Leaked Intel document shows Core i9-10900K with 30% performance gains over Core i9-9900K

[midian182]

Something to look forward to: We know that Intel’s 10th-generation Comet Lake chips are on their way, and they include the flagship Core i9-10900K. Now, what’s purported to be an internal document from Chipzilla has shown some of the official performance gains of the ten-core CPU.

The document comes from MebiuW, a Weibo forum user, via Tom’s Hardware. The main takeaway is that the 10900K is up to 30 percent faster than the 8-core/16-thread Core-i9 9900K in threaded workloads.

The upcoming chip was tested with single- and multi-core workloads in benchmarks, including SYSmark, SPEC, XPRT, and Cinebench. The biggest performance increase over its predecessor is seen in SPEC, where the 10900K has a 30 percent uplift, while there’s a 26 percent increase over the 9900K in Cinebench R15, thanks to its two extra cores.

Single-threaded performance increases aren’t as large, with the 10900K just 3 percent faster than the 9900K in XPRT. The new CPU uses the same 14nm process, though its maximum 5.3GHz core clock is higher than the older chip’s 5.0GHz.

While the 10900K has a 125W TDP, the results also list a 250W TDP. The latter is the chip’s PL2 power state, which is required for all cores to hit the turbo frequency.

Last week, slides that revealed Intel’s entire 10-gen series leaked online. They showed the 10900K with a base frequency of 3.7GHz and a boost of 5.1GHz, reaching 5.2GHz with Turbo Boost Max 3.0. And by using Thermal Velocity Boost, which requires a high-end cooling solution, the chip can hit a single-core frequency of 5.3GHz and reach 4.9GHz on all cores. It also has 20MB of cache, Enhanced Core and Memory Overclocking, and Active Core Group Tuning, while supporting DDR4-2933, Wi-Fi 6 and 2.5G ethernet.

With these results, it appears that Intel’s 10th-gen processors could face a challenge from AMD’s Zen 2-based Ryzen 3000, which offer better multi-threaded performance, if inferior single-core performance. And with the future Zen 3/Ryzen 4000 rumored to feature massive IPC gains, Intel could be looking at some tough times ahead.

We still don’t know when Intel's new CPUs will arrive but expect to learn more at CES.

Permalink to story.

https://www.techspot.com/news/83374-leaked-intel-document-shows-core-i9-10900k-30.html

 

[hahahanoobs]

"While the 10900K has a 125W TDP, the results also list a 250W TDP. The latter is the chip’s PL2 power state, which is required for all cores to hit the turbo frequency."

Can we put the whole "misleading TDP" discussion to rest now? Please?

 

[scavengerspc]

Cant wait to see them try to stuff the desktop version of that in a laptop.
But then I said the same thing about the 9900K and now I have one so, who can say.

 

[jgraham11]

So this chip will be the same efficiency at the 9900k, 25% more power, about 25% more performance... so its still approx. 85% behind the Ryzen 3950.

I wonder if this chip will be shipped with known CPU vulnerabilities: just like Kaby Lake, and Coffee Lake, and Comet Lake, oh wait that's this one!

Did anyone notice in the disclaimer that not all security patches are installed for this testing, just another Intel contrived benchmark... And where is the FCC required disclaimer that Intel is legally required to post on all of their benchmarks because of their prior contrived bechmarks:
"Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchase, including the performance of that product when combined with other products.“
- Shortform, "our benchmarks cast our CPUs in the best possible light and don't actually reflect real world performance, and as a result, you shouldn't trust us, you have been warned!"

 

[trparky]

"While the 10900K has a 125W TDP, the results also list a 250W TDP. The latter is the chip’s PL2 power state, which is required for all cores to hit the turbo frequency."

Can we put the whole "misleading TDP" discussion to rest now? Please?

I'm glad that Intel is finally posting a TDP value that's semi-close to what you'll really see with their chips when you put them under load. However, it's a bad thing; it's showing that just how inefficient their chips really are and is showing just how badly they really need to get off of 14nm. Too bad 10nm is a complete flop and they don't have an answer until (some say) 2023.

So until then, get ready for chips that run hot as all get out!

 

[Black Paper]

Will the Core i9-10900K come with a chiller too?

 

[hahahanoobs]

I'm glad that Intel is finally posting a TDP value that's semi-close to what you'll really see with their chips when you put them under load. However, it's a bad thing; it's showing that just how inefficient their chips really are and is showing just how badly they really need to get off of 14nm. Too bad 10nm is a complete flop and they don't have an answer until (some say) 2023.

So until then, get ready for chips that run hot as all get out!

What TDP Intel or AMD put on the box didn't matter in the first place. 5mins of research will tell you what cooling is required from people that have tested/reviewed whatever CPU/GPU you're looking at. Most will use the stock cooler and it's a perfectly good cooler for 80% us. TDP is an estimate. Just as we buy PSU with ~100W of headroom, we do the same with our CPU cooling. No one goes out looking for "95W coolers" then returning them because they can't handle anything more than that. TDP has never made buying a cooler with the performance you wanted an issue. Most consumers probably barely push their chips 45W for what they do on a computer, but Intel has to put something on the box to give users a general idea. They have no idea what you're going to be doing.

 

[hahahanoobs]

So this chip will be the same efficiency at the 9900k, 25% more power, about 25% more performance... so its still approx. 85% behind the Ryzen 3950.

10900 is the entry i9 with 10 cores and the 18 core 10980XE was just reviewed.
Why would you expect a 10 core part to go up against 16 cores?

 

[Rock Dirty]

"250W TDP"

Umm wut?

250? "Black Paper" ask if it came with a chiller. I'm thinking more along the lines of a vat of liquid nitrogen.

Intel, being Intel, will probably require you to buy a new motherboard for this beast also which will need six layers and cost as much as a used car.

 

[GreenNova343]

Hmmm, let's look at the math here:
-- 30% more performance on multi-core tests
-- 25% more cores/threads than its predecessor
-- 1.3 / 1.25 = 1.04, or about 4% improvement after accounting for the additional cores
-- article says the clock rates are roughly 3% higher than the predecessor
-- 1.04 / 1.03 = 1.00971

So, either they managed to maybe get 0.97% more IPC out of it, or the "30%" is the result of a small bit of rounding. And since the power usage also went up 25%, they didn't make it any more efficient in power usage.

Sounds like "OK' to "meh" news for high-end non-HEDT desktop systems, & "meh" to "why, God, why waste our time" for gaming systems...

 

[kevbev89]

"While the 10900K has a 125W TDP, the results also list a 250W TDP. The latter is the chip’s PL2 power state, which is required for all cores to hit the turbo frequency."

Can we put the whole "misleading TDP" discussion to rest now? Please?

Wait, can you explain this?

 

[Morphine Child]

Hmmm, let's look at the math here:
-- 30% more performance on multi-core tests
-- 25% more cores/threads than its predecessor
-- 1.3 / 1.25 = 1.04, or about 4% improvement after accounting for the additional cores
-- article says the clock rates are roughly 3% higher than the predecessor
-- 1.04 / 1.03 = 1.00971

So, either they managed to maybe get 0.97% more IPC out of it, or the "30%" is the result of a small bit of rounding. And since the power usage also went up 25%, they didn't make it any more efficient in power usage.

Sounds like "OK' to "meh" news for high-end non-HEDT desktop systems, & "meh" to "why, God, why waste our time" for gaming systems...

I'm pretty sure they will squeeze few extra frames in most games, and also some extra productivity numbers due to extra cores. No intel fanboy will care about consumption or heating or anything else, as long as BFV runs at 5 extra frames...

I am also curious about the price...

 

[jgraham11]

10900 is the entry i9 with 10 cores and the 18 core 10980XE was just reviewed.
Why would you expect a 10 core part to go up against 16 cores?

The comparison is for technological capability: AMD chips can fit 16 cores (32 threads) and all that performance in less of a thermal envelope than the 9900k (16 threads) and now much less than 10900k (20 threads).

AMD's 16 core is going up against the 10980XE and beating it in most cases, all while consuming approx. half the power and not completely tanking in games because of its mesh interconnect topology. And not having to pay for Intel's expensive high end desktop motherboards...

 

[hahahanoobs]

The comparison is for technological capability: AMD chips can fit 16 cores (32 threads) and all that performance in less of a thermal envelope than the 9900k (16 threads) and now much less than 10900k (20 threads).

AMD's 16 core is going up against the 10980XE and beating it in most cases, all while consuming approx. half the power and not completely tanking in games because of its mesh interconnect topology. And not having to pay for Intel's expensive high end desktop motherboards...

Majority of conusmers don't care about who has the lowest power consumption.

 

[JimboJoneson]

Keyword: "Projections" ... Does that mean that this is their expectations ... or did they actually not run any benchmarks?

Edit: I guess its part of their security patch disclaimer, lol. From that I assume these are unpatched results.

Overall, this doesn't look too great at all ... exactly as expected from adding 2 cores and tweaking the boosting a little. It looks like the light threaded stuff barely gets any increase, so I doubt we'll see much gains in gaming. But people, you generally don't game with a 2080ti and settings that bottleneck the CPU, so don't sweat it.

Even if one doesn't care about power draw, one might care about how much is needed to spend on cooling when using all 10 cores, or how much fan noise you are willing to put up with ...

 

[JimboJoneson]

...
Intel, being Intel, will probably require you to buy a new motherboard for this beast also which will need six layers and cost as much as a used car.

New motherboard(s) and socket(s) is confirmed. But it looks like they may have to split out the top end, due to to the massive power requirements needed for the higher end parts. Another leak showed that there might be two separate sockets altogether in this division (LGA1200, and LGA1159).

If true, that means that a lower end CL CPU won't fit in a high end comet lake mobo and vice versa - so no upgrading within the same CPU series from low to high end without buying a new mobo as well.

Me "You can't possibly get worse in the motherboard compatibility department"
Intel: "Hold my beer"

Its just speculation though ... as to what exactly this LGA1159 comet lake socket will be for ... take it with a grain of salt.

 

[PatrickEagle]

STAR CITIZEN is my favorite game and the only one that will benefit the most out of that CPU. That game is coded to utilize as many CPU cores/threads as available as well as multi-GPUs, and also requires SSDs bandwidth. So the combination of high core count and single thread performance to get the most out of my Intel Optane 905P SSD 4k random read/write speeds will be perfect for my needs. And Optane only work with latest Intel motherboards.

 

[JimboJoneson]

"While the 10900K has a 125W TDP, the results also list a 250W TDP. The latter is the chip’s PL2 power state, which is required for all cores to hit the turbo frequency."
...

But what does this mean? The sentence seems to imply that 250w is "required" for all cores to hit turbo. That makes zero sense.

Uncannily, 205W as listed in the article's document, for the 9900k is exactly what a stock 9900k will pull at the socket under full AVX load. The number then likely represents that maximum draw the CPU will allow before throttling the boost - allowing to maintain that number as a maximum. The sentence from the article seems to imply the opposite.

So expect that the 10900k will pull 250w under extreme loads before throttling all boosting - this lends to the reasoning that the platform may end up with two separate sockets (as I posted above) -- to keep pricing on mobos reasonable for the lower end.

Note that this is the first time most of us have ever seen that PL2 TDP number on Intel CPUs ... so the "95w" was a bit misleading, but with the extra "max" TDP of 205w thrown in the mix, everything is about exactly right, and with both numbers included, its actually more accurate than AMDs TDP numbers - which at least for Zen2 have been a bit out in left field.

 

[neeyik]

And Optane only work with latest Intel motherboards.

As storage, Optane will work on any motherboard that has the required connector (I.e. PCI Express or M.2). Specific Intel boards are required for it to be used as Optane memory.

Hmmm, let's look at the math here:
-- 30% more performance on multi-core tests
-- 25% more cores/threads than its predecessor
-- 1.3 / 1.25 = 1.04, or about 4% improvement after accounting for the additional cores
-- article says the clock rates are roughly 3% higher than the predecessor
-- 1.04 / 1.03 = 1.00971

So, either they managed to maybe get 0.97% more IPC out of it, or the "30%" is the result of a small bit of rounding. And since the power usage also went up 25%, they didn't make it any more efficient in power usage.

It's possible the newer chip can maintain a higher clock rate across all cores, compared to the 9900K, so with more cores and higher clocks, one can see how the SPEC base2006 results are up to 30% higher.

 

[neeyik]

Slightly off topic, but if anyone has an Intel processor and they want to see what power consumption it's running at full load, then use Intel's own CPU diagnostic tool along with OCCT.

The former is very good at loading up the processor, and the latter can display power consumption in real time. For example, my i7-9900K hits 160 W (BIOS set to allow this) in CPUload test (which uses AVX2).

 

[JimboJoneson]

As storage, Optane will work on any motherboard that has the required connector (I.e. PCI Express or M.2). Specific Intel boards are required for it to be used as Optane memory.


It's possible the newer chip can maintain a higher clock rate across all cores, compared to the 9900K, so with more cores and higher clocks, one can see how the SPEC base2006 results are up to 30% higher.

How long they can maintain their boost clock for is also a consideration with Intel CPUs and certain benchmarks. Any tweaking they did there can make a fair bit of variation in benchmarks as well.

 

[JimboJoneson]

Slightly off topic, but if anyone has an Intel processor and they want to see what power consumption it's running at full load, then use Intel's own CPU diagnostic tool along with OCCT.

The former is very good at loading up the processor, and the latter can display power consumption in real time. For example, my i7-9900K hits 160 W (BIOS set to allow this) in CPUload test (which uses AVX2).

A curiosity ... do you know if the numbers that HWMonitor produces are accurate, or if the results line up with the Intel tool?

 

[neeyik]

A curiosity ... do you know if the numbers that HWMonitor produces are accurate, or if the results line up with the Intel tool?

Well I just ran the AVX2 test in OCCT (8 threads, ran for 90 seconds) and used OCCT itself, AI Suite 3, HWinfo64, and Intel's own power consumption tool to monitor the total package consumption.

OCCT, Intel, and HWinfo64 all reported 150 W; AI suite 3 suggested 83 W was being consumed.

 

[bandit8623]

4 more threads and 30%? meh not good.

 

[Irata]

"While the 10900K has a 125W TDP, the results also list a 250W TDP. The latter is the chip’s PL2 power state, which is required for all cores to hit the turbo frequency."

Can we put the whole "misleading TDP" discussion to rest now? Please?

Depends how it‘s marketed and shown in reviews. I have a very strong doubt that Intel will show the 250W in any ad or that this will appear next to the CPU‘s name in benchmarks.