The vast majority of users don't run open loop cooling. The vast majority of reviewers aren't going to use open loop coolers in their reviews. They might have a separate article/video addressing that, but they won't put it in the main review. Even when not running strictly spec, reviewers aren't going to post reviews of CPUs with hand tuned, overclocked to the brink results.
So you posted all of that and the above is all I had to read to tell that you weren't paying attention.
I was responding to a post about GN's "Highly Tuned / Overclocked 10600K VS" video. This is constantly used to show that you can't OC an Intel rig to beat Zen 3.
That is a completely false conclusion as I demonstrated. I'm *not* talking about "normal users". The guy in the video debunking GN even says this is for the 1%.
If we are talking about normal users we'd be talking about Dell and so on with plug and play settings, and Anandtech's benchmarks would be the right ones to look at.
For plug and play users, Zen 3 wins, no question and I never said it didn't.
But if you are going to DIY and you are going to OC and go for higher performance, the landscape changes quite a bit.
Bottom line is that Gamers Nexus' "highly tuned \ overclock" benchmarks are an absolute joke.
4Ghz ring bus with DDR4-3200 on a 10600K with a $700 motherboard is laughable.
So you posted all of that and the above is all I had to read to tell that you weren't paying attention.
I was responding to a post about GN's "Highly Tuned / Overclocked 10600K VS" video. This is constantly used to show that you can't OC an Intel rig to beat Zen 3.
That is a completely false conclusion as I demonstrated. I'm *not* talking about "normal users". The guy in the video debunking GN even says this is for the 1%.
If we are talking about normal users we'd be talking about Dell and so on with plug and play settings, and Anandtech's benchmarks would be the right ones to look at.
For plug and play users, Zen 3 wins, no question and I never said it didn't.
But if you are going to DIY and you are going to OC and go for higher performance, the landscape changes quite a bit.
Bottom line is that Gamers Nexus' "highly tuned \ overclock" benchmarks are an absolute joke.
4Ghz ring bus with DDR4-3200 on a 10600K with a $700 motherboard is laughable.
But the examples you were posting were from GN's standard review, not their extreme tuning video. You haven't addressed their actual extreme tuning results (AMD v Intel) at all.
But the examples you were posting were from GN's standard review, not their extreme tuning video. You haven't addressed their actual extreme tuning results (AMD v Intel) at all.
Yeah, I did, I just didn't spoon feed you an image. It is linked to in the post I was responding to. The guy that I linked to that debunked it showed GN's extreme tuning results, used GN's memory settings on his rig, and lost 21% of his performance in Tomb Raider.
The absolute numbers aren't even that relevant, the point is that GN's "tuning" is garbage.
Yeah, I did, I just didn't spoon feed you an image. It is linked to in the post I was responding to. The guy that I linked to that debunked it showed GN's extreme tuning results, used GN's memory settings on his rig, and lost 21% of his performance in Tomb Raider.
The absolute numbers aren't even that relevant, the point is that GN's "tuning" is garbage.
The link you provided is a guy comparing his extreme overclock, hand tuned memory with cache ratio overclock and unknown stability, to GN's standard review settings which GN's uses for a very good reason. He did NOT compare to GN's extreme tuning video where GN overclocks the cache to 4.9 GHz (they highest they could achieve with stability) with memory at 4 GHz and hand tuned timings. BTW, they could go higher than 4 GHz on the memory for Intel, but they say that it forces them to start to loosen the hand tuned timings which then makes the higher frequency a wash.
If we're going to go off of random extreme overclocking videos, why not find some LN2 videos for comparison? Might as well, right? It's not like what cooling they use matters or how unstable their system may be matters, right? It's not like we can't do extreme things on AMD systems as well. . .
But the examples you were posting were from GN's standard review, not their extreme tuning video. You haven't addressed their actual extreme tuning results (AMD v Intel) at all.
The entire first section of that post is about GN's extreme tuning, which was linked to by the person I was responding to. The first paragraph is entirely about his settings :
Everyone's big assumption here is that Tech Jesus really had the optimum settings. He didn't. His settings are 5.1Ghz all core with a 4Ghz ring bus using DDR4-3200 CL14. Someone using AMD might think that is good. It isn't for Intel.
The entire first section of that post is about GN's extreme tuning, which was linked to by the person I was responding to. The first paragraph is entirely about his settings :
You're still ignoring that GN has more than 1 video and that the specs you are talking about are from their standard review video, not their extreme tuning video. I don't know what else say at this point.
Edit: You are also ignoring Igor's labs results and the long list of review sites that don't test just bog standard 3200MHz RAM and yet still show the same results. If you can find any actual valid comparison between a highly tuned Intel vs highly tuned AMD machine (not just some random guy only testing one side with no testing control whatsoever) that shows your claim that Intel can pull ahead, then let's see it. Otherwise all you're doing is trying to create doubt and speculate what could be based upon extremely anecdotal evidence when there is a mountain of evidence against your claims.
Everyone's big assumption here is that Tech Jesus really had the optimum settings. He didn't. His settings are 5.1Ghz all core with a 4Ghz ring bus using DDR4-3200 CL14. Someone using AMD might think that is good. It isn't for Intel.
The Tech Jesus specs you listed are dead wrong. For Intel they used 4000 Mhz 15-15-15-34 with custom subtimings and Cache Ratio at 49x (so ALMOST your state 50). Anything beyond that wasn't stable or had frame-time inconsistencies. This was stable 24/7. And yeah, on some systems you get a bit more milage out of them, but this is borderline on something someone could use daily without fear of constant tunings/crashing.
Your video has the CPU @ 5.4 Ghz, 5 Ghz ring and ultra low latency memory. Even the vast majority of 10900K's wont reach it, no matter the cooling and definitely not in a 24/7 gaming rig.
Besides this is getting very far from what interests most people. From my experience turning XMP on is all the overclocking 99% of people do (and many don't know that even).
The entire first section of that post is about GN's extreme tuning, which was linked to by the person I was responding to. The first paragraph is entirely about his settings :
No, as i timestamped in the post above. It was 4000Mhz CL15 and 4.9 Ghz cache-ratio (they tried to go beyond- it had inconsistencies/wasn't stable). There are plenty i5's out there that won't even reach that while being 24/7 stable.
It's clear you haven't watched the video, as these are even listed on some of the bar-graphs ....
And never mind the Tech Jesus.. Why are you ignoring Igorslab results that are also heavily tuned? The best cases for AMD look like this:
No, as i timestamped in the post above. It was 4000Mhz CL15 and 4.9 Ghz cache-ratio (they tried to go beyond- it had inconsistencies/wasn't stable). There are plenty i5's out there that won't even reach that while being 24/7 stable.
It's clear you haven't watched the video, as these are even listed on some of the bar-graphs ....
Actually the best case for Intel on that benchmark is this.
For some reason as resolution goes up, the Intel rigs tend to win. Yes I know GPU limited and all that hence 720P, but something has to be going on with Zen 3's ability to move data from memory to the GPU as this kind of flip shows up in multiple benchmarks - and this is obviously not within the margin of error :
Same thing happens on Tomb Raider, as the resolution goes up the winners flip, though not as dramatically, it is very consistent with the top 4 slots occupied by the only 4 Intel chips on the chart - and by the time you get to #5 it's not in the margin of error anymore :
Igors labs are not heavily tuned. DDR4 4000CL18 is not heavy tuning. Loading XMP memory profile is not heavy tuning either.
XMP profiles are made for average Joe, even if they contain decent primaries ( like say 3600C15-15-15 kit ), one can rest assured that after loading that profile motherboard will apply those settings and relax a ton of secondary and tertiary settings while jacking up SA/IO voltages to the moon.
In fact even memory manufacturers setup XMP profiles in a way that even most horribad motherboard with horrible memory routing still has a chance to run it after loading said profile.
For example this was elite B-DIE mem for like 2-3 years
Settings for XMP 3600CL15 marked as red are outright bad and meant for those stupid MBs. People run like 280 RFC, RRD_S should be 4, RRD_L 6, TFAW 4xRRD_S with this RAM.
There is also a question of what is not in profiles, MB is free to set tertiaries as it likes.
For example
If you want to discredit Intel's performance, set those to 30+ and see memory BW drop by 20GB/s and performance not so great anymore.
My friend had 9900K with ~400GFlops in Linpack on XMP, and once tuned on same DDR4 3600 it was almost 500Gflops with unhealthy increase in power usage and voltage requirements.
The rule no1 of memory tuners is: never load XMP profile, and if it was loaded you need to clear CMOS to get rid of what MB maker has done to exposed and hidden knobs of memory tuning.
Igors labs are not heavily tuned. DDR4 4000CL18 is not heavy tuning. Loading XMP memory profile is not heavy tuning either.
XMP profiles are made for average Joe, even if they contain decent primaries ( like say 3600C15-15-15 kit ), one can rest assured that after loading that profile motherboard will apply those settings and relax a ton of secondary and tertiary settings while jacking up SA/IO voltages to the moon.
In fact even memory manufacturers setup XMP profiles in a way that even most horribad motherboard w
The rule no1 of memory tuners is: never load XMP profile, and if it was loaded you need to clear CMOS to get rid of what MB maker has done to exposed and hidden knobs of memory tuning.
Thanks for the info. I use Custom timings myself and knew XMP was bad, due to no secondary timings, but didn't know it was THAT bad. Thought the timings were st least no-worse than stock 1.2v settings
Taiphoon Burner has a pretty complete database, that allows you to browse different brands sub-timings. It also allows you to create your own high performance profile.
Changing those can have huge effects. As example, on my RAM tREFI XMP is 11730 and tRFC is 560. tREFI is how many clocks before the memory needs to be charged to hold contents, during which it must be idle. tRFC is how many clocks it must charge.
560/11730 means 4.77% of the time my RAM would be idle, recharging, using XMP.
I actually run 520/44000 = 1.18% of the time it is idle, recharging.
The vast majority of users don't run open loop cooling. The vast majority of reviewers aren't going to use open loop coolers in their reviews. They might have a separate article/video addressing that, but they won't put it in the main review. Even when not running strictly spec, reviewers aren't going to post reviews of CPUs with hand tuned, overclocked to the brink results. First, because it takes a lot of time to get to that point and they don't usually have near enough time for that between receiving hardware and embargo ending. Second, hand tuning and overclocking is not a widely developed skill and very few even enthusiasts bother with it anymore. Third, getting RAM that can clock really high at really tight timings is typically very expensive and most people would benefit far more from spending that money going up a tier in CPU or GPU. Lastly, hand tuning everything to the max is also highly dependent on sample quality, both of the RAM and CPU so even if someone posts amazing results and you spend the money buying exactly what they did, there's absolutely no guarantee you'll be able to match their overclocks/tuning/performance.
In the video I posted earlier, GN uses faster memory than what any of their AMD CPUs could support, with hand tuned timings, and the cache ratio overclocked to 4.9 GHz and the 10600K still lost to a stock 5800x with 3200 MHz RAM in 4/5 tests. So no, this is not true at all.
One example from someone with an open loop cooler, no testing controls, and unknown stability is not proof of anything.
That's not a lower score, it's called margin of error. The difference is only 0.17% and indicates a non-CPU bottleneck.
You're right, GN doesn't do extreme tuning in their base reviews. They do tune the memory, but not to the extreme. They do separate videos for extreme tuning which I already posted a link to that you seemed to completely ignore.
I don't put much stock in random internet postings, but even still, look at his CPU game/render numbers and compare to a Zen 3 CPU.
Neither of those architectures was designed for 14nm. They need both needed the increase in xtor density and power reduction to effectively replace Skylake. Rocket Lake is so late because the engineers had to figure out how to squeeze a modified *Cove architecture onto 14nm dice.
Yes of course that is true. But the point I was trying to make it that when Intel stalled at 14nm they should not have stalled architectural development. From what I've read processor architecture is first designed and tested using computer simulations, this way they can compare various changes to other known designs and see how they will perform, then that design is moved to silicon.
In order to increase performance while holding costs the trend over the last 30 or 40 years was to add transistors to increase IPC while shrinking the process to maintain die area (profits) and hopefully increase frequency and power efficiency.
So this breaks down for Intel at 10nm. They just can't get it happening.
What I'm saying is instead of halting architectural development until the 10nm process is viable they should have simply continued with new architectures at 14nm. Sometimes things don't go according to plan and you have to improvise and make the best with what you have. This is exactly what they are doing now with Comet Lake. I'm saying they should have done this 3 years ago. Yes, profits will suffer as will thermals with the large core but it's better than the same core over and over again. Something they are realizing now, a few years too late.
I'm sure they never thought AMD and Apple would catch up so quickly, which is why they ultimately "froze" instead of continuing to innovate. It's easy to compete when you are way ahead of the other team, once they catch up or pass you then you see what's inside. That's where Intel is now. The next year or two is extremely critical for them. If I were running that company I would take the heat from the board and stock holders and focus on advancing the technology. Use those billions to save the company. If they don't understand the precarious position they are in they are not going to be the Intel we remember. Unlike 10 or 15 years ago there are really competitive forces in their marketplace. AMD, Apple, Samsung, TMSC, just to name a few. And those guys are hungry for a piece of the pie Intel has been hoarding.
While TMSC 7nm node for Zen 3 is fantastic, a big part of where Zen 3 is beating Intel is IPC (architecture). Intel has a huge clock advantage yet still can't beat Zen 3. Imagine Intel had the equal of Zen 3 IPC-wise but at 10% better clocks. Sure the thermals would be a lot higher but they'd be holding ground on the architecture.
As far as I know, and please correct me if I'm wrong, we have no good IPC data for comparing Zen 3 to Sunny Cove and Willow Cove. All of the Cove's are mobile with clocks flying all over the place. If someone could do some testing using HWinfo's average effective clocks we know something.
From what I have seen here how I think this will pan out. Rocket Lake (Sunny Cove) overall is going to be a few percent behind Zen 3 and the clockspeed advantage will make up for it in some apps and not others. So we'll be seeing pretty even performance with Zen 3 being much more efficient. Moving the Willow Cove core to the desktop will help with apps which require the additional cache.
I think Willow Cove and Zen3 are pretty comparable IPC-wise with Zen 3 having a slight overall advantage. We will see the Anandtech review of RL showing pretty even performance but RL needing double the watts. Of course Intel can only compete to 8 cores. If you need more than that then it's AMD all the way.
You seem to expect ridiculously much out of Rocket Lake. On what are you basing that assumption?
AMD was doing poorly with the 3xxx and earlier because it had similar amount of usable L3 cache for games (2x16 vs 16MB) and higher memory latency. While the memory latency is still higher, 5xxx series improved it by 5-10ns and doubled the usable cache (32MB vs 16MB) which with many workloads cuts the actual latency in half (depends highly on workloads and cache hits but on average it's right around 50% less for double the cache) .
The result is that:
AMD does particularly well in games where the games fast-path fits into L3 the cache. These are usually lighter eSports titles (look at Rainbow Six, CS:GO, Valorant, Overwatch (which is clearly bottlenecked on all CPUs).
But it still beats Comet Lake on average even when the latter is heavily overclocked and uses optimized memory as Gamers Nexus and igorslab show)
The problem with Rocket Lake is, that while L2 cache size is increased, L3 size remains the same at 16MB. Cache bandwidth seems to be considerably increased (particularly L1), but not latency. Also the memory controller was also already very aggressive on Skylake. There really isn't that much room to go beyond 40ns of memory latency. So where do you expect to see the gains? If it were "general IPC" Zen 3 would be much stronger than it is.
Going back to the "GPU limited" point:
1. While Comet Lake pulls some wins on igorslab benches, some look like this (I would hope you agree that RTX 3090 @ 1280x720 is NOT GPU limited):
Please do check out the World War Z results above, this is an excellent example of a game fitting well into L3 cache. Do You really expect Rocket Lake to make up the difference AND add 20% to that?
2. Take Techpowerup's pure draw-call tests (where it's very easy to see when it shifts from 100% CPU to 100% GPU limited):
The problem I saw with "Rocket Lake Gamining king!" results is that it all but certainly won't retake the crown in benchmarks that play well with L3 cache, as AMD simply has 2x more.
Games that have loads of cache-misses are already very competitive on 10900K and yes, it will improve there, but nowhere near 20%.
I mean, just look how 6700K performed vs 2700K or 3770K for instance, although IPC uplift is in the same ballpark as Comet Lake -> Rocket Lake.
You seem convinced that Ryzen 5000 will match Rocketlake - what gives you this impression?
Personally, I see that the 10900k is on average, just a tiny % behind Ryzen 5000. It's very logical to assume the new Intel 11th gen, with new architecture, will be faster than the current generation, making it faster in gaming and thus the new gaming performance champion.
Not really rocket science here, and doesn't really matter since you can't buy a Ryzen 5000 without going through great effort, due to AMD not being able to afford their own fabs and having to beg TSMC for scraps.
You seem convinced that Ryzen 5000 will match Rocketlake - what gives you this impression?
Personally, I see that the 10900k is on average, just a tiny % behind Ryzen 5000. It's very logical to assume the new Intel 11th gen, with new architecture, will be faster than the current generation, making it faster in gaming and thus the new gaming performance champion.
Not really rocket science here, and doesn't really matter since you can't buy a Ryzen 5000 without going through great effort, due to AMD not being able to afford their own fabs and having to beg TSMC for scraps.
I was mostly talking about the 20% performance increase over AMD that I found unlikely, not that it will probably be faster in gaming overall (which is pretty much a given).
Not really rocket science here, and doesn't really matter since you can't buy a Ryzen 5000 without going through great effort, due to AMD not being able to afford their own fabs and having to beg TSMC for scraps.
Demand is outstripping supply. AMD isn’t begging for scraps, they have a contract with TSMC. They are a top customer (2nd only to Apple, though it looks likely that they will be switching places). Why are you making stuff up to defend your position?
You seem convinced that Ryzen 5000 will match Rocketlake - what gives you this impression?
Personally, I see that the 10900k is on average, just a tiny % behind Ryzen 5000. It's very logical to assume the new Intel 11th gen, with new architecture, will be faster than the current generation, making it faster in gaming and thus the new gaming performance champion.
Not really rocket science here, and doesn't really matter since you can't buy a Ryzen 5000 without going through great effort, due to AMD not being able to afford their own fabs and having to beg TSMC for scraps.
This entire post is totally FUBAR lol. For one thing, TSMC actually has customer oriented business relationships, unlike a certain company that starts with "in' and rhymes with "hell".
What I'm saying is instead of halting architectural development until the 10nm process is viable they should have simply continued with new architectures at 14nm.
Heh. What makes you think they halted anything? Put it another way: does anyone seriously think Rocketlake would have been a compelling product in late 2019, or would it have been just as DOA then as in early 2021?
Heh. What makes you think they halted anything? Put it another way: does anyone seriously think Rocketlake would have been a compelling product in late 2019, or would it have been just as DOA then as in early 2021?
Sure, maybe at half the power consumption. Turns out AMD is too busy spending their engineering resources on things that matter instead of pumping every last amp they can into the package in a desperate attempt to cling onto an irrelevant benchmarketing win.
I don't understand people who think Intel releasing new architectures while stuck at 14nm would have been a bad strategy.
It's as if Zen3 using the same 7nm as Zen2 didn't happen or something. Now answer this, if Zen3 can get a solid performance increase over Zen2 while still having decent efficiency gains, one and a half year later, why Intel trying to achieve something similar at 14nm would have been a bad idea? and no, not talking about backporting, what I'm saying is to design a new architecture knowing 10nm is broken, instead of refreshing for 5 years.
I'm tired of this 10nm thing, as if all the issues for Intel come from inability to move to 10nm, which is false, the issue was tying every new architecture to a node shrink, which is absolutely not a necessity, as evidenced by Zen3, Apple A13, ARM cores and more that prove that you can get solid gains on the same node, instead of stupid refreshes.
Krzanich's major strategic mistake was not doing that backport to 14nm starting in 2017/2018 when they knew 10nm might be delayed. In fact, they probably should have done it whether they were going to be delayed or not. It's called a fall-back plan, hedging your bets, an "insurance policy".
Apple has been porting Darwin / MacOS / OS X to ARM for years in its labs, "just in case". Apple did the same thing back in the days of PowerPC, keeping an x86 port handy. A few hundred million to do that each year for a company like Intel would have been nothing.
Now they are racing against the clock to see if they can get Rocket Lake out before AMD can get real supply of Zen 3 into the market, all AMD really has right now is bragging rights from some chip sales via NewEgg or mindshare.de to about 2% of the market. I figure Intel has about 3 months. Rocket Lake has to be good.
Put it another way: does anyone seriously think Rocketlake would have been a compelling product in late 2019, or would it have been just as DOA then as in early 2021?
The power consumption would have been bad, but an 11900k would have been better than the 9900ks for the intended target market.
Rocket Lake has already been banished from the mobile space, and Intel isn't launching any 4c-or-smaller variants either. It's solely an enthusiast's part for the DiY sector. With an 8c limit, you know it isn't for "serious" productivity workloads either.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
Facebook
Twitter