I agree, they have had a hard time entering even when they had competitive products, they are trying to put more focus to one place at a time instead of spreading all the talent to thin to make an impact. Hopefully as they gain traction in other spaces we'll see them bring more competition to mobile as well.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Intel 10nm/7nm CPU/GPU Rumor Thread : Cannonlake, Icelake, Tigerlake, Sapphire Rapids, etc.
STEM
Gawd
- Joined
- Jun 7, 2007
- Messages
- 614
Have to love Intel marketing. Whoever is in charge of their marketing department needs to get fired Today along with the entire team. Start fresh on Monday at 9:00 am 
Did you even bother reading my post?
I have been reading this since Bulldozer. The next product is always the game changer.
GoldenTiger
Fully [H]
- Joined
- Dec 2, 2004
- Messages
- 29,643
LOL, it's always the next one with amd. Yep, keep waiting.... Waiting... Waiting...
Classic, I called you out for not bothering to read the link due to your comment, so you imply I didn't read your 1 sentence of a post that was implying the link only covered rendering. Guess it was both, you didn't read it and your trolling? I have no care, I even stated AMD gets trounced in some games (although, majority they are pretty even, but behind on average). But to imply that the only time they are useful is in rendering is misleading at best.
I mean, I hope they come out with good products but I don't preach that their next release is going to stomp Intel or anything. But sometimes they come out with a great product (check out Rome for a good example). I think people were getting tired of no competition and where wishing way to hard that AMD had some magic up it's sleeve. Ryzen was a pretty good attempt and has made pretty good inroads. A lot of people have stopped waiting and started buying. Hard to find numbers everywhere, but in many markets AMD is vastly outselling Intel (again, limited to certain markets because it's hard to find #'s in some markets). Either way, it's a good indicator. Now if they can come up with something for laptops...
https://www.extremetech.com/computi...mily-is-dominating-sales-at-european-retailer
https://www.extremetech.com/computi...igh-end-ryzen-3000-cpus-still-in-short-supply
https://www.techspot.com/amp/news/81202-amd-cpu-sales-move-ahead-intel-japan.html
STEM
Gawd
- Joined
- Jun 7, 2007
- Messages
- 614
History rhymes more often than it repeats. I didn't say AMD will beat Intel. I said that they already did and next year they will dominate Intel with their Zen 3 architecture. And this isn't because Intel doesn't have capable engineers, it's more like upper management, their board and CEO can't get their heads out of their asses. Here's a hint: for once, give consumers and businesses what they ask for and need instead of fucking with them and fleecing them to no end.
5150Joker
Supreme [H]ardness
- Joined
- Aug 1, 2005
- Messages
- 4,568
I for one am happy AMD is competing again in the cpu market as the intel monopoly was no good for anyone. Hopefully they keep the momentum going and don't let up because intel will push back. Those who think intel is dead are ignorant of the past.
And it happens on both on CPU and GPU. How many times the next red team GPU was the one that would kill Nvidia?
I take that as a "no".
AMD didn't beat Intel. Zen2 is behind Intel in both IPC and frequency. And Icelake is here with a nice IPC bump.
Intel's only problem is the delay of the 10nm node. The foundry engineers were too aggressive and got very difficult problems by working at so small scales and with novel materials. Recall that the 10nm node used in the half-broken Cannonlake has a transistor density of about 100 MTR/mm², whereas the 7nm node used by AMD for Zen2 has a density of only 66MTr/mm². So the Intel 10nm node is about 50% more dense than the 7nm node from TSMC. That is the problem. Intel foundry engineers were too aggressive and they are paying with a huge delay.
Funny as I summarized your post for you... but sure, ignore the obvious, ignore what the benchmarks say, ignore what other say, and then point at others and say they aren't reading what you have to say.
Meh, beat is a relative term, chose your benchmarks to prove you win, just as someone else can do. They are competitive and depending on the workload one can be better than the other. They are behind IPC in a few workloads (mostly some games and some less threaded loads) and AMD is up in IPC in other work loads. It isn't a 100% win, but they win in IPC in many instances. To say they are behind Intel in IPC is again, leaving a lot of information out and cherry picking. I know you and facts don't like to agree (or at least the ones that don't support your claims) but if you take off your bias lenses, you'll see you are making statements with 1/2 truths to rebut others who are making 1/2 truths... that makes you just as bad as the person your arguing with, if not worse because you know they are doing it and to prove them wrong you are doing it.
In multithreaded rendering, AMD wins as normal. Single threaded Intel is slightly ahead. In the few games they tested, Intel was ahead by a very slim margin to a slightly larger margin depending. Was trying to find another review I read a while back that had a bit more indepth review, but having a hard time.
https://hothardware.com/reviews/amd-ryzen-9-3900x-vs-core-i9-9900k-ipc-shootout
IPC bump with icelake... great, when can I buy one for my desktop? Maybe next year? Maybe 2021? Great, apples to oranges comparison, that's useful to prove your point. I don't think anyone is arguing AMD needs to do better in the mobile market, so this argument means what to this discussion? Cool, if I was in the market for a mobile chip, then this would be on my list to check out. I know it's difficult to actually find real information (google searches can be complicated, I know), but please at least pretend you're not an Intel fanboy (or just wear it proud and loud and state it in your responses). Otherwise understand it's not as black/white (cut and dry) as you are implying.
I am talking about average IPC. Average obtained by testing 32 applications (no games).
I have no idea of when desktops will be ready. Icelake mobile is ready. It is shipping to OEMs and partners.
Got a link for that first one? I haven't been able to find a really good comprehensive comparison. Everything I have found said Intel wins in games in general (some are within 1% others quite a bit more, and some AMD wins) and AMD goes back and forth in most non gaming (and non single threaded) apps..
https://www.pcgamesn.com/amd-intel-zen-2-coffee-lake-benchmark-performance-comparison
https://www.thefpsreview.com/2019/07/07/amd-ryzen-9-3900x-cpu-review/9/
I'll reserve to much judgement on this one until it actually matters (to me, obviously if you're shopping for a laptop, then it applies).
Quick Edit:
After doing a bit more reading, it seems ice lake isn't much better than whiskey lake for normal tasks, graphics are improved over the 610-620, would like to see it compared to Iris Plus 655. It also seems like it is a bit more lax with limits as well, boosting much higher for a little while when a task starts and then drops off. This could easily explain the (normally) minor differences from previous generations.
From Anandtech: "As for Ice Lake itself, our results lean towards Ice Lake outperforming Whiskey Lake, if only by a small margin."
Extremetech: "Gains for Ice Lake relative to Whiskey Lake are fairly anemic, though this can vary dramatically depending on which Whiskey Lake system you own now."
PC Mag: "We looked at one Ice Lake-based laptop sample, with the second-from-top processor option installed, in a (presumably, given its size) thermally forgiving chassis."
Keep in mind, these are all from Intel setup systems for testing/comparing, have any independent reviews come out yet? I'm interested in what kind of real world gains they have gotten (and possibly other models).
Last edited:
Funny, your go to source for IPC comparisons on 1st and 2nd gen Ryzen says otherwise, it's strange that you didn't link to stilts review this time.....
https://www.overclock.net/forum/10-...nical-matisse-not-really.html#/topics/1728758
"For the first time in over a decade, AMD has reached IPC parity with Intel.
On average, based on the results of 32 individual workloads Zen 2 even manages to provide slightly higher average IPC than Coffee Lake-S Refresh.
Thanks to it AVX-512 resources Skylake-X manages to stay a head in this test suite however, not by a large margin."
Guess I'll have to deal with the performance hit on all those AVX-512 desktop apps.
edit: Good to see you back though, it was weird that you disappeared during the Rome launch.
STEM
Gawd
- Joined
- Jun 7, 2007
- Messages
- 614
I'd like to see how the 9900K performs in all those cool AVX-512 desktop apps versus the 3900X
In other news: I find it hilarious how far the brand loyalty for Intel goes. Brand loyalty and fanboyism suck big time. However, being brand loyal to Intel is like being a triple-masochist. First, you enjoy the torture provided the performance hit from all those layered security patches, then you enjoy the uncertainty of the unpatched security flaws, and lastly, you enjoy paying more for less. That's triple masochism.
Some people are into that kind of thingI'd like to see how the 9900K performs in all those cool AVX-512 desktop apps versus the 3900X
In other news: I find it hilarious how far the brand loyalty for Intel goes. Brand loyalty and fanboyism suck big time. However, being brand loyal to Intel is like being a triple-masochist. First, you enjoy the torture provided the performance hit from all those layered security patches, then you enjoy the uncertainty of the unpatched security flaws, and lastly, you enjoy paying more for less. That's triple masochism.
.
It is showing a 6% IPC advantage for SKL-X, not 20%, and he tested stuff that other reviews don't even know that there exist.
https://www.overclock.net/forum/10-...nical-matisse-not-really.html#/topics/1728758
That's totally disingenuous. When you remove the outliers and compare equal platforms, cough cough skl-x has quad channel memory, they come out exactly the same.
Coming from dude that railed for the past year about not running memory out of spec to keep things equal.
Got it, I didn't realize the baseline was not what we were talking about. Makes much more sense now. I see how they came to this conclusion now...
"Thanks to it AVX-512 resources Skylake-X manages to stay a head in this test suite however, not by a large margin." It's due to one single feature that allowed certain tests to show it in a better light. This is why I was asking for links, to see how they were getting results.
And this statement from the same link says "For the first time in over a decade, AMD has reached IPC parity with Intel.', so in their eyes, they are on parity (which is basically what I've been seeing). This is more inline with other reviews that I've seen.
And it's more power efficient while doing so "As expected, Matisse provides significantly higher performance per watt than its competition, thanks to its leading edge 7nm manufacturing process". (Not that it matters for what we're talking about, just though it was neat they finally caught up in that as well).
Either way, pretty good read, thanks for the link (and correcting my mistaken self about the %'s).
edit: Looking through some of the individual, noticed there were a few outliers that are really the reason for the IPC difference, possibly uses AVX512 or something specific to that CPU because it even beats the pants off of previous coffee lake CPU's in the same test. If you take the biggest outlier out, it's a 3% difference, so that one out of 32 benchmarks makes up 1/2 the difference, lol. If you take the second outlier, the difference is less than 1%. So if you remove 2 of the 32 benchmarks you could paint an entirely different picture (which supports his conclusion they are pretty much even at this point). Even including these both, AMD won 17 of the 32 benchmarks. This is a pretty good read, but someone using this to point out AMD is still behind leaves out some basic facts and information. These aren't exactly a benchmark list that a normal user would use (the outliers, like LINPACK that shows the AVX lead Intel has).
Last edited:
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
Same shit they pulled during a64 days with sse2 benchmarks. When they can't win outright, they focus on an obscure instruction set which is rarely used.
The reason that someone wasn't here during Rome or Zen2 was because it's more lucrative shilling other sheep facing platforms during a launch. It always happens with him then he slinks back here when other places are burnt out from the shilling. Just like that 'friend' who always used your shit and never repaid you in any way, then always left town for a while once people got sick of them.
This is why I was asking for links, I could guess what was going on, but now I know for sure
. This is also why I provide links when I post something, so people can verify and find mistakes (which can and do happen) and I can learn from it. I don't care what it is (or if I'm the one wrong), but I prefer to have the actual facts to go from.After reading the full breakdown, if you remove the 2 apps from the 32 that used AVX to give Intel a rediculous look, they are pretty much even. So, 2 apps out of 32 (Average) where accounting for over 5% of Intels lead.. saying this another way, 6% of the apps (2/32) attributed to 83% of intels lead... wonder if this is how Intel is going to give 2x the $/perf of their next iteration, cherry pick a benchmark that uses a new opcode, that give almost 0 real world benefit to 99.99% of their users.
So when comparing Rome to Xeon we have to disable memory channels because Xeon has only six?
It is parity with CoffeeLake, not with Skylake-X.
Let us remove anything where Intel wins. Got it!
So when comparing Rome to Xeon we have to disable memory channels because Xeon has only six?
In the context of determining IPC. Yeah.
OutOfPhase
Supreme [H]ardness
- Joined
- May 11, 2005
- Messages
- 6,574
I'd suggest that if you're attempting to determine an architecture's aggregate IPC, you shouldn't disable parts of that architecture.
The whole idea of normalizing something is to remove factors that bias what you are trying to measure. Measuring a single core on a platform that is designed to support magnitudes more, will most likely skew your results.
OutOfPhase
Supreme [H]ardness
- Joined
- May 11, 2005
- Messages
- 6,574
No, you don't "normalize" something by artificially changing the exact metric you're trying to measure. In this case that metric is per-thread per-clock instruction completion rate. What you're calling "bias" is an advantage one architecture has (or may have), and you do quite specifically want to measure that as part of the result.
Thus, you would normalize the number of threads, and frequency. You should not disable cache, memory lanes, etc.
Again context.
This whole off ramp we've taken stems from a desire to achieve a single metric that distills down to a binary choice.
If one says intel is x IPC faster because sky-x with all its platform advantages shows that over the desktop part, they are being disingenuous. There are a myriad of factors that can contribute to this. If one ran the battery of tests with one platform having a slow (say HD) and the other having a close factor SSD (optane or pci4 mvme), the results would skew.
Also taking what I said and throwing hyperbole at it, fine disable everything inside the processor as well, is equally disingenuous.
N
. Keep on ignoring what people say and push your agenda. I was simply pointing out there were some very pronounced outliers due to a specific opcode being used and in all other benchmarks where this wasn't the case they are about equal. If you have a workload with avx512, no choice, Intel hands down. Everything else... Pretty much the same. Don't know why this is so hard to comprehend.
Lol, not even worth arguing. Yes, that's exactly what I said, you can read so well. I was trying to hold an intelligent conversation about it, but guess I need to just keep talking to myself for thatSo when comparing Rome to Xeon we have to disable memory channels because Xeon has only six?
It is parity with CoffeeLake, not with Skylake-X.
Let us remove anything where Intel wins. Got it!
. Keep on ignoring what people say and push your agenda. I was simply pointing out there were some very pronounced outliers due to a specific opcode being used and in all other benchmarks where this wasn't the case they are about equal. If you have a workload with avx512, no choice, Intel hands down. Everything else... Pretty much the same. Don't know why this is so hard to comprehend.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
This doesn't really 'compute'.
HEDT usually shows lower performance in desktop applications, if only by a hair, due to higher latencies with more cores and slower memory, in my experience. There are cases where extremely highly threaded consumer workloads scale well enough, or the extra memory bandwith helps, or the extra cache that's usually available helps, but overall, single-core performance tends to go to the desktop parts.
Now, I'd still want to see the exact parts in question compared but I don't find the use of SK-X to be disingenuous.
Newest leaks showing 14nm desktop until 2022:
https://www.techpowerup.com/260130/...top-brazen-it-out-with-14nm-skylake-till-2022
More confirmation that not only is Intel 10nm late, but it is simply trash altogether.
We always new the first iteration of 10nm would not make it to desktop in the form of Cannon Lake, then this year it became clear the next gen of 10nm in the form of Ice Lake would cut it in desktop use. Now, it looks like NO 10nm will ever go in consumer desktop and Intel will transition to 7nm Meteor Lake in 2022.
https://www.techpowerup.com/260130/...top-brazen-it-out-with-14nm-skylake-till-2022
More confirmation that not only is Intel 10nm late, but it is simply trash altogether.
We always new the first iteration of 10nm would not make it to desktop in the form of Cannon Lake, then this year it became clear the next gen of 10nm in the form of Ice Lake would cut it in desktop use. Now, it looks like NO 10nm will ever go in consumer desktop and Intel will transition to 7nm Meteor Lake in 2022.
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
Yup. Got to wonder if Intel has massaged Skylate in Comet Lake - S and / or Rocket Lake -S to eliminate vulnerabilities and the performance cost of mitigations such that the core is still competitive clock-for-clock with Zen 2 while also maintaining higher clockspeeds.
It's very surprising that Rocket Lake is Skylake and not Willow Cove. By the time RKL is released, we will have Zen 3 which should see an IPC and clock boost once again. RKL will NOT see an IPC boost and SKL sees diminishing gains after 5.0 ghz, which is true for any architecture it seems.
jeremyshaw
[H]F Junkie
- Joined
- Aug 26, 2009
- Messages
- 12,511
It does seem entirely backwards. For the purpose of a 10 core CPU, the IGP is probably among the less important parts. Why spend so much time porting over their next gen IGP to 14nm, only to replace it with 10nm or 7nm design and IP a year later?
The only guess I have is the current IGP is already very dated and probably on driver life support by now. Or they are still using AMD or Nvidia patents and want to get away from them.
First test of the 9900KS:
https://www.tomshardware.com/features/intel-special-edition-core-i9-9900ks-benchmarked
The new stepping seems to be doing wonders for power consumption and this heat.
At stock 5.0 ghz, it is using around 25% less power than the 9900k at those clocks. It looks like a 10 core Comet Lake could draw similiar or less power than a 9900k using this stepping.
The price increase may be worth it for those that want to run at 5.0ghz in sff. This is not easily done with a 9900k.
The other positive light is that the new stepping was designed to mitigate some of the vulnerabilities and did little to no impact on IPC.
https://www.tomshardware.com/features/intel-special-edition-core-i9-9900ks-benchmarked
The new stepping seems to be doing wonders for power consumption and this heat.
At stock 5.0 ghz, it is using around 25% less power than the 9900k at those clocks. It looks like a 10 core Comet Lake could draw similiar or less power than a 9900k using this stepping.
The price increase may be worth it for those that want to run at 5.0ghz in sff. This is not easily done with a 9900k.
The other positive light is that the new stepping was designed to mitigate some of the vulnerabilities and did little to no impact on IPC.
they didnt test how much performance penalty comes from these mitigation inside the hardware..
IdiotInCharge
NVIDIA SHILL
- Joined
- Jun 13, 2003
- Messages
- 14,675
Three not entirely exclusive possibilities:
- The porting work had been started before there was a solid understanding of when 14nm would be supplanted in volume by 10nm / 7nm
- The porting work was not difficult
- The new IP provides a foundation, not just from a driver perspective but from a feature perspective, that Intel finds important for the marketability of the CPUs that it will reside in
Upgrades to efficiency reduce the load for mobile and they also reduce whatever overclocking / TDP overhead that is present for enthusiasts using the IP block on top-end parts.
Upgrades to the transcode block are even more important, as improvements to speed (and efficiency, which are coupled here) as well as upgrades to codec support and output quality are useful for say video conferencing, video editing, video stream pulling and serving, which also includes various forms of streaming.
Given that one of the arguments for Ryzen is that software encoding for streaming is still desirable due to the limited quality of the encoding in most fixed-function IP blocks in GPUs to include Intel's current Skylake-generation IGPs and with Ryzen being available with 'extra' cores that are otherwise extraneous for gaming, Intel's backport of their latest IGP seems to make sense.
Last, and likely lower for most except gamers, hopefully the new GPU IP block will include support for HDMI 2.1 and HDMI VRR / DP Freesync, such that VRR may be supported on laptops and other devices that include a display panel as well as on outputs to VRR-capable displays on both interface types.
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
Agreed, this is what I've been saying for a while. Only way that crazy IPC boost claim is possible is with exploit fixes and other tweaking in specific benchmarks. I bet the average is lower.
I don't think It will catch them up everywhere to AMD IPC though, as exploits also don't slow them down everywhere..