What makes a modern CPU running at the same core speed faster

>>61399631
>I'll probably just wait
If you don't have a reason to upgrade, you probably should just wait.

>>61399644
I'm starting to feel the age of my CPU but it's not so awful that I can't do anything

>>61399631
Modern Intel CPUs have a four times as large IPC.

https://en.wikipedia.org/wiki/Instructions_per_cycle#FLOPs_per_cycle_for_various_microarchitectures

>>61399651
Just wait then, you don't have an upgrade path so any worthwhile change isn't going to be "budget"

Try >>>/g/pcbg

>>61399631
bloomfield still kicks ass

though if you want to squeeze 1366 to the max get a x5675, that's the last cpu (i believe) made for that socket. that 6 core/12 threaded can overclock to 4.5ghz easily. just get a decent cooler if you haven't (something like a scythe will do)

>>61399681
oh and to mention it's only $50 on ebay, you're lucky to have a 1366 board because decent ones go for over $100

>>61399631
Best budget upgrade: $25 Xeon X5660 on ebay.
6 cores, 12 threads, overclock it to 4.2+ and you're good to go.
4.5+ is possible with great cooling.

File: y9tvy3.png (116KB, 415x411px)

116KB, 415x411px

>>61399692
>>61399681
Thanks man, I'm using a noctua d15 so I'm doing alright on the cooling front

this looks really promising, wonder if this dude's OC is stable

even if I can only hit 4-4.2GHz that'd be great

>>61399722
Main downsides are power draw and heat when overclocked, but performance is more important to me
General consensus is not to go above 1.40v for a 24/7 oc, but it's not like replacements are expensive either

>>61399722
just for some quick reference, x5675 @4.5ghz gets about 1000 in cinebench. that's all i remember =(

>>61399722
>wonder if this dude's OC is stable
>2 cores left enabled
>no HT
lolno

>>61399631
Because of faster architecture. Older processors have larger dies resulting in less power efficiency and higher temperatures. Don't bother with the Xeon CPUs even though they are very good for ~20 dollars. The catch is you have to get a motherboard that supports those CPUs and they are generally >100 dollars. For such an old CPU it's hard to justify in my opinion. I'd recommend the Ryzen 1600 for it's versatility in pretty much everything. Its price is good and would get you on the AM4 platform which will have many upgrades down the road. So if you game, stream or encode a lot then the Ryzen 1600 can really flex its muscles. It has more cores, more threads, better power efficiency etc. than your current CPU. If you don't need something that strong then you can get the Ryzen 1400 or maybe the Ryzen 1300 which comes out in 10 days. I think it will cost ~100 USD.

>>61399719
How fast would I be able to encode h265 1080?
How many fps?
I need to build an encoder rig and I want something cheap

>>61399792
that picture is actually my CPU, I have an old motherboard

>>61399792
>being unable to read
op already has an x58 motherboard

>>61399796
Ryzen will annihilate the Xeon in multithreaded performance. Those CPUs are old and while you probably will see a performance boost from what you are currently using, it's hard to justify moving to a dead socket. AM4 is the way to go.

>>61399826
shill pls

File: Untitled.png (1KB, 367x31px) Image search: [Google]

1KB, 367x31px

>>61399826
>>61399826
>>61399826

fuck off

>>61399631
clock speed is a misleading number.

Processing power has a lot to do with cpu architecture and transistor density.

All clock speed tells you is how many electrical signals per second it can handle.

If you have two cpu's, both operating at the same clock speed they can be vastly different.

example:
cpu1@3GHZ - 20 cycles to execute an instuction A
cpu2@3GHZ - 5 cycles to execute an instruction A

both run at 3GHZ but cpu 2 will be able to perform instuction A 4 times quicker than cpu 1

its all about architecture

>>61399841
Encode something in h265 1080 and post stats please

>>61399667
>>61399853
Thanks for the response fellas

>>61399854
diy

>>61399841
I thought we were discussing the 6 core Xeon. My bad.

File: GAAFET.png (630KB, 999x532px)

630KB, 999x532px

>>61399667
wrong
Its not even close to 4x the IPC of nehalem it might be 1.5x @ the same clock speed
Raw IPC per cycle is irrelevant to cpu performance as not all instructions are equal or even similar in design. Pipeline length also has a huge impact on the way IPC works
Basically a long ass pipeline is bad but it can allow much higher clocks. This is why P4 chips are terrible even though some are clocked to nearly 4ghz. Flip side is AMDs K8 which has a much more efficient pipeline but it limits the clock speed of the cpu so they ran at mostly half ghz of the pentium of the day but overall performed better. Its a balancing act in that regard. Fat instruction windows mean you need more cache and more transistors over all.
Also ram speed and thus bus speed has a much larger impact on modern CPUs than it did in the past.
CPUs have not progressed much in the past 10 years at least compared to the 98-08 timeframe. Mainly due to clock speeds being limited by silicon itself so we are chasing small gains on the IPC side.
Ryzen has VERY good ipc when paired with fast ram. Intels new i9 CPUs are a regression from what we saw on broadwell e this is evident by the smaller L3 cache size yet much larger L2 cache.


https://www.youtube.com/watch?v=8caCYb0G0ec

>>61399722
Wish we had cheaper 1366 motherboards. All those xeons but the perfect mobos are expensive as fuck.

Makes the whole thing kinda quite expensive.

>>61399841
>>61399834
>"Don't buy new things! Go back several generations and buy something that was obsolate four years ago!"

>>61399719
>>61399681
Now I just have to decide between x5675 and x5660

do I want to save $25? I could eat like 2 mcdonalds meals with that

>>61399916
And it doesn't help that people keep fucking breaking the good ones leaving nothing but cheap chinkshit boards behind.

>>61399930
This is a thread about budget CPUs and upgrading well-dated hardware

talking about modern hardware is pointless

>>61399937
x5675 overclocks better and is newer

>>61399943
i actually have a classified x58 in my closet with no use.

>>61399901
>Raw IPC per cycle
opinion discarded

>>61399885
But we are. That's a dual socket X58 system posted.

Anyways I've upgraded from an i7 950 to a X5675 too. I'm getting 1k Cinebench with my stable OC and generally don't limit my 1070@1440p. Amazing for an 8 year old platform.
Lots of fun to OC these chips too. Just be careful with the vcore, the consensus with these Xeons is that they degrade heavily above 1.35V.

File: 990X.png (226KB, 410x393px) Image search: [Google]

226KB, 410x393px

>>61399722
>>61399681
>>61399841
These are basically the same exact chip as the i7 extreme 980x/990x

one of those xeons @ 4.5ghz should peform as good or better than any modern locked cpu sold today

>>61399963
That's usually the case. My X5660 goes to 4.5GHz, but it's so hot i back it off to 4.3.
990 in Cinebench R15 tho

File: AMDramspeed.png (1MB, 1266x669px)

1MB, 1266x669px

>>61399968
The numbers linked in the wiki is theory maximum IPC which is pretty much useless as no instruction set is as basic as those bitch ass numbers state. Average instructions could be 1/10th of what is listed as the theory max.
That wiki says bulldoze and ryzen have the same IPC and skylake is double ryzen?
Even MIPS is nearly irrelevantt to determine cpu work output for modern instruction sets

>>61399667
>>61399901

The wikipedia link says floating point instructions per cycle. Given the recent trend of APUs and improved on board graphics 4x doesn't seem that far fetched.

>>61400057
The numbers linked in the article deal only with floating point. You didn't even fucking read the header did you? It also says per two cores for bulldozer.

File: cinebench.png (1KB, 360x25px)

1KB, 360x25px

>>61399971
>>61400002
my cinebench stats for anyone interested

File: i9way.png (36KB, 756x719px)

36KB, 756x719px

>>61400096
Yes I know but what I am saying is those numbers are the theory max per clock cycle. Actual instructions are much more complex and will never represent the numbers stated.
Modern CPUs do not run 4x as many instructions per cycle compared to older CPUs. In theory they might but in practice is not even close. All of the chokes add up to a much slower process. We are improving each area but it takes time. You cant simply double the floating point and then double the cpu output. Fetch rates have to be doubled as does write rates everything thus is very hard to actually improve the IPC and has slowed down a ton in the past 10 years.

>>61400154
So you can get nearly double the performance for little money, just do it m8
there's a big thread on overclock.net on x58 overclocking that has been very helpful in getting that stable 1k cinebench score

>>61400240
Yeah I'm basically convinced at this point

>>61399796
encoding locally is a meme, you will be many thousands of episodes in before the box you build pays for itself vs massive cloud parallelism
and in the meantime it will be far slower