Jump to content

IPC vs clock speed

Hey guys, can you tell me what programs are ipc dependant and which are clock speed dependant and please, WHY?

Please quote me so that I know that you have replied unless it is my own topic.

Link to comment
Share on other sites

Link to post
Share on other sites

2 minutes ago, Ezio Auditore said:

Hey guys, can you tell me what programs are ipc dependant and which are clock speed dependant and please, WHY?

Every program. Because thats what they use to process info

He who asks is stupid for 5 minutes. He who does not ask, remains stupid. -Chinese proverb. 

Those who know much are aware that they know little. - Slick roasting me

Spoiler

AXIOM

CPU- Intel i5-6500 GPU- EVGA 1060 6GB Motherboard- Gigabyte GA-H170-D3H RAM- 8GB HyperX DDR4-2133 PSU- EVGA GQ 650w HDD- OEM 750GB Seagate Case- NZXT S340 Mouse- Logitech Gaming g402 Keyboard-  Azio MGK1 Headset- HyperX Cloud Core

Offical first poster LTT V2.0

 

Link to comment
Share on other sites

Link to post
Share on other sites

2 minutes ago, Ezio Auditore said:

Hey guys, can you tell me what programs are ipc dependant and which are clock speed dependant and please, WHY?

No, because IPC is architecture dependent and clock speed affecting performance depends on the IPC. Software has nothing to do with it.

 

The question you should be asking is "what programs are single core performance dependent and multi-core performance dependent?" In which case it depends on if the program is I/O bound or CPU bound. That is they're waiting on something to happen or they aren't.

Link to comment
Share on other sites

Link to post
Share on other sites

What you're likely thinking about is the "Cores vs Clocks" debate. Otherwise IPC just refers to how much a CPU can get done with the same clock speed.

CPU: I5 4590 Motherboard: ASROCK H97 Pro4 Ram: XPG 16gb v2.0 4x4 kit  GPU: Gigabyte GTX 970 PSU: EVGA 550w Supernova G2 Storage: 128 gb Sandisk SSD + 525gb Mx300 SSD Cooling: Be Quiet! Shadow Rock LP Case: Zalman T2 Sound: Logitech Z506 5.1 Mouse: Razer Deathadder Chroma Keyboard: DBPower LED

Link to comment
Share on other sites

Link to post
Share on other sites

As @M.Yurizaki say you want to go for single thread vs multi thread instead. Higher clockspeeds are just a brute force way to get more IPC but better architecture always win: it's why a 4440 usually matches and even outperforms the 8350 from AMD it's just doing as many or more Instructions Per Clock due to better architecture instead of just throwing more and more power at the chip and forcing the clock higher and higher to try and make up for the IPC disadvantage.

-------

Current Rig

-------

Link to comment
Share on other sites

Link to post
Share on other sites

7 minutes ago, M.Yurizaki said:

No, because IPC is architecture dependent and clock speed affecting performance depends on the IPC. Software has nothing to do with it.

 

The question you should be asking is "what programs are single core performance dependent and multi-core performance dependent?" In which case it depends on if the program is I/O bound or CPU bound. That is they're waiting on something to happen or they aren't.

There are some programs out there that take more advantage of later generation processors though generally it isn't too big of a deal.

Make sure to quote me or tag me when responding to me, or I might not know you replied! Examples:

 

Do this:

Quote

And make sure you do it by hitting the quote button at the bottom left of my post, and not the one inside the editor!

Or this:

@DocSwag

 

Buy whatever product is best for you, not what product is "best" for the market.

 

Interested in computer architecture? Still in middle or high school? P.M. me!

 

I love computer hardware and feel free to ask me anything about that (or phones). I especially like SSDs. But please do not ask me anything about Networking, programming, command line stuff, or any relatively hard software stuff. I know next to nothing about that.

 

Compooters:

Spoiler

Desktop:

Spoiler

CPU: i7 6700k, CPU Cooler: be quiet! Dark Rock Pro 3, Motherboard: MSI Z170a KRAIT GAMING, RAM: G.Skill Ripjaws 4 Series 4x4gb DDR4-2666 MHz, Storage: SanDisk SSD Plus 240gb + OCZ Vertex 180 480 GB + Western Digital Caviar Blue 1 TB 7200 RPM, Video Card: EVGA GTX 970 SSC, Case: Fractal Design Define S, Power Supply: Seasonic Focus+ Gold 650w Yay, Keyboard: Logitech G710+, Mouse: Logitech G502 Proteus Spectrum, Headphones: B&O H9i, Monitor: LG 29um67 (2560x1080 75hz freesync)

Home Server:

Spoiler

CPU: Pentium G4400, CPU Cooler: Stock, Motherboard: MSI h110l Pro Mini AC, RAM: Hyper X Fury DDR4 1x8gb 2133 MHz, Storage: PNY CS1311 120gb SSD + two Segate 4tb HDDs in RAID 1, Video Card: Does Intel Integrated Graphics count?, Case: Fractal Design Node 304, Power Supply: Seasonic 360w 80+ Gold, Keyboard+Mouse+Monitor: Does it matter?

Laptop (I use it for school):

Spoiler

Surface book 2 13" with an i7 8650u, 8gb RAM, 256 GB storage, and a GTX 1050

And if you're curious (or a stalker) I have a Just Black Pixel 2 XL 64gb

 

Link to comment
Share on other sites

Link to post
Share on other sites

1 minute ago, DocSwag said:

There are some programs out there that take more advantage of later generation processors though generally it isn't too big of a deal.

That's by way of using instruction sets available only to that generation and later, and only if the program was compiled to take advantage of it (if the OS or CPU doesn't have some kind of magical sauce to dynamically recompile the code).

 

That also has nothing to do with IPC. A Bulldozer with AVX will smoke a Phenom II in floating point instructions that use it, even though the Phenom II has better IPC.

Link to comment
Share on other sites

Link to post
Share on other sites

Tnen why in linus's biggest mistake video does he say that a core i7 4960x will beat a core i7 6950x in gpu accelerated video rendering as 4gen has higher clock speed. @Misanthrope @M.Yurizaki @Moress @Clanscorpia

Please quote me so that I know that you have replied unless it is my own topic.

Link to comment
Share on other sites

Link to post
Share on other sites

Just now, Ezio Auditore said:

Tnen why in linus's biggest mistake video does he say that a core i7 4960x will beat a core i7 6950x in gpu accelerated video rendering as 4gen has higher clock speed. @Misanthrope @M.Yurizaki @Moress @Clanscorpia

The difference in ipc between the two isn't big enough to make up for the higher clocks of the 4960x. I'm assuming that's what he's getting at. I'm not sure how that has to do with GPU accelerated video rendering though...

5 minutes ago, M.Yurizaki said:

That's by way of using instruction sets available only to that generation and later, and only if the program was compiled to take advantage of it (if the OS or CPU doesn't have some kind of magical sauce to dynamically recompile the code).

 

That also has nothing to do with IPC. A Bulldozer with AVX will smoke a Phenom II in floating point instructions that use it, even though the Phenom II has better IPC.

Some architectures have higher IPC though because of more ALUs within each core so if a benchmark is more parallel it could take advantage of the higher ipc more easily than a not so easily parallelized benchmark.

Make sure to quote me or tag me when responding to me, or I might not know you replied! Examples:

 

Do this:

Quote

And make sure you do it by hitting the quote button at the bottom left of my post, and not the one inside the editor!

Or this:

@DocSwag

 

Buy whatever product is best for you, not what product is "best" for the market.

 

Interested in computer architecture? Still in middle or high school? P.M. me!

 

I love computer hardware and feel free to ask me anything about that (or phones). I especially like SSDs. But please do not ask me anything about Networking, programming, command line stuff, or any relatively hard software stuff. I know next to nothing about that.

 

Compooters:

Spoiler

Desktop:

Spoiler

CPU: i7 6700k, CPU Cooler: be quiet! Dark Rock Pro 3, Motherboard: MSI Z170a KRAIT GAMING, RAM: G.Skill Ripjaws 4 Series 4x4gb DDR4-2666 MHz, Storage: SanDisk SSD Plus 240gb + OCZ Vertex 180 480 GB + Western Digital Caviar Blue 1 TB 7200 RPM, Video Card: EVGA GTX 970 SSC, Case: Fractal Design Define S, Power Supply: Seasonic Focus+ Gold 650w Yay, Keyboard: Logitech G710+, Mouse: Logitech G502 Proteus Spectrum, Headphones: B&O H9i, Monitor: LG 29um67 (2560x1080 75hz freesync)

Home Server:

Spoiler

CPU: Pentium G4400, CPU Cooler: Stock, Motherboard: MSI h110l Pro Mini AC, RAM: Hyper X Fury DDR4 1x8gb 2133 MHz, Storage: PNY CS1311 120gb SSD + two Segate 4tb HDDs in RAID 1, Video Card: Does Intel Integrated Graphics count?, Case: Fractal Design Node 304, Power Supply: Seasonic 360w 80+ Gold, Keyboard+Mouse+Monitor: Does it matter?

Laptop (I use it for school):

Spoiler

Surface book 2 13" with an i7 8650u, 8gb RAM, 256 GB storage, and a GTX 1050

And if you're curious (or a stalker) I have a Just Black Pixel 2 XL 64gb

 

Link to comment
Share on other sites

Link to post
Share on other sites

17 minutes ago, DocSwag said:

Some architectures have higher IPC though because of more ALUs within each core so if a benchmark is more parallel it could take advantage of the higher ipc more easily than a not so easily parallelized benchmark.

I already mentioned that IPC is architecturally dependent. But newer architectures doesn't automagically mean better IPC. e.g.,  K10 to Bulldozer.

Link to comment
Share on other sites

Link to post
Share on other sites

18 minutes ago, DocSwag said:

The difference in ipc between the two isn't big enough to make up for the higher clocks of the 4960x. I'm assuming that's what he's getting at. I'm not sure how that has to do with GPU accelerated video rendering though...

Some architectures have higher IPC though because of more ALUs within each core so if a benchmark is more parallel it could take advantage of the higher ipc more easily than a not so easily parallelized benchmark.

Dont think that is correct cause see,lowering of ipc as ell as clock speed is net decrease in performance. I don't think intel would reduce perfo tnrough the generations

Please quote me so that I know that you have replied unless it is my own topic.

Link to comment
Share on other sites

Link to post
Share on other sites

10 minutes ago, Ezio Auditore said:

Dont think that is correct cause see,lowering of ipc as ell as clock speed is net decrease in performance. I don't think intel would reduce perfo tnrough the generations

He didn't say they lowered the IPC. The IPC went up but clock speeds were reduced which makes them more or less equal. Theres a reason why 6700k>9590 even though the 9590 is at 5ghz. It's IPC. 

Main Gaming PC - i9 10850k @ 5GHz - EVGA XC Ultra 2080ti with Heatkiller 4 - Asrock Z490 Taichi - Corsair H115i - 32GB GSkill Ripjaws V 3600 CL16 OC'd to 3733 - HX850i - Samsung NVME 256GB SSD - Samsung 3.2TB PCIe 8x Enterprise NVMe - Toshiba 3TB 7200RPM HD - Lian Li Air

 

Proxmox Server - i7 8700k @ 4.5Ghz - 32GB EVGA 3000 CL15 OC'd to 3200 - Asus Strix Z370-E Gaming - Oracle F80 800GB Enterprise SSD, LSI SAS running 3 4TB and 2 6TB (Both Raid Z0), Samsung 840Pro 120GB - Phanteks Enthoo Pro

 

Super Server - i9 7980Xe @ 4.5GHz - 64GB 3200MHz Cl16 - Asrock X299 Professional - Nvidia Telsa K20 -Sandisk 512GB Enterprise SATA SSD, 128GB Seagate SATA SSD, 1.5TB WD Green (Over 9 years of power on time) - Phanteks Enthoo Pro 2

 

Laptop - 2019 Macbook Pro 16" - i7 - 16GB - 512GB - 5500M 8GB - Thermal Pads and Graphite Tape modded

 

Smart Phones - iPhone X - 64GB, AT&T, iOS 13.3 iPhone 6 : 16gb, AT&T, iOS 12 iPhone 4 : 16gb, AT&T Go Phone, iOS 7.1.1 Jailbroken. iPhone 3G : 8gb, AT&T Go Phone, iOS 4.2.1 Jailbroken.

 

Link to comment
Share on other sites

Link to post
Share on other sites

18 minutes ago, Hunter259 said:

He didn't say they lowered the IPC. The IPC went up but clock speeds were reduced which makes them more or less equal. Theres a reason why 6700k>9590 even though the 9590 is at 5ghz. It's IPC. 

Sorry I was a bit unclear. What I meant was that if intel did not increase ipc so that it compensated loss in clock speed then it would result in lower net performance which intel is not crazy enough to do.

Please quote me so that I know that you have replied unless it is my own topic.

Link to comment
Share on other sites

Link to post
Share on other sites

IPC and Clockspeed are strongly related within a given architecture.

IPC is instructions per clock. So they are related by definition. 

IPC reliant programs will benefit from higher clock speed and higher clockspeed (under certain parameters) result in higher instruction count. Not higher IPC, just higher instructions executed over a period of time (usually measured in one second). Remember the clockspeed is meassured in Hertz, KiloHertz, Megahertz, Gigahertz. All those units imply the use of "A second" as period of time measurement. 

 

Instructions are different and take different time executing.

An SSE2 takes much less time and physical resources than an AVX instruction, but it also does less things. If you take the time an AVX takes to finish, see the amount of work it did and try to do the same thing with SSE2 in same amount of time, you'll find you just cant do it. Hence, we call AVX more efficient than SSE2. (theoretically speaking). 

 

Intel and AMD have different implementation of those instructions.

Even those with similar names, they "do the same, but in different ways", so they may not be directly comparable. They will take different amount of machine instructions adding up to a diferent execution time (and different IPC of course). 

 

Programs have their code optimized for certain instruction sets, but they can use many.

If the instruction set is not found on the CPU, the program will rollback into a simpler (less efficient) instruction. For instance, if you download PCSX2 (a playstation 2 emulator for PC) it will let you use from SSE2,3,4,4.1,4.2,AVX,AVX2. It will attempt to use the better (newer) available in its own code. As you can see here, Phenoms and FX are trading blows, within the same clock, while FX are clear winners once they bump the clockspeed way beyond Phenom capabilities: 

PCSX2 Benchmark Thread

 

In another benchmark, you can see Phenom resulted 20% faster in single thread clock per clock against FX. 

 

obr5.jpg

So in short: Different programs use instructions in different ways, A 20% improvement in IPC can be opaqued by a 20% clockspeed increase in the counterpart, ending up with similar results overall. To know exactly how much will they matter, you have to factor clockspeed and IPC assuming the program can make use of the same instruction on both CPU's, and this assuming core count does not take any impact. 

 

IPC is an Averaged number, meaning it will vary IRL scenarios:

Other slight changes that affect IPC is cache speed, latency and hit ratio, but i believe it's too much for a single post here. Let's say a cpu does not always take the best path when they fetch the instructions to be executed and they may end up wasting time searching for the proper one when they fail. FX has a lesser hitratio than Phenom and a higher clock penalty for the re fetch process, and that does not account on IPC or clockspeed but it may make a huge impact on server apps like SAP for industries or database accesses. Increasing cache size is a huge factor to prevent misses but it has different impacts across every platform. That's why intel and AMD have different cache allocation sizes and channel accesses to that cache layer. 

 

Recommended reading of how the cache and other things may affect CPU microarchitecture.

 

CPU performance can be tweaked by microcode and better software adaptations.

Nothing here has the last voice on how a CPU performs across every workflow. From one year to the other, FX went from unacceptable to highly valuable once their micro architecture was understood and taken advantage of same can happen with future Ryzen and Kaby. When Skylake came out, it wasn't impressive compared against Haswell. But time took it's course and now we do know Skylake can archieve higher FPS in games due to it's optimized architecture. 

 

----------------

 

This may seem as a lot of info, but it's just the tip of the icebeg. CPU's are hard to understand, and the final result is even harder to predict or interpret.

Depends on too many factors that are already averaged, grouped up and averaged again, so the margin of error for predictability is big. Even more considering software adaptations come years after the new technology releases to the masses. So this will change constantly giving different results across every test. 

 

In short: Don't get too drawn by IPC and clockspeed until benchmarks are here. 

 

cheers!

Link to comment
Share on other sites

Link to post
Share on other sites

6 hours ago, M.Yurizaki said:

I already mentioned that IPC is architecturally dependent. But newer architectures doesn't automagically mean better IPC. e.g.,  K10 to Bulldozer.

I know, I was just using that as an example :).

 

6 hours ago, Ezio Auditore said:

Dont think that is correct cause see,lowering of ipc as ell as clock speed is net decrease in performance. I don't think intel would reduce perfo tnrough the generations

The 6950x has higher IPC than the 4960x. However, the 4960x has higher clocks than the 6950x. In the end, the 4960x has better single core performance. Intel didn't reduce the performance because instead the added more cores. The 6950x is 10 cores vs the 4960x is 6 cores. So in multi threaded tasks the 6950x is MUCH better. Intel is just focusing on multi threaded, not single threaded.

Make sure to quote me or tag me when responding to me, or I might not know you replied! Examples:

 

Do this:

Quote

And make sure you do it by hitting the quote button at the bottom left of my post, and not the one inside the editor!

Or this:

@DocSwag

 

Buy whatever product is best for you, not what product is "best" for the market.

 

Interested in computer architecture? Still in middle or high school? P.M. me!

 

I love computer hardware and feel free to ask me anything about that (or phones). I especially like SSDs. But please do not ask me anything about Networking, programming, command line stuff, or any relatively hard software stuff. I know next to nothing about that.

 

Compooters:

Spoiler

Desktop:

Spoiler

CPU: i7 6700k, CPU Cooler: be quiet! Dark Rock Pro 3, Motherboard: MSI Z170a KRAIT GAMING, RAM: G.Skill Ripjaws 4 Series 4x4gb DDR4-2666 MHz, Storage: SanDisk SSD Plus 240gb + OCZ Vertex 180 480 GB + Western Digital Caviar Blue 1 TB 7200 RPM, Video Card: EVGA GTX 970 SSC, Case: Fractal Design Define S, Power Supply: Seasonic Focus+ Gold 650w Yay, Keyboard: Logitech G710+, Mouse: Logitech G502 Proteus Spectrum, Headphones: B&O H9i, Monitor: LG 29um67 (2560x1080 75hz freesync)

Home Server:

Spoiler

CPU: Pentium G4400, CPU Cooler: Stock, Motherboard: MSI h110l Pro Mini AC, RAM: Hyper X Fury DDR4 1x8gb 2133 MHz, Storage: PNY CS1311 120gb SSD + two Segate 4tb HDDs in RAID 1, Video Card: Does Intel Integrated Graphics count?, Case: Fractal Design Node 304, Power Supply: Seasonic 360w 80+ Gold, Keyboard+Mouse+Monitor: Does it matter?

Laptop (I use it for school):

Spoiler

Surface book 2 13" with an i7 8650u, 8gb RAM, 256 GB storage, and a GTX 1050

And if you're curious (or a stalker) I have a Just Black Pixel 2 XL 64gb

 

Link to comment
Share on other sites

Link to post
Share on other sites

5 hours ago, Ezio Auditore said:

Sorry I was a bit unclear. What I meant was that if intel did not increase ipc so that it compensated loss in clock speed then it would result in lower net performance which intel is not crazy enough to do.

IPC has not changed much in the last few generation of Intel CPUs.

They have been small increments, approximately 10% ~ 15% each generation -- compared to 25% - 50%+ before.

 

The i7-6950X compensated clock speed in exchange for MORE CORES.

That chip is a 10-core 20-thread, as a result, will generate more heat.

The i7-4960X is a 6-core, 12 -thread.

Both are 140W TDP chips. To maintain it within 140W TDP, you have to drop the frequency down.

 

 

If the program cannot make use of the extra cores found on the i7-6950X, then those cores will just sit idle.

Let's say...some random rendering program only uses a maximum of 4-cores and 8-threads.

  • i7-6700K, 4-core, 8-thread, 4.0 GHz
  • i7-4960X, 6-core, 12-thread, 3.6 GHz
  • i7-6950X, 10-core, 20-thread, 3.0 GHz

In THIS case, the extra 2-cores on the i7-4960X, and the extra 6-cores on the i7-6950X will do nothing. To simplify it, you are essentially now looking at:

  • i7-6700K, 4-cores, 8-thread, 4.0 GHz
  • i7-4960X, 4-core, 8-thread, 3.6 GHz
  • i7-6950X, 4-core, 8-thread, 3.0 GHz

The i7-6700K running a 4.0 GHz will out-perform the other i7's.

 

 

You also mentioned, "...GPU accelerated video rendering..."

The CPU isn't doing much of the work, the GPU (graphics card) is.

The GPU is doing the rendering work, not the CPU.

Some further testing may need to be done on this, but in that case, something like a i7-6800K or i7-6700K may give you similar results compared to a i7-6950X.

Intel Z390 Rig ( *NEW* Primary )

Intel X99 Rig (Officially Decommissioned, Dead CPU returned to Intel)

  • i7-8086K @ 5.1 GHz (still tweaking) -- i7-6800K
  • Gigabyte Z390 Aorus Master -- ASUS X99 Deluxe
  • Sapphire NITRO+ RX 6800 XT Special Edition Sapphire NITRO+ RX 5700 XT Special Edition -- 2x Sapphire NITRO R9-Fury in Crossfire
  • 32GB G.Skill TridentZ DDR4-3000 CL14 (16GB TridentZ RGB + 16GB Red/Black TridentZ)
  • SanDisk 480 GB SSD + 1TB Samsung 860 EVO + 1TB WD SN750
  • EVGA SuperNOVA 850W P2 + Red/White CableMod Cables
  • Phanteks Enthoo Luxe Tempered Glass Edition
  • Ekwb Custom loop
  • Logitech G502 Proteus Spectrum + Corsair K70 (Red LED, anodized black, Cheery MX Browns)

AMD Ryzen Rig

  • AMD R7-5800X
  • Gigabyte B550 Aorus Pro AC
  • 32GB (16GB X 2) Crucial Ballistix RGB DDR4-3600
  • Gigabyte Vision RTX 3060 Ti OC
  • EKwb D-RGB 360mm AIO
  • Intel 660p NVMe 1TB + Crucial MX500 1TB + WD Black 1TB HDD
  • EVGA P2 850W + White CableMod cables
  • Lian-Li LanCool II Mesh - White

Intel Z97 Rig (Decomissioned)

  • Intel i5-4690K 4.8 GHz
  • ASUS ROG Maximus VII Hero Z97
  • Sapphire Vapor-X HD 7950 EVGA GTX 1070 SC Black Edition ACX 3.0
  • 20 GB (8GB X 2 + 4GB X 1) Corsair Vengeance DDR3 1600 MHz
  • Corsair A50 air cooler  NZXT X61
  • Crucial MX500 1TB SSD + SanDisk Ultra II 240GB SSD + WD Caviar Black 1TB HDD + Kingston V300 120GB SSD [non-gimped version]
  • Antec New TruePower 550W EVGA G2 650W + White CableMod cables
  • Cooler Master HAF 912 White NZXT S340 Elite w/ white LED stips

AMD 990FX Rig (Decommissioned)

  • FX-8350 @ 4.8 / 4.9 GHz (given up on the 5.0 / 5.1 GHz attempt)
  • ASUS ROG Crosshair V Formula 990FX
  • 12 GB (4 GB X 3) G.Skill RipJawsX DDR3 @ 1866 MHz
  • Sapphire Vapor-X HD 7970 + Sapphire Dual-X HD 7970 in Crossfire  Sapphire NITRO R9-Fury in Crossfire *NONE*
  • Thermaltake Frio w/ Cooler Master JetFlo's in push-pull
  • Samsung 850 EVO 500GB SSD + Kingston V300 120GB SSD + WD Caviar Black 1TB HDD
  • Corsair TX850 (ver.1)
  • Cooler Master HAF 932

 

<> Electrical Engineer , B.Eng <>

<> Electronics & Computer Engineering Technologist (Diploma + Advanced Diploma) <>

<> Electronics Engineering Technician for the Canadian Department of National Defence <>

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×