Intel processor architectures tested clock-to-clock in gaming - results all over the place

[That Norwegian Guy]

On 7.1.2017 at 4:04 AM, Energycore said:

Ivy is just die shrunk Sandy though. It was a tock

Yet the IPC difference between Ivy and Sandy was 6%, but for Haswell vs. Ivy it was 7.04%

 

However Haswell lost 12% avg. overclocking headroom compared to Ivy while Ivy had lost almost nothing (3%~) compared to Sandy.

 

Seems tick vs. tock in Intel terms is the difference between wiping your ass with your left or your right hand

[Energycore]

22 minutes ago, That Norwegian Guy said:

Yet the IPC difference between Ivy and Sandy was 6%, but for Haswell vs. Ivy it was 7.04%

 

However Haswell lost 12% avg. overclocking headroom compared to Ivy while Ivy had lost almost nothing (3%~) compared to Sandy.

 

Seems tick vs. tock in Intel terms is the difference between wiping your ass with your left or your right hand

Well Haswell may have lost some overclocking headroom but Haswell Refresh gained it back, then Skylake lost a little bit and Skylake Refresh gained it back and then some.

 

Yes it seems like if you want high clocks you should buy every other generation

[crystal6tak]

Sweet, my 4790k on 4.4 should last me a good few years.

 

I'll probably delid to OC it further in the future once AMD starts competing again.

 

Seems like delidding is super easy nowadays, there are tools and even re-lid tools as well.

[straight_stewie]

On 1/6/2017 at 4:50 PM, Levis95l said:

The testing methology in more detail for example.

I would like to know way more about their testing methodology. These results are different than one would expect. 

[VerticalDiscussions]

What..... in..... the..... actual..... ah screw it!!!

 

[Orangeator]

41 minutes ago, VerticalDiscussions said:

-Snip-

That is the single best post I have ever seen on this forum.

[potoooooooo]

On 1/6/2017 at 6:09 PM, RGProductions said:

IDK why but I always knew Ivy was a really strong architecture 

Everything in this chart is based directly on Sandy Bridge. Sandy bridge is the base current intel CPUs are built off of

[Luke]

Guys should i upgrade my 3770K?

 

Hah jk intel you won't get my money until you deserve it.

[LAwLz]

On 2017-01-07 at 5:21 AM, Zodiark1593 said:

Was Sandy that big of a jump over Westmere? I figured the jump to Core (2) was far larger than from Westmere to Sandy. Probably closer to what Nehalem was to Core. Or was there something I missed from the launch reviews? 

The thing with Sandy Bridge was not only the massive IPC increase, but also the far higher overclocks you could get, and on top of that the GPU was pretty good too (and for video encoding it was the best HWA encoding ever).

 

The 2600K was performing as well as the 980X, which was a 1000 dollar 6-core CPU (on a much more expensive platform), plus it overclocked far better (once both were overclocked, the 2600K came out on top).

Imagine if Intel released a 300 dollar i7 which matched their 12 core chip today (in programs which actually used all threads), and overclocked better, and had extra features like an onboard GPU.

 

When you compared mainstream platform vs mainstream platform, Sandy Bridge was something like 30-40% faster in your average application, and overclocked probably close to 100% higher than the old chips (so instead of only managing a 500MHz overclock, you could now get 1GHz overclock with ease).

 

It was such a massive upgrade it basically killed the X58 platform.

 

 

It might not be that impressive to you because you were comparing X58 to Sandy Bridge, but what you have to remember is that those were meant for two different markets. It's just that Sandy Bridge was so good it even outperformed the enthusiast stuff which was ~4 times the price.

 

 

My friend bought an i7-920 based system just a few months before Sandy Bridge came out. I told him to wait, and to not spend so much money (something like 2000 dollars) on his computer, but he didn't listen. He wanted to buy such an expensive computer because it would last him a long time. Then like 4 months later I bought a Sandy Bridge CPU, a 6850 GPU and then my ~800 dollar computer ran circles around his.

[Thony]

4+ years and all they have done (according to benchmarks) is lowering heat and power consumption.

 

Not sure they struggle with progress (more research needed, someone with new ideas etc.) or they are just not motivated.... 

[Cyborgsmith]

An argument can be made that performance has plateaued since sandy bridge. Unlike GPUs cpus haven't improved significantly. 

[Levis95l]

12 hours ago, straight_stewie said:

I would like to know way more about their testing methodology. These results are different than one would expect. 

There's nothing really really detailed about the tests themselves but gather here everything I can get.

 

-Testing platforms-

Sandy- and Ivy Bridge: Asrock Z77 Pro3 -motherboard and 8GB of DDR3 (run at 2133MHz for Sandy and 2400MHz for Ivy).

Haswell: Asus Z97-A USB 3.1 -motherboard and 8GB of DDR3 @ 2400MHz.

Sky- and Kaby Lake: Asus Z270 ROG Maximus IX Formula -motherboard and 8GB of DDR4 @ 2400MHz.

- Open test bench

- Nvidia GTX 1080 reference card with 376.33 drivers (the newest at the time of testing)

- 1080p monitor

- 64 bit Windows 10 Pro installed on 120GB Corsair Force GT SSD disc

- Games installed on Samsung 960 Pro

- Silverstone 750W Strider Gold power supply

 

-Benchmarks-

Cinebench R15: Run with all cores enabled and with only one core. Both results in the graph.

Blender 2.78a: RyzenGraphic_27 file published by AMD was rendered.

Geekbench 4: Results include scores for all cores and a single core.

X265 HEVC Benchmark 2.0.0

3DMark Fire Strike physics test

JetStream 1.1 Javascript Benchmark Suite: Run with Google Chrome version 55.0.2883.87 m

Octane 2.0 Javascript Benchmark: Run with Google Chrome version 55.0.2883.87 m

AIDA64 Memory Benchmark

 

Unfortunately that's about it. Everything else can be seen from the graphs.

 

I guess the most interesting part is the description below BF1 benchmarks but I pretty much refered that already. Here's a translation word by word:

"Battlefield 1 was tested on DX12 API and fps measured with OCAT software. The results describe well the challenges faced when testing the processor architectures in games.

 

The clearly best performing processor in Battle Field 1 was Ivy Bridge, even after several re-runs. For example, with Kaby Lake and Skylake processors the test hitched in a couple of places and the fps dropped distinctly, but with Ivy Bridge the problem didn't appear as severely, even though it used an old Z77 motherboard and DDR3 memory."

[BingoFishy]

On 1/6/2017 at 11:22 PM, Energycore said:

I was born in 2011 when Sandy Bridge released so

Youngest person the forum!

[Gibbons]

Haven't read the whole thread so this might have already been mentioned. The DDR3 ram and DDR4 ram modules are running at different timings so that's why things look mostly normal for synthetic/pro benchmarks but games are all over the place.

[Nena Trinity]

I would like to see more CPUs added to the test!

[TidaLWaveZ]

I wonder where Haswell-E would sit if they turned off two cores.

[ZothOmmog]

1 hour ago, Levis95l said:

The clearly best performing processor in Battle Field 1 was Ivy Bridge, even after several re-runs. For example, with Kaby Lake and Skylake processors the test hitched in a couple of places and the fps dropped distinctly, but with Ivy Bridge the problem didn't appear as severely, even though it used an old Z77 motherboard and DDR3 memory.

Damn, and I was hoping to make a move for the newer CPU's and MoBo's as an upgrade.

 

But um I have a brother who wants my system because I've talked about upgrading for the past 6 months and I was really hoping to go for SL/KL but as I've seen here it doesn't really seem to benefit me at all, if I were to take SL/KL out of the equation, to which platform would it be the most smart to upgrade to in that case, LGA2011-v3? Thoughts anyone?

 

PS he doesn't want to go through the trouble of building a system.

[Fetzie]

22 minutes ago, ZothOmmog said:

Damn, and I was hoping to make a move for the newer CPU's and MoBo's as an upgrade.

 

But um I have a brother who wants my system because I've talked about upgrading for the past 6 months and I was really hoping to go for SL/KL but as I've seen here it doesn't really seem to benefit me at all, if I were to take SL/KL out of the equation, to which platform would it be the most smart to upgrade to in that case, LGA2011-v3? Thoughts anyone?

 

PS he doesn't want to go through the trouble of building a system.

Wait for Ryzen, and either 5820k or 6800k if Ryzen falls on it's face.

[SpaceGhostC2C]

On 1/6/2017 at 11:50 PM, Levis95l said:

All processors were run at 4.0 GHz and with 2400 MHz memory speed (except Sandybridge which only supports 2133 MHz).

 

24 minutes ago, Gibbons said:

Haven't read the whole thread so this might have already been mentioned. The DDR3 ram and DDR4 ram modules are running at different timings so that's why things look mostly normal for synthetic/pro benchmarks but games are all over the place.

This is a potentially important point: given the differences in typical timings for DDR3 and DDR4, 2400MHz across the board most likely means effectively faster RAM for the DDR3 CPUs (Sandy to Haswell)

On 1/7/2017 at 4:04 AM, Energycore said:

Ivy is just die shrunk Sandy though. It was a tock

It also introduced tri-gate transistors.

On 1/7/2017 at 4:00 PM, Prysin said:

Then again, Hyperthreading makes any product way more relevant. Even if IPC isnt a big factor, HT is

On 1/7/2017 at 9:03 PM, Zodiark1593 said:

 One could say that Hyperthreading is an integral part of the architecture to get the most out of it, and not just a feature. 

I would dispute that. Although maybe, for gaming, it is.

 

2 hours ago, Thony said:

4+ years and all they have done (according to benchmarks) is lowering heat and power consumption.

 

Not sure they struggle with progress (more research needed, someone with new ideas etc.) or they are just not motivated.... 

And what did Intel do in the 4 years before Sandy Bridge? How long did Core 2 last? I don't understand why the mere passing of time is supposed to mean anything. Incremental, marginal improvements on an existing architecture are to be expected, but a whole new architecture achieving performance leap? No amount of money and time can ensure it, much less control its timing.

And even if it was possible, who needs such a boost every 6 months? It's seems to me that 5-year old CPUs still working fine is a good thing for us (don't we start crying "planned obsolescence" when the opposite happens?). If anything, it's a problem for Intel when trying to sell us more CPUs. It is in general a good thing not to have to buy a replacement (unless, of course, consumerism).

 

Now AMD is apparently catching up with this lot and plans to keep the same socket for at least 4 years: how many revolutionary improvements can you achieve without changing the socket? Bulldozer to Excavator was an impressive overall IPC increased, when compared to Intel's past 4 years, but it required FM1->FM2-FM2+ (I guess the reason they never got past Piledriver in AM3+ was the inability to keep the same socket - but just guessing). So, AMD's bet is to remain competitive 4 more years without stellar upgrades in IPC, which requires Intel to spend 4 years more in the same ballpark...

 

[LAwLz]

8 hours ago, Gibbons said:

Haven't read the whole thread so this might have already been mentioned. The DDR3 ram and DDR4 ram modules are running at different timings so that's why things look mostly normal for synthetic/pro benchmarks but games are all over the place.

8 hours ago, SpaceGhostC2C said:

This is a potentially important point: given the differences in typical timings for DDR3 and DDR4, 2400MHz across the board most likely means effectively faster RAM for the DDR3 CPUs (Sandy to Haswell)

Source?

I don't see them mentioning timings in their page.

 

8 hours ago, SpaceGhostC2C said:

And what did Intel do in the 4 years before Sandy Bridge? How long did Core 2 last? I don't understand why the mere passing of time is supposed to mean anything. Incremental, marginal improvements on an existing architecture are to be expected, but a whole new architecture achieving performance leap? No amount of money and time can ensure it, much less control its timing.

Intel is holding back though.

They could easily make it like this:

Pentium - dual core with hyperthreading

i3 - quad core

i5 - quad core with hyperthreading

i7 - 6-core with hyperthreading

 

And then on their enthusiast platforms starts at 6-cores with extra PCIe lanes, quad channel memory, and even more cores at the higher end.

But they have 0 reason for doing this. Why give customers more when they are forced to buy whatever you sell them anyway? No reason to improve. The only threat to Intel is ARM, and ARM's strengths are:

1) Efficiency

2) Heat output

3) Size

 

And wouldn't you know it, the first two are mostly what we have been getting ever since AMD couldn't compete anymore.

[Mira Yurizaki]

I was reading this article (http://www.realworldtech.com/bulldozer/2/) on a review of AMD's Bulldozer and an overall theme of it was what AMD has to do is change the rules of the game. When they released the AMD64 architecture, they changed the rules of the game. Intel was too busy with damage control on IA-64 and now all the server customers were excited they could get 64-bit x86. The author posits that AMD wanted to change the game again by battling on the core count front, considering that AMD could not match the might of Intel's manufacturing capabilities.  But we know they failed there. So AMD is going to take a crack at it again.

 

However I'm not so hopeful that more cores is going to be the computing revolution we're hoping for. What's the biggest use case for computer usage? Going on the internet and/or consuming media. You can do that on an ARM. It doesn't take much power to meet the use cases for most people.

 

The other thing is AMD really needs to reach out. Even your grandmother probably knows what Intel is. Intel and NVIDIA both have a strong marketing team. It doesn't matter how good your product is if you can't sell it to other people. (and they should really drop the aggressive act, it makes them sound like the in-your-face N-Gage ads)

[dfsdfgfkjsefoiqzemnd]

Well, at least we now know the answer to "should I upgrade from my 6600K to a 7700K?".

 

The answer is : Get an Ivy Bridge CPU

[Mozenator]

On 1/6/2017 at 3:54 PM, cj09beira said:

now we will see the core race,

 

You mean the race AMD won before it even started? AMD doubled down on having more cores for the main mainstream and it almost put them in an early grave. I doubt we will see any quantum jumps in core count outside of enthusiast platforms until developers better optimize gaming and productivity for more cores. Software is going to to have to put the pressure on hardware before CPU architects will prioritize core count again.  

[Scruffy90]

On 1/6/2017 at 8:35 PM, crisro996 said:

*looks at 3770k*

 

*Looking at my 3770k*

[Scruffy90]

I really dont understand how the 3770k is pushing ahead that far, unless my ass lost the silicon lottery with my cpu too.