Single threaded performance

[Seminole]

Just wanted to share these graphs I made mostly surrounding the STP of Skylake and AMD processors, but mostly showing the dominance of the i3 in price/performance  enjoy.

 

The i3-6320 looks to be really worth the value dominating the under $200 processor market, while the 6300 seems to lag behind.

The i7 6700k and 4790k obviously would have topped this chart but I cut them out because there wasn't enough space. But you can see that even the Pentium G3258 annihilated all of the Skylake i5's in terms of single threaded performance.

 

Note: these graphs use the lowest data point as the base, so they show the difference of the performance, not the total performance. It's similar to looking at just the top of a chart. Passmark scores are from https://www.cpubenchmark.net/singleThread.html

 

Raw statistics (AMD fanboys look closely)

FX9590   1,720   $229.99
X4 845    1,795   $68.99
i5-6500   1,942   $204.99
i5-6600   2,091   $229.99
i3-6100   2,107   $121.99
i5-6600k 2,127   $239.99
i7-6700   2,137   $314.99
i3-6300   2,159   $153.99
pG3258  2,175    $69.99
i3-6320   2,259   $164.99
i7-6700k 2,328   $359.99
i7-4790k 2,529   $339.99

[harrynowl]

Nice effort. Might want to add some Axis  

[Glenwing]

It only looks like the G3258 "annihilates" the Skylake i5's when you cut off the bottom of the chart like that. It's barely more than a 2% difference. 

[stconquest]

Looks like they are all the same to me.  Maybe the stock clock has something to do with the minor differences?

[Đỗ Đức Huy]

Boo, your chart is bad and you should feels bad. Here is the actual chart with the X axis intact

 

 

 

TLDR does not make that much of a difference 

 

 

[Seminole]

1 hour ago, Glenwing said:

It only looks like the G3258 "annihilates" the Skylake i5's when you cut off the bottom of the chart like that. It's barely more than a 2% difference. 

A 2% difference is a pretty big difference for single threaded performance. That's like saying "oh the new architecture has only an 8% IPC improvement." You have to keep in mind the passmark score is literally going from 0 to 2500+. You're talking about 2% of over 2000. It might not sound like a lot but just in that 2% improvement you could fit half of a Pentium 4. That sounds pitiful but that's half of a Pentium 4 you didn't have.

Of course I'm not going to argue that it's a massive difference, it isn't. And of course it's going to be small percentages because nowadays we all have 2 cores or more (unless someone here has a Pentium 4) so the STP is going to increase a lot slower than it did 10 years ago.

Still matters though. Especially if you're debating between an i3 and an i5 on a tight budget.

1 hour ago, Đỗ Đức Huy said:

Boo, your chart is bad and you should feels bad. Here is the actual chart with the X axis intact

 

 

 

TLDR does not make that much of a difference 

 

 

The LTT forums are not for flame wars. At least don't try to start one ffs. And single thread performance does matter, because not all tasks can efficiently spread workloads, some just can't.

 

 

If the i3 had the same STP as the i5 in these benchmarks then the i3 would have been annihilated (at least a lot worse than it was).

Of course the i5 is going to win though. Multithreaded performance matters a lot too. My point was just to show how much single threaded performance really helps the lower end processors catch up a bit.

[Seminole]

1 hour ago, stconquest said:

Looks like they are all the same to me.  Maybe the stock clock has something to do with the minor differences?

The stock clock speed does in fact make a difference, but not as big as the actual STP. Around the higher end CPUs they are all basically the same clockspeed though, maybe 0.4GHz off at the most, and you can see even the 3.2GHz pentium on the old architecture still pulls ahead.

[stconquest]

1 hour ago, Seminole said:

The stock clock speed does in fact make a difference, but not as big as the actual STP. Around the higher end CPUs they are all basically the same clockspeed though, maybe 0.4GHz off at the most, and you can see even the 3.2GHz pentium on the old architecture still pulls ahead.

That's the one that threw me off a bit.  3.2GHz and no boost clock.

[Seminole]

3 minutes ago, stconquest said:

That's the one that threw me off a bit.  3.2GHz and no boost clock.

Generally for CPUs with more than 1 core the clock speed doesn't affect the STP that much.

[Sintezza]

If you compair a core i3 with a core i5 from the same architecture clocked at the same clock speed.

Then they will have roughly the same single thread performance yes.

Thats ofc logical, because they are basicly all the same thing.

A i3 is basicly a miss production of an i7 with 2 defective physical cores disabled..

[Prysin]

lol, that test got to be seriously bugged if a 4790k can beat a 6700k. Both have the same frequency, but the IPC of the 6700k alongside DDR4 means that it is way faster.

G3258 being that high means it is overclocked. It isnt a very fast CPU out of the box, so it is definetively OCd to some extent.

 

Honestly, the only scores i think is valid is the AMD scores. They should be lower then intel by a wide margin, which they are.

 

However the Intel scores are so broken up and all over the place, i have little to no real faith in that benchmark. Seems inconsistent with every other benchmark there is

[Lays]

Just for reference I wanted to do a test at 5ghz to see how it compares, now I want to see if I can get 3000

 

 

 

[Seminole]

36 minutes ago, Prysin said:

lol, that test got to be seriously bugged if a 4790k can beat a 6700k. Both have the same frequency, but the IPC of the 6700k alongside DDR4 means that it is way faster.

G3258 being that high means it is overclocked. It isnt a very fast CPU out of the box, so it is definetively OCd to some extent.

 

Honestly, the only scores i think is valid is the AMD scores. They should be lower then intel by a wide margin, which they are.

 

However the Intel scores are so broken up and all over the place, i have little to no real faith in that benchmark. Seems inconsistent with every other benchmark there is

They have multiple sources for every single score they have, and there is strictly no overclocking. And the 4790k still does beat the 6700k in a lot of benchmarks

https://www.pugetsystems.com/labs/articles/Haswell-vs-Skylake-S-i7-4790K-vs-i7-6700K-641/#CPUPerformance-UnigineHeavenPro4_0

 

And the G3258 is actually pretty damn fast for a Pentium. Just a bit outdated that's all.

[Prysin]

4 hours ago, Seminole said:

They have multiple sources for every single score they have, and there is strictly no overclocking. And the 4790k still does beat the 6700k in a lot of benchmarks

https://www.pugetsystems.com/labs/articles/Haswell-vs-Skylake-S-i7-4790K-vs-i7-6700K-641/#CPUPerformance-UnigineHeavenPro4_0

 

And the G3258 is actually pretty damn fast for a Pentium. Just a bit outdated that's all.

G3258 is haswell. it is clocked at 3.2 GHz. It does not have Turbo. It has 3MB of L2 Cache and 0MB of L3 cache.

 

Do you understand how retarded you are if you think that a Haswell CPU at 3.2GHz can score higher then a Haswell CPU at 4GHz? or even a Skylake CPU at 3.2GHz?

 

Skylake is proven to be 7-10% faster then haswell in every single way. This is a fact, it is proven by every respectable review site there is. So explain to me how a CPU with 7-10% lower IPC, same clock speed, less cache, no L3 cache, slower system memory can beat a faster CPU.

Newsflash! IT CANT. It is physically impossible to overcome that deficient without increasing clock speed.

This is how AMD CPUs stay even remotely relevant atm. Low IPC but high clock speed. The higher clocks negate the IPC deficient.

 

If you think Pugetsystems are "the best source" to provide, you are wrong.

It is a known fact that companies that specialize in building PCs uses whatever software they latch onto, then they STAY with said software. Even if it is a bit dated and or mis-representing. Because after all, aslong as they can refer to a score for your system and a "reference score" using the same test, regardless if it is flawed or not, then they wont care. Not, one, bit.

 

Its like a PC building/retail company here in Norway, they use Intel Burn test to test if parts are malfunctioning. Yeah, way to go lol.... You wont even know if a GPU is artifacting if you use that software. All you will know is whether it can give you a "OK" signal. Which it can do, while still being broken.

 

Either way, you are banned now, it seems. Which is for the better if you are posting malicious bullshit like this.

[harrynowl]

3 hours ago, Prysin said:

G3258 is haswell. it is clocked at 3.2 GHz. It does not have Turbo. It has 3MB of L2 Cache and 0MB of L3 cache.

G3258 is 3MB L3 FYI. Intel has't made L2 cache sizes larger than 256K per core in some time.

 

As a haswell die it will be 3MB L3. 256K L2 (per core) 32K of L1 instruction (per core) and 32K of L1 decode (per core).

[Prysin]

1 hour ago, harrynowl said:

G3258 is 3MB L3 FYI. Intel has't made L2 cache sizes larger than 256K per core in some time.

 

As a haswell die it will be 3MB L3. 256K L2 (per core) 32K of L1 instruction (per core) and 32K of L1 decode (per core).

yeah, i saw that later. I was reading off intels own data sheets, but they only state total size, not individual cache size (god knows why)