or £195 for a EVGA GTX 970.

Guru3D did something: https://www.guru3d.com/news-story/amd-zen-engineering-sample-aos-further-analysis.html
they took a  i7 5960X Haswell-E (because it matches the core count) and downclocked it to match the ZEN engi sample, put a RX480 and ran Ashes at the same settings - it does not look good for ZEN
 
example:

Because if AMD can't even match previous generation technologies, it won't make any headway in datacenters, a place it really needs to for financial security.
 
Also, I wouldn't bet their architectures are so different.