Wednesday, January 1st 2020
Core i9-10900K up to 30% Faster than i9-9900K: Intel
Intel's upcoming Core i9-10900K desktop processor is up to 30 percent faster than the Core i9-9900K according to the company, which put out a performance guidance slide that got leaked to the web. Based on the 14 nm "Comet Lake-S" silicon and built for the new LGA1200 platform (Intel 400-series chipset motherboards); the i9-10900K is a 10-core/20-thread processor that leverages increased TDP headroom of 125 W to sustain higher clock-speeds than 9th generation "Coffee Lake Refresh," while also offering a 25% increase in processing muscle over the i9-9900K, thanks to the two additional CPU cores.
In its performance guidance slide, Intel shows the i9-10900K scoring 30% more than the i9-9900K in SPECint_rate_base2006_IC16.0. There's also a 25% boost in floating-point performance, in SPECfp_rate_base2006_IC16.0, which roughly aligns with the additional core count, as both these tests are multi-threaded. Other noteworthy results include a 26% gain in Cinebench R15, and 10% in SYSMark 2014 SE. In tests that don't scale with cores, Intel appears to rely entirely on the increased clock-speeds and improved boosting algorithm to eke out performance gains in the low-to-mid single-digit percentages. Intel is introducing a new clock-speed boosting technology called Thermal Velocity Boost, which can dial up clock-speeds of the i9-10900K up to 5.30 GHz.
Sources:
ITHome, Tom's Hardware
In its performance guidance slide, Intel shows the i9-10900K scoring 30% more than the i9-9900K in SPECint_rate_base2006_IC16.0. There's also a 25% boost in floating-point performance, in SPECfp_rate_base2006_IC16.0, which roughly aligns with the additional core count, as both these tests are multi-threaded. Other noteworthy results include a 26% gain in Cinebench R15, and 10% in SYSMark 2014 SE. In tests that don't scale with cores, Intel appears to rely entirely on the increased clock-speeds and improved boosting algorithm to eke out performance gains in the low-to-mid single-digit percentages. Intel is introducing a new clock-speed boosting technology called Thermal Velocity Boost, which can dial up clock-speeds of the i9-10900K up to 5.30 GHz.
144 Comments on Core i9-10900K up to 30% Faster than i9-9900K: Intel
Intel is giving you marginal higher single core performance IF you go out and buy a high end motherboard and an expensive cooling solution to throw on the chip. And of course we have to conveniently ignore the much higher power consumption and the almost certain possibility to learn about more Intel security vulnerabilities in the near future.
Still worth a giggle I guess :)
Threading and frontend should not be that much different. Intel does appear to have more load/store but actual compute is the key to SMT. My bet would be on AMD's much cleaner execution stage setup. They have more execution units that are far more specific in what they do. Intel has a couple powerful ones but the management to get the stuff efficiently into these must be pretty crazy.
- Zen2 has 4 INT ALUs (with slightly varying capabilities) and 4 FP ALUs (2 FMA/FMUL and 2 FADD) plus some AGU/Load/Store stuff.
- Coffee Lake has 4 execution pipes with strange range of capabilities. One that can do everything (INT, FP, both INT/FP Vector etc), second can do a little less and remaining two are largely INT stuff.
What AMD said and went for with Bulldozer with regards to execution units sounds true here - FP is less critical than INT. The moment FP is used and used heavily, Intel's scheduler will need to make hard choices. This is simplified but while Zen can do 4 INT and 2 FP instructions (or 4 in case of 2 FMA and 2 FADD) at once, Coffee Lake has to choose whether it does 4 INT, 3 INT+1 FP or 2 INT + 2 FP.If you look at the link, 9900K R0 stepping has everything but TAA and V3a already fixed in hardware. P0 stepping lacks MDS fixes.Sure they did. John_ even pointed out exactly that.
25% performance increase is the baseline expectation (10 cores is 125% of 8 cores). Some of the tests rely on single/fewer cores resulting in no real performance increase.This is probably the first time the longer-term platform argument is not true with Ryzens. AM4 is expected to get Ryzen 4000 but AMD has not even given hints about what happens after that. They have said AM4 will last to 2020. Intel's sockets have been very predictible - 2 CPU generations per socket. Which in a weird way puts whatever socket 10-series will come out for, on par with AM4 at this point in its life cycle.
So you go out and buy an Intel processor because reviews show you that the top Intel CPU, under the best cooling solution, on an expensive high end motherboard in a case with plenty of air flow, while paired with an RTX 2080 Ti is 5%-10% on average faster than the 3900X at 1080p? And that's while ignoring power consumption because power consumption is important only when AMD CPUs are less efficient.
What about a typical system with a mid range motherboard, mid range air cooling, probably an RTX 2070S or 5700XT at best, in a case with mediocre air flow running games at 1440p? What is the difference in games there? And can you really see it and feel it in games?Yes they will see it and they will say "10 cores much faster than 8 cores"...... IN EVERYTHING. ;)
AMD power consumption wins are at 8 cores or more. 6-core CPUs are not doing too well. There is a baseline higher power consumption of +10W (compared to both Intel's 9000 series and Ryzen 2000 series) and scaling kind of helps but not completely. If I remember correctly in TPU CPU reviews, 8700K puts up hell of a fight to 3600/3600X that includes power consumption/efficiency. It is worth noting here that 8700K does indeed have all the software mitigation problems right now.
Intel is just struggling right and (and regurgitate same things, from struggling?) and we as users hope to get it together and show some real competition soon.
2011-2012: s1155 - Sandy Bridge (2000) and Ivy Bridge (3000)
2013-2014: s1150 - Haswell (4000) and Broadwell (5000)
2015-2017: s1151 - Skylake (6000) and Kaby Lake (7000)
2018-2019: s1151v2 - Coffee Lake (8000) and Coffee Lake refresh (9000)We have no price reference other than 9000 series for now. Intel is probably screwed in 8-10 core space primarily due to power issues but 6-core CPUs are competitively in a pretty nice position if priced correctly. 9400F is simply put the best bang-for-buck gaming CPU today, there is no contest. What it lacks is futureproofing both due to lack of HT and platform. 10400F can change both of these. 9600K is currently at ~220€ with the same problems. Again, 10600K can get HT and competitive enough platform and should perform close enough to 3600/3600X in production while being faster in games.
We enthusiasts like our powerful CPUs but majority of the people do not need or want an expensive CPU. Today, the best bang-for-buck CPU - in new prices, not accounting for the EOL sales of Ryzen 2000 series - is Ryzen 5 3600. This is all about the 200+-50€ price range that has been midrange for a long time and sells a lot. A few years back Intel got a lot or criticism for 300€ i7 CPUs and much more justified criticism about the 500€ 9900K. The market has not changed but our perception of it has, making new price points seem normal. This really is not normal for a layperson who wants a good enough or bang-for-buck computer.
Not so sure about that 30% performance increase will come for free, especially with that 250 watt max TDP. I'm not into watercooling of any kind, and I already consider my 3700X as hot.
Still, I bought the AMD because I believe Intel needs a competition. I'm in the position now hoping that Intel will clean up their mess and be the leader in computing solution once again... Because AMD now have the capability to match it, which in the end, we as the consumer will win with the available solutions.
You can look at the 3900X Review, it got beaten over all games relative Performance by my old 8700K. But when it comes to pricing a german Hardware dealer asks the following Prices:
- Ryzen 3700X 325,90€/Intel i7 9700K 379,90€
- Asrock Taichi X570 305,90€/Asrock Z390 Taichi 232,45€
The Ryzen System costs you 631,80€/The Intel System costs you 612,35€
So the Intel System is cheaper and it beats the Ryzen System in games by nearly 10%. So if your System is a gaming rig, what would you choose?You can cool it with a good air cooler or an AiO, no custom Loop or similar necessary here. 10% gaming benefit over all cards
It’s really that simple!
When it comes to compression and especially decompression Ryzen beats Intel by a lot. The bigger the archive being decompressed for example the bigger the difference. Few ms is far from true.For gaming 3600 is good enough and pretty much as fast as 8700K.You forget to factor in a cooling solution for the Intel system. Ryzen already includes a cooler. Include even a 20€ budget cooler and you have price parity with X570 being a more modern platform. Plus thanks to not limiting compatibility to one chipset it's possible to get B450 for less 1/3rd the price of X570 if the user does not need PCIe 4.0 and put that saving toward a class higher GPU instead.
So which one would i buy if i had to?
9700K+Z390+Cooler+ let's say 1660S.
Or 3700X+B450+5700.
or even 3600+B450+2070S
Assuming the price will be the same the AMD system will be faster thanks to a faster GPU.
People argue over how much a CPU affect framerates. Guys - a GPU is still the #1 when it comes to gaming.
So if AMD offers good enough performance at significantly lower prices that allows the user to get a faster GPU.
9700K and 9900K make sense only if you already have the fastest card and want those few extra frames at the top end.
For most users AMD makes tons more sense.
Besides would you buy Ryzen - a CPU that is faster at everything except gaming (losing 5-20% in edge cases in 1080p with 2080Ti) or would you buy Intel - a CPU that is great only at gaming (losing 20-100% in everything else that needs cores).
To me the choice is clear - AMD. A more user friendly platform with upgrade options that excels at nearly everything instead of being great at only one thing.
Intel is really in a sad state at the moment. Their 10nm issues are truly unfortunate, I'll give them that, but it seems clear to me they got too comfortable with their lead over AMD when I think about how many iterations of Skylake we've seen at this point. The Zen architecture, and the growing list of *lake security flaws have been around for a while now... and so far we've been shown nothing more than more *lake chips. I don't think they had any plans to move on from *lake any time soon...
Intel's counterpart to IF is UPI. By the way, both are used in pretty much the same way in multi-socket/CPU systems. There is a reason EPYCs top out at 2 socket configurations and while marketing says this is market optimization and best most optimal section of the market, the actual reason is purely technical - with the amount of dies EPYCs use there are not enough to facilitate the links between separate CPUs in optimal manner.
what Intel is developing with their EIMB is different,separate modules working together as one cpu.
- IF and UPI are interconnects, basically a logical/electrical specification of the connection. Whether IF is more universal has little bearing in this context.
- EMIB is a physical, packaging solution. using a silicon substrate isntead of PCB traces for connecting two dies which has multiple advantages, primarily concerning power and speed. EMIB is in the middle of tracing over PCB and having a full interposer.
and same type too.tweaked 14nm vs tweaked 7nm.
more for amd to squeeze out from 7nm+ than for Intel to gain from another 14nm revision,same as a better IMC on ryzen 4000 would benefit IF speeds.
Still,the path is mostly the same.was that developed for core+vega apus only ? seems wasteful.
It is likely that APUs with Vega were a kind of mass-production test of sorts. In these APUS, EMIB was used for GPU-HBM connection, meaning connecting the very wide memory bus from GPU to HBM without having a full interposer.