Ryzen vs i7 (Mainstream); Threadripper vs i9 (HEDT); X299 vs X399/TRX40; Xeon vs Epyc

that is a yuuuuge hurt for intel.

my laptop and OS will never be patched so i can compare the performance in terms of IPC per different software when i do get a ryzen rig.

AMD's biggest leak, if that CB15 score to be true, this is an excellent architectural improvement over zen1 because CB15 only gets like 2-4% max difference from memory OC, rest are all IPC gain which means 10-12% IPC for CB15 seems real. that ST IPC i'd guess be around 190-195, which is still damn amazing already.

in another news, we have the biggest Intel leak, a performance reduction of up to 40%!! due to security patch.

 

How long has AMD been making APU's now? 10 years? And still, the graphics processing power is on the weak side.
People buy intel because they want a strong cpu, and they are satisfied with weak intel integrated graphics.
Seems to me, AMD should offer a product with a low to mid CPU but powerful graphics processing power.

Recently, I got excited when I read about "chiplets".
Why can't/won't amd put a 4 core / 8 thread AMD cpu next to a mid-level discrete amd gpu, and both under one cpu cooler?
I know it's a lot of heat, but there are plenty of aftermarket cpu coolers, and high performance TIM.

When I go on youtube, I see lots of PC builds for "cheap gaming PC" under $500.
And I think .... if only AMD made a APU which had enough graphics power to run any modern game at 1080p/60fps.
People could simply build a mini PC with a APU (and no graphics card) and this would satisfy all their cpu/gpu needs up to 1080p/60fps,
and at a lower cost (compared to a PC with discrete graphics card).

Imagine the AsRock Desk mini with a 30W TDP cpu chip and 60W TDP graphics chip, combined into one 90W APU.

 

Amd's core count is more than making up for the Intel overclocks. Agreed it is not a boon for overclocking hobby.

 

@ole!!! @tilleroftheearth @Papusan @hmscott - this is the thread to discuss the comparison and contrasting of the different products. It may have died off recently, but this is the proper thread for it.

 

@tilleroftheearth @ajc9988 @TANWare

this is the new icelake IPC https://www.anandtech.com/show/14664/testing-intel-ice-lake-10nm/9
looks like zen4 got some catching up to do!

"WARNING" this post is directed related to zen 4 has catching up to do it is not entirely unrelated thus belong in this thread!. the link may contain content which may result in emotional damage to some, view at your own discretion!

 

May be a nice boost if true.

 

I disagree that Zen 4 does. This is best compared to Zen 2, but also doesn't compare instruction per second, and is also a mobile CPU.

With that said, Zen 2 was originally planned to just be a die shrink. When they realized that the frequency was going to regress, they moved up certain uarch advances to Zen 2 to bolster IPC to compensate for the frequency regression. Then, TSMC found a way to get a bit more frequency out of it, which is now what we have.

Zen 3 is more about the IPC increase and to a degree using the 18% area reduction to add new features and likely go wider on design. So, by the time Intel releases Ice Lake server chips next year, it will go against AMD's Zen 3 Milan chips.

Meanwhile, it is proper to compare the Ice Lake-U chips to the 3000 series APUs, which both won't have retail availability until Q3 into Q1 of 2020.

Zen 4 is a further reduction in area, as it is planned for the 5nm long lived node (compared to recent reduction in size of process nodes). 3nm will be the really long lived node, which Samsung claims for 2021 volume, TSMC claims for 2023, but their 5nm lines are supposed to be able to move to 3nm and the 2023 is for the ground up 3nm fab they are building, and Intel is claiming their 7nm for 2021 volume in the second half, which will be between or closer to the 3nm node at the other two fabs.

There is nothing to suspect that after 7nm frequency will go up anymore, rather it likely will continue to decline. That means going wide is key.

Intel did a great job on going wide with Ice Lake, showing great jumps in IPC for Sunny Cove. I've been singing Sunny Cove's praises on IPC for awhile now, but also crapping on Intel's 10nm node process.

So, with the process reducing frequency, but increasing IPC dramatically, we will see single digit performance gains. The question comes, with Tiger Lake this year, how much more performance can Intel squeeze from the 10nm process to increase frequency through decreasing density, while also wondering what IPC increase they will get, as Sunny Cove's replacement did not focus on ST performance, meaning IPC performance.

Moreover, the Ocean Cove uarch DOES focus on IPC again, which is about the time of 7nm for 2021. It is unknown if it will be able to be used on the refined 10nm process or only used for 7nm.

 

OHHH BOY look at dat iPC. not looking good for zen !! @ajc9988 @TANWare @tilleroftheearth

now gotta wait till tigerlake, that 30% IPC improvement doesnt seem too far fetched now

 

By the time they get that IPC to desktop, you have Zen 3 or Zen 4, which is two more iterations of IPC for AMD. How does that spell trouble?

With that said, it does spell heavy competition for AMD's 3000 series APUs, which use Zen+ and Vega (although the Vega APU does outperform Intel's graphics).

But, Intel has lost around 1GHz in clock frequency (cycles per second). Because of that, the increased IPC isn't telling the whole story.

On the other hand, it says AMD needs to get the Zen 2 cores into their mobile and APU lines sooner rather than later.

 

naw zen4 wont have anything other than garbage of an overclock again. at best maybe some ipc improvement. thank god to AMD giving competition so i can buy intel again!

besides, thats just icelake. 9-13% over zen2 PHEWWWWWWWWWW dayuummmm.

tigerlake is even more agressive shown on the userbenchmark for ST performance. i'd bet a good 5-7% over icelake. zen4 gotta have at least 15-20% IPC over zen2 or it'll be left in the dust!

 

I'll believe the numbers when I actually can get a hold of the silicon, till then.

 

So, let's map this out.

For Intel, what is known is they have two 14nm designs coming for desktop: Comet and Rocket. We know they plan on 10nm and 7nm co-existing in 2021, meaning they will move mobile and server chips to 7nm in 2021 and 2022. That means if the wide cores cannot be back ported to 14nm due to area footprint, then Comet and Rocket will have iterative improvements, but not the IPC gains of Ice or Tiger. That means the soonest Intel will have those IPC gains for Desktop market is likely around 2021, which is Zen 4.

For AMD, they had 56% IPC increase with Zen, 4% IPC with Zen+, 13% to 15% for Zen 2 (double that on FP, which they sucked at compared to Intel before that, so just say that is roughly catching up), but have a frequency deficit versus Intel 14nm++ node. Zen 3 is where they planned for the larger IPC increase, although some of that got ported to Zen 2 over frequency fears, so we don't know how much IPC they have left for the Zen 3 uarch on 7nm+. They then go to 5nm, which will also have an IPC increase.

So, let's add up the IPC gains. Intel has 9-13% over Zen 2 with Ice lake. They will likely get 4-6% with Tiger, but let's use your 5-7%. That means that Tiger, coming late next year, but not used on desktop until 2021, will likely have 13%-19% IPC over Zen 2. That is a large IPC amount. Now, if AMD does the 3-4% like Zen+ (which signs say it will be more than that), then another 13-15% jump, that means Zen 4 will have 16-19% IPC over the current Zen 2 chips on desktop by the time Intel uses that on desktop.

Notice how the IPC gains over that time are practically flat? What that means is it will come down to if TSMC can maintain the frequency on their 5nm process and whether Intel can increase their frequency on the 10nm process enough by the time 10nm reaches desktop.

If AMD can do more than that on IPC, or if Intel can, then things change on that calculation. But using known information, I do not see either company as superior in 2021. Just good competition.

Now, once Intel gets 7nm on desktop, and AMD is still on 5nm waiting for 3nm around 2022, there may be a difference. But, the next year, TSMC 3nm fab comes online, and there likely will be no process advantage for Intel at that point, which is around Zen 5 or 6.

As I mentioned, though, Intel's Ocean Cove on 7nm is supposed to be ANOTHER large jump in IPC, which will hit desktop potentially in 2022. That means Intel likely won't have any significant performance advantage until then, if then.

 

i disagree!! intel will lose enterprise market share but i dont care about any of that. on desktop once tigerlake hits, zen4 or 5 gets obliterated!

i sure love my high frequency + high IPC intel cpu!! if amd can give the same i'll go amd, but they can't only thing they can give is more cores, which intel will be forced to give more cores anyway so i'll buy intel!

 

Show me a road map of Intel's showing an Ice Lake or Tiger Lake desktop CPU, please. I'll wait.

Intel doesn't seem to have the capacity on 10nm at the moment. They plan for server to get Ice Lake-SP chips around Q2 2020. But, we already know that they have Comet and Rocket lake planned for desktop, with rocket lake coming next year.

Ice lake-u and Tiger Lake-u, due next year, are mobile parts.

So please show me when tiger lake gets to desktop. Or 10nm Intel comes to desktop.

 

i do know for a fact that tigerlake and zen4/5 be out similar times. zen4 no 20% IPC improvement? then intel i go!

 

I wish the 3xxx series were better OCers for sure, that would make it more fun. I do miss my 7920x in regards to benchmarks, but this 3900x is no slouch. For the money it's quite the bang for the buck. I'm glad that AMD competitive with Intel, because we all know what happened with Intel these past few years (nothing, no innovation).

 

maybe after like 4th yr optimization on 7nm TSMC might get to 5ghz like intel does, but they wont do that until 5nm.

 

Now THAT is a fair statement, since even under my assumptions Zen 4 and a 10nm Intel desktop CPU will be out in 2021.

As such, tigerlake and Zen 4 should be the head to head on desktop CPUS. But, as I mentioned, the IPC should be the same around then. Just a question of frequencies and if they can meet expected IPC gains. Also SOFTWARE OPTIMIZATIONS!

 

i'll see zen 4 once it comes out, but we can already guess what its probably capable of. i would love to be proven wrong and so we can have a fat zen 4 cpu with 30% IPC improvement at 5ghz but we know that is impossible. now that icelake has some numbers, even more so from Ian who tested it himself and this give a bit of credibility to that tigerlake benchmark too.

zen4 has no leaks nothing, prob garbage of an OC just like zen2 and zen+.

 

Why do you focus on frequency when ice lake still has not been seen over 4GHz?

 

Why show all cards? They will come. I prefer not showing my cards in my gaming before it’s needed. Show it before is stupid.

 

eventually they will? intel will likely stuck on 10nm for awhile even if they claim to move to 7nm. 10nm++ will have some good frequency in a few yrs time, though theres always chance thing may change.

do you not want a 12 core 5ghz cpu that has a superb ipc? snappy computer is my fav

 

I am looking at how crappy the frequency was on cannon lake. It was like 2GHz range. For 10nm+, they got it to mid to high 3GHz range. Do I expect near 5GHz on 10nm++? NO! I expect low to mid 4GHz tops. That's it. And that is only if the tiger lake uarch allows for it.

Also, what I want doesn't matter. Double the IPC of ice lake and give me 2.8GHz and I'd be happier than ice lake IPC at 5GHz. Why? Because doubling the IPC, then giving me 10% more frequency than half the 5GHz CPU is faster than the original IPC at 40% more speed.

I don't care about the frequency, and I've told you that numerous times.

Very true. I just don't think Intel has cards to hold back.

Cannon was a 2-core chip that never broke 3GHz. Ice is faster, but not rated over 4GHz. Now, the libraries for high performance are less dense. But you also have way more cores. That is why I have my doubts.

 

who cares about how crappy the performance is now? i mean we always looking whats ahead thats why technology is fun.

besides even if i buy anything now, AMD would cost more as i need to buy entire rig, for intel i already got the laptop and biosi just need the cpu you know.

but with ST performance, intel at 5ghz sitll win even with worse IPC, simply because ipc + frequency is the way to go if ipc isnt too far apart.

 

That's why I mentioned and recommended the 9700K, since you didn't want to use HT.

I do agree people should weigh whether they have to buy an entire new platform or not.

If you planned on sticking with AMD for Zen 3, then I'd say make the switch.

But, Zen 4 and 2021 will bring DDR5, PCIe 5 potentially (or at least on servers), and a new socket, meaning you would need a new platform anyways.

For Comet, there are changes to the power delivery, so it likely will take a new socket for real this time, not just the cause Intel said so, at least for the 10-core parts.

Plus, they also could be adding PCIe 4.0 support with it (don't know yet), but I'd bet Rocket lake does have PCIe 4.0 support.

And IPC*Frequency = Instruction per cycle*cycles per second = instructions per second, which is the main metric for performance. So you are fully correct to look at IPC*Frequency to figure out the likely performance of the CPU, and that the frequency does matter, just like the IPC does.

That is why I've discussed more and more that they may have to trade off frequency to get IPC moving forward on the smaller nodes. As long as it is more performance, who cares?

Edit:

To show my example above on 2.8GHz and double IPC, or 5GHz and IPC, I'll write it out here.

2.8*2=5.6 instructions per second
5GHz*1=5 instructions per second

That shows the slower frequency is fine so long as the IPC is high enough.

 

but no, i'll prob use 9900ks depending on the price. assuming it allows lower voltage than current 9900k.

 

I don’t think Intel will give AMD a free-card to lose the game. None knows what’s behind Intel’s doors. If I was the boss at Intel (or AMD) I also wouldn’t put out everything and show all cards. That would be stupid.

 

I think Intel just underestimated AMD. That also was because they relied on partner companies to get EUV ready, and that was delayed over 4 years. So I think part is Intel sleeping on it, the other part wasn't their fault.

That isn't to say Intel is sitting still. They had Sunny cove waiting forever, they were hiring for Ocean Cove back in 2016 or so, which is around 2021-22 when it hits the market, and they have at least one or two uarchs they made ready to go for 10nm. Just, without EUV, they are having to use quad patterning or more to do the transistors, and they got rid of one of the two dummy gates between transistors, which means a single defect in a dummy gate destroys two transistors or more, especially if at a critical location.

It is ingenious if it worked. But they likely needed EUV to execute it, and EUV is not something they make in house, it is suppliers that make the lithography equipment and masks, and pellicles, etc. Intel is NOT to blame for them not being ready to market when those companies said they would be. It IS Intel's fault for not making contingency plans.

Intel is trying to do their best to right the ship, and I think they will, just that it will take until 2022 to do so, assuming 7nm is on time.

 

also why frequency is so important.

since i dont understand the real deep stuff about cpu, my general idea is this. let say going from skylake to icelake intel showed a bunch of chart on what was improved, assuming along with latency improvement as well.

i dont know much about legacy code and how it flow through a CPU and if it actually benefits from the improvement. take instruction set as an example legacy codes has no use for avx so on. so ultimately comes down to frequency and my experience with higher ghz = better performance on all front including GPU, storage, ram latency etc, everything.

that is why zen2 was a big improvement because ccx design was pretty much the same, with just some architecture difference while they are able to lower the latency by a HUGE margin within ccx itself and ccx to ccx communication, anything you do will literally benefit from that latency reduction regardless what type of workload, as long as they go through zen 2 cores.

also people say IPC is important because lifts all workload. but we know thats not the case cause IPC is actually dependent on the software. what really lifts ALL workload is frequency.

 

With time you’ll see instructions as AVX will be more important. As well high clocks will still be important.

 

IPC depends on the instruction set used. It is in the first word. Frequency is how many cycles can be done per second. So it increases the frequency of each cycle, thereby increasing when the work finishes.

If you compare SSE to AVX512, SSE will take more cycles to complete the task. AVX allows for it to process the instructions more efficiently, reducing the number of cycles to complete. So, even when Intel chips down clock to do AVX2 and AVX 512 instructions, thereby reducing the number of cycles per second, it still finishes earlier than doing more instructions using SSE, even with the higher cycles per second number.

This is why we test with multiple workloads, using multiple instruction sets, etc.

IPC is how many instructions are done each cycle. There are fewer instructions that need done with AVX512, so even if the IPC is lower for that workload, and the frequency is lower, the instruction set can be better overall.

That is what makes it confusing and why they test with different software, that way you can try to figure out which software best matches your uses so you can choose the best processor for your purposes.

 

im just not good at expressing my point unless i give an example.

i understand all that. but what im trying to say is this. if you ask me to bet $100000 that increasing IPC will mean all software improve by a decent amount i would say no because deep down i do not know what the cpu or software is doing at ALL. but if its reversed for frequency i would do it because i can't think of a way to increase frequency while having less performance outside of throttling.

so to speak, i rely on my experience which i can easily give it up because i know that i know nothing about what its doing at a given time and that my experience is similar an value/result of many factors affecting it. if i think i know something is better for a reason i'll easily jump on it regardless of my experience. i know zen2 isnt better, if we have a way to just use it's inner core it'd blow intel away. zen2 arch doesnt allow that to happen because they design it for eypc, not for consumers. and frankly it already won and beat the mesh system that intel has.

 

I've got an example for you then:

I'll sell you either an FX8350 at 5GHz or a Ryzen 3700X running at 4.3GHz. Your choice. One has higher frequency, but the other has better IPC.

 

then i will need to test on my majority of workload and come back to you and tell you i cant buy it because.
1. performance is **** or
2. i cant fit it into my laptop
3. other i havent think of yet.

besides thats going to the extreme. fx is very old i could rule out that it prob has a lot of performance issue. so to save time i'd take that chance and not measure my workload on it. heres the thing though. 5ghz with poor performance because of IPC, but IPC is result of reducing latency, optimize size cache and bunch of other cpu infrastructure. so there MIGHT be 1 or 2 thing that doesnt touch any other part of CPU and it'll out perform the current CPU. again im simply not ruling those out and at the sametime, pointless because most software do use bits and parts of cpu.

 

This is somewhat wrong comparison. You can somewhat have both. 5.0 GHz is good but not the expense of/at the other. Most of both will always be preferred. But go for one don’t make it good.

 

@ajc9988 wont be surprised if u sell it to someone who only after numbers they'd go for FX lmao. garbage dumped mission accomplish.

 

I was using it as a VERY black and white example of IPC versus frequency.

I wasn't meaning it as saying you must pick one or the other. As I mentioned, true performance is IPC*Frequency. So, there are times the lower frequency will win on something with a high IPC over a CPU with high frequency, but lower IPC.

For Intel, they now have lower IPC. But, because they can overclock higher and hit such high frequencies, it can outperform the AMD CPUs at certain tasks, especially on older code not made to use Ryzen.

I'm also not discussing driver optimizations, etc. at the moment, which further compounds the problem. Also not talking software optimizations, which Intel, because of years at the top and spending billions has created that software ecosystem, AMD is playing in an Intel world, not an open world. As such, AMD has to try to optimize itself to work with things optimized for Intel, which is a sandbag, but not unfair, for AMD.

 

Yeah, I don't own one. Only reason I ever would is if I got into LN2, just to shoot for the frequency world record! LOL!

 

i guess breaking it down it goes like this:

frequency you dont really break it down though it is determined by silicon quality and design density, what you get is the base of speed? i guess thats a way to put it.

IPC, improvement in IPC can result from reducing latency in FCLK, UCLK, increase in cache, pipeline etc etc. but if a software is used doesnt touch cache or go to another CCX, all it take advantage of is the in core improvement of latency reduction. so that IPC wouldnt be entirely accurate, which is reverse for frequency, will ALWAYS give the boost.

 

I think it funny all you guy's are basing on future What If's. I would base any purchasing decisions on current silicon. I have the 1950x and will look at the new Zen2 TR core with 32 cores but have not made a decision., why no silicon yet!

If you have an 8700k that currently suits your needs you are nuts to upgrade to a 3900x etc.. If however I am looking again to upgrade in 2022 I will base that on the available hardware at the time.

 

That is closer, but isn't quite right either. Older software cannot use the newer instructions on a CPU. In fact, if you go back far enough, the newer CPUs may not have the right instruction sets to even run the code. LOL.

For IPC, as you mentioned, you don't know which instruction set the software is using, or telling the CPU to use. If you for sure knew the instruction set, then IPC can tell you precisely how many instructions it runs each cycle. It is a very accurate measure. But, as mentioned, few know what instruction set their favorite software is running, and there are numerous instruction sets a CPU can use.

http://www.cpu-world.com/CPUs/Core_i9/Intel-Core i9 i9-9900KF.html

• MMX instructions
• SSE / Streaming SIMD Extensions
• SSE2 / Streaming SIMD Extensions 2
• SSE3 / Streaming SIMD Extensions 3
• SSSE3 / Supplemental Streaming SIMD Extensions 3
• SSE4 / SSE4.1 + SSE4.2 / Streaming SIMD Extensions 4 ?
• AES / Advanced Encryption Standard instructions
• AVX / Advanced Vector Extensions
• AVX2 / Advanced Vector Extensions 2.0
• BMI / BMI1 + BMI2 / Bit Manipulation instructions
• F16C / 16-bit Floating-Point conversion instructions
• FMA3 / 3-operand Fused Multiply-Add instructions
• EM64T / Extended Memory 64 technology / Intel 64 ?
• HT / Hyper-Threading technology ?
• VT-x / Virtualization technology ?
• VT-d / Virtualization for directed I/O

Those are the instruction set extensions. Then you have were it recognizes what the CPU can use, picks it, then presents the instructions. You have a lot of moving parts on the cache, such as instruction and op caches, etc.

Some elements of a CPU, like effects of latency, cannot be isolated as often from IPC. Now, if you feed it a memory intensive workload that cannot stay on the CPU cache to perform the task, then memory latency will be more weighted.

That is what makes this a PITA to describe. There are lots of moving parts, some of which are just contributors to IPC.

That is what Ryan Shrout focused on. He would focus on a nuanced latency, rather than looking at how that latency would impact overall performance. Why? Because the latency disparity was larger than the performance impact overall.

Now, understanding the latency is important, but it can also lose the forest through the trees.

For example, does it matter that the memory has 72ns vs 64ns? We like the lower number, but that doesn't tell us much. When we say that the 64ns latency gave us a boost of 10%, now we care.

It is about the overall performance, which is instructions per second, that we care about. That is broken into IPC and Frequency. You can further examine what contributes to IPC, but that is your high level number to multiply against frequency to get what you really want: PERFORMANCE!

 

memory is another factor and eventually it'll just be a part of the cpu all tier based storage pretty much from cache to ram to storage.

72 to 64 may not give 10% in performance increase but what if ramdisk is used? storing junk files or using ram for cache need, benefits a lot. really comes down to if you have a use for it though.

if i have an AMD system i could make a few test on its ipc all pinned to a single core. assuming it REALLY doesnt go into other ccx but i doubt it donno what window might be doing.

i have done this test however is CB15 ST test if you let it run u'll realize thread usage hops around cpu cores but only uses a single thread. however if u pinn it to a single thread you actually lose performance which baffles me. if data doesnt flow to over to other core and stays in tact wouldnt it make it faster. should be like that in theory, but ultimately i have no idea what its doing. i could make guesses on scenarios how shifting to different thread would out perform pinned thread but i would have no proof.

 

Funny thing, with the new scheduler, even though it dances around on Intel still, it gets locked to a single core on the new Ryzen CPUs. LOL

I don't even want to get started on the effects of schedulers. If you want, look for white papers on the cutting edge, which are latency aware schedulers! It's some good reading, even if dense.

 

Kingpin 2080 Ti with Intel 9900K, he did with the AMD chip a couple days ago.

 

IPC? GHz? Magic Pixie Dust? Those are just parts of the answer I look for.

Platform, Platform, P latform. The whole is always more than the sum of the parts (when it's done right).

Sunny Cove/Ice Lake later this year is what is really exciting to me. Especially on this notebook forum.

Catching and exceeding in a notebook what desktop systems are doing today in a few short months is newsworthy, even when compared to Ryzen 2 desktop platforms. This is a tangible improvement that anyone will be able to relate to (yeah; assuming the shipping models match or exceed the previews we've seen so far).

Far different than the zombie-like chant 'more cores today will keep you future proof tomorrow' that is at best, marketing fluff that the masses seem to have bought hook line and sinker so far...

The tiny peek I have had at Ryzen 2 platforms in real-world workloads (not my own, this was other's systems) did not impress me at all.

AMD should be given a bronze star for participating and a silver star for playing catch up to Intel in a few short years (or, was that 13 or so)? But Intel is still in the driver's seat for anyone that is not a fanboy of any, one company, but a fan of true performance for the workloads most people do today (and still).

Are the 'many' cores the new bling of today? I guess. But those cores were sitting there mostly unused and yeah, the performance that was left on the table to have 'AMD' inside instead of Intel, today (and still) was very evident to me and acknowledged by the owners of those Ryzen platforms too (for the 'normal' workloads they do daily).

 

I find when talking about 2021 or 2022 it no longer is a few short months. Especially when it is vaporware we are talking about in that time frame.

 

I guess I'm of a different mindset then your friends who own those Ryzen platforms. I find my 3900x to be a great machine, even when compared to my 7920x (or my 7900x, 9800x, 9900k, 8700k, 7700k, 8750h...). Intel has been stagnant since Skylake, their only real improvements were adding more cores. I'm sure if AMD didn't catch up (and surpass) Intel, then we would be on 14++++++ for the next 10 years...

 

Start at the end. Ice Lake-U doesn't TOUCH the performance of desktop systems. Period. Exclamation point. It exceeds on IPC alone. But, it also is 1-1.5GHz SLOWER than Intel's own 9900K. The IPC doesn't make up for it getting SMASHED by frequency. And that is before we discuss its measly 4 cores compared to double that on the 9900K and 9700K (HT amounts to about 30% more performance, introduces security vulnerabilities on pre-Ice Lake Intel chips, and still relies on software able to use higher thread counts, which is growing, but is addressed below).

Sunny Cove is a great architecture, nonetheless, and is something I've applauded for months, even to you liking my posts (you only like my posts saying positive things about Intel, not their faults, or AMD's positive attributes, but you will like negative things I say about AMD; almost like there is a bias instead of weighing the merit of each statement on its face... no matter). To that, I will not let anyone denigrate those engineers work, because it was GOOD WORK!

Moreover, if all you took from the more detailed description of how IPC and frequency interact, so be it. But, there is no platform advantages to Intel anymore. For certain software, there is an ecosystem advantage, but platform, not really. Intel doesn't have optane advantage unless you want a boot drive, which we now know isn't its best use. You can use optane drives on AMD systems. It has Optane DIMMs, but so far, as far as consumers go, it isn't worth it, so says EVERY review out there. That is a server product and niche for use at that. Thunderbolt is open source as of this month, taking away that reason for Intel, with some MBs for X570 already integrating TB ports directly on the MB. So what exactly, specifically, is the "platform" advantage? Software ecosystem, you have a point. AMD is playing in an Intel world (say that while thinking of the song material girl, you will laugh!!!). But platform, THAT is the real joke.

Even when PCIe 5.0 gets here, AMD already signed on to CXL for coherence (which is really exciting), while Intel signed onto gen Z connects between nodes, designed in part by AMD. When those are put together, seriously, coherence between board components and nodes, including cache and memory, things are going to get bonkers in the best of ways! But that is server and datacenter. Don't know when that will filter down to mainstream.

Except that this has been found correct. If you look at everyone that revisited their 7700K vs 1600/X and 1700/X reviews, any that said go for the 4-core did NOT age well. In fact, across the board in modern titles, it looks like the worst advice imaginable. My how a couple years of software updates can change that contour.

Sure, certain workloads still haven't given way, like photoshop, but that is few and far between. I'll touch more on that in a minute. Either way, you are missing what has changed if you are saying that, and there are now a ton of articles to prove it, as they revisited that very concept leading up to the release of the 3000 series. So your opinion doesn't matter, hard facts say otherwise.

Really? Considering you have, in real world workloads, AMD crushing the 9900K in many of them, especially multithreaded workloads, which is the MAJORITY of workloads, unlike photoshop, how did you find a very narrow group that uses only single threaded programs? Even Adobe Premiere shows AMD winning it. So I'm very confused by your statement when EVERY review shows practically the opposite conclusion, along with a member here saying they are impressed over their prior owned Intel offerings.

Or maybe you haven't read about OEMs being tired of Intel's crap, Intel not delivering on time, making excuses, etc. for the past couple years? Maybe that is it.

So your statement is silly in the face of hard data. Intel still has its places where it wins. That has NOT been heavy multithreading for years. Now, it is even fewer wins. Gaming and Photoshop type things, mostly. So how about instead of your generalities, you just sing where each CPU excels? If you did that and acknowledged Intel's shortcomings, people might trust you more.

I can agree. Once again, speculation. But I wanted to show how even Intel's expected IPC changes in the next years don't mean AMD is done for at that time, while also getting the chance to discuss in more nuance and depth what IPC and Frequency are and how they relate to performance. That is why I found it a good talk.

Definitely this! If Intel felt no pressure, they wouldn't have put out the token Cannon Lake, which did have IPC improvements, but Intel left that out of their comparison to make Ice Lake IPC look more impressive and to not bring the focus on how delayed 10nm volume is. Intel was planning on putting Kaby-X 4-cores on HEDT to try to get people to move to the more expensive platform. When Ryzen dropped, they had to reverse tactics and put more cores on mainstream rather than moving more people to HEDT. And now, a couple years later, they are selling half the HEDT chips by volume as AMD in some regions (referring to DIY numbers, as Intel still has the largest presence in OEM builds, although, as referenced above, Intel is really angering them as of late). So, but for AMD, we'd still have quad core mainstream, which, as also mentioned, those reviews suggesting a 7700K over a Ryzen gen 1 did NOT age well.


Edit: For comparison, here is EposVox discussing encoding for streaming.


Summary:
Quicksync from Intel = image quality issues, don't use
AMD GPU encoding = broken back to 480, don't use when streaming, although can encode if not streaming (he has a separate video specifically on the 5700 and 5700XT encoding worth a watch)
AMD CPU encoding = if 3600 or above, set to medium and forget about it, 3900 = magic
NVENC = great for lower bitrates when comparing with Fast and Very Fast
Intel CPU = you have to test, because on the mainstream with cores and threads, you have to find your sweet spot; gave advice on how to test.

 

Have you heard the phrase 'a poverty of words'? Your 'book reports' don't change anything and you keep missing the point that most people don't use computers for streaming, etc. ...

Have a good look at the Anandtech article. In some workloads, it destroys desktop systems.

This is Intel's first real response and only a preview at that. The future (from 2006, today and forward) is still pretty much an Intel world except for very specific workloads where many cores do matter. Intel only has to deliver now (and there is no reason to think it can't at this point).

A balanced platform that is optimized for real-world workloads as Intel has been striving and delivering for many years now is all that matters and is the only thing worth pursuing (and that includes their SSD's too). It includes IPC, it includes the optimum number of cores for today's workloads (and the options for 'many-core' workloads were always there, at a price) and it also includes high GHz speeds as possible.

I'm not trying to change your mind. I really don't care. But there is a huge disconnect between what you 'analyze' and present as AMD superiority and what the reality is.

From what I saw, in person, of the few AMD systems I've seen running for workloads that most everyone uses daily, they are as good a deal as going back to the late '90's and putting monster wings and racing stripes on your dads Econo Japanese import.

If work is performed faster if battery life is greatly increased on mobile systems if lighter designs are possible (or bigger batteries in the same size used today) then the entire platform matters. This is what I see with Intel. They understand that.

AMD may look like it's firing on all cylinders in contrast to Intel the last few months, but digging a little deeper things fall apart quickly, IMO.

Give me a balanced platform where the 'experience' is better than what I'm currently using and you can keep your stats, scores, and analysis of why AMD is (or; should be) the better choice, but somehow isn't.

 

we need to give credit where it is due and imo icelake is pretty nice. its a damn shame intel kept it and not giving us that improvement for nearly half a decade, could have given that to us on 14nm++. i can't wait to see what tigerlake will be capable of, brand new arch. too bad that review wont come out till something like 2022 lmao