Sunday, January 27th 2019

Intel Readies Energy-efficient 35-Watt Core i9-9900T Processor

Intel succeeded in bringing down the TDP of its 8-core/16-thread "Coffee Lake-Refresh" silicon all the way down to a staggering 35 W, from its currently rated 95 W, which in real-world usage easily exceeds 110 W, given Turbo Boost, and other performance enhancements enabled by DIY motherboards. The new Core i9-9900T achieves its TDP with a combination of significantly lower clock-speeds, and an aggressive on-die power-management system. Its nominal-clock is down to 1.70 GHz from 3.60 GHz of the original i9-9900K, while 1~2 core Turbo Boost frequency is down to 3.80 GHz from 5.00 GHz of the original. The all-core Turbo clock-speed could be as low as 3.30 GHz. Intel hasn't tinkered with the L3 cache amount, which is still set at 16 MB, and the UHD 630 iGPU retains its EU count and clock-speeds. The chip features its 4-character product code of QQC0.
Source: Tom's Hardware
Add your own comment

31 Comments on Intel Readies Energy-efficient 35-Watt Core i9-9900T Processor

#1
hat
Enthusiast
Intel measures TDP at "base" clock, not Turbo... so I would expect this chip to take a lot more than 35w while Turbo is active. I speculate this is an attempt to take a stab at AMD's efficient Ryzen chips with misleading labeling and fine print.
Posted on Reply
#2
Zubasa
hat, post: 3982348, member: 32804"
Intel measures TDP at "base" clock, not Turbo... so I would expect this chip to take a lot more than 35w while Turbo is active. I speculate this is an attempt to take a stab at AMD's efficient Ryzen chips with misleading labeling and fine print.
Yup, 35W but make sure very board comes with MCE enabled and violate the TDP.
Posted on Reply
#3
hat
Enthusiast
Zubasa, post: 3982354, member: 30988"
Yup, 35W but make sure very board comes with MCE enabled and violate the TDP.
That 35w TDP is flying out the window at warp speed the instant Turbo is active anyway... with or without MCE.
Posted on Reply
#4
efikkan
Intel has offered similar designs for years, not only as i7/i5 T variants, but also various Xeon CPUs with "many" cores, low TDP and very low clocks.
Some of these might be interesting for like a home firewall or server, but I would still probably go for a low TDP 4-core rather than a low TDP 8-core for that purpose.
Posted on Reply
#5
XXL_AI
dear chipzilla, "we don't care", take that cpu and shove it up to your ... you know.
stop keeping cpus in your factory stock and send them to the market, your chipzilla crap is way too overpriced.
Posted on Reply
#6
ZoneDymo
man, you make it run way slower and it consumes less energy? what a shock!!
wait, does that mean then if you overclock a cpu it will consume MORE energy and run hotter as well?!!!?
Wait, THATS why people have like custom water cooling loops or after market better performing coolers? it all makes sense now!
Posted on Reply
#7
lynx29
3.4ghz is non-boost speed? so at 3.4ghz it should run at 35 watts if turn off intel speedstep. honestly if the price is right, that is still a great gaming chip. my i7-7820hk cpu only runs at 2.9ghz due to overheating issues otherwise, and it plays all games on my laptop with gtx 1070 at 100hz 100 fps on medium ot high settings.

i have it underclocked to 36 watts as well. thats a lot of energy savings and most games won't be able to tell the difference
Posted on Reply
#8
hat
Enthusiast
lynx29, post: 3982413, member: 153071"
3.4ghz is non-boost speed? so at 3.4ghz it should run at 35 watts if turn off intel speedstep. honestly if the price is right, that is still a great gaming chip. my i7-7820hk cpu only runs at 2.9ghz due to overheating issues otherwise, and it plays all games on my laptop with gtx 1070 at 100hz 100 fps on medium ot high settings.

i have it underclocked to 36 watts as well. thats a lot of energy savings and most games won't be able to tell the difference
No, 1.7 is the base speed and 3.3 is the all core turbo. TDP is measured at base clock speed, which is how the 9900k gets away with being "95w" and this chip is 35w. Once Turbo is active, kiss those figures goodbye.
Posted on Reply
#9
ArbitraryAffection
1.7Ghz lol. I still think it's impressive how well Ryzen competes in efficiency even built on an objectively inferior process technology. So realistically we are looking at 50-60W usage at all core turbo? Still seems reasonable but should be sold as 65W, it's a very fine line to false advertising.
Posted on Reply
#10
Patriot
ArbitraryAffection, post: 3982439, member: 145270"
1.7Ghz lol. I still think it's impressive how well Ryzen competes in efficiency even built on an objectively inferior process technology. So realistically we are looking at 50-60W usage at all core turbo? Still seems reasonable but should be sold as 65W, it's a very fine line to false advertising.
Yeah, my guestimate is 50-65w depending on that all-core turbo final clock... something intel no longer advertises and can vary chip to chip...
You should check out anand's review of coffee lake laptop that intel is hiding in the Chinese education market... kabylake basically is better in everything that isnt avx512.
https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-and-core-i3-8121u-deep-dive-review

This will probably be the only 10nm product, what actually comes to market will be 10nm+
The only reason intel didn't cancel 10nm after years of failure is they expect the same issues on the 7nm node they are having on 10nm.
Solve on one, solve for both, time not as wasted. Currently 14nm+++ s better than 10nm.
Posted on Reply
#11
NicklasAPJ
hat, post: 3982436, member: 32804"
No, 1.7 is the base speed and 3.3 is the all core turbo. TDP is measured at base clock speed, which is how the 9900k gets away with being "95w" and this chip is 35w. Once Turbo is active, kiss those figures goodbye.
Thats how they both are doing, AMD are doing the same thing.
Posted on Reply
#12
notb
hat, post: 3982356, member: 32804"
That 35w TDP is flying out the window at warp speed the instant Turbo is active anyway... with or without MCE.
This CPU is made mainly for OEMs and they will know how to tune it to keep under certain limit they need (be it 35W or 45W). Don't worry too much. :-)
Posted on Reply
#13
ghazi
notb, post: 3982480, member: 165619"
This CPU is made mainly for OEMs and they will know how to tune it to keep under certain limit they need (be it 35W or 45W). Don't worry too much. :)
Yeah, you can always configure the BIOS to not allow the chip to exceed TDP. It'll be really funny to see how this chip performs when it actually has to stay within 35W...

NicklasAPJ, post: 3982475, member: 136240"
Thats how they both are doing, AMD are doing the same thing.
Nope, not true. AMD's chips stay within TDP.

ArbitraryAffection, post: 3982439, member: 145270"
1.7Ghz lol. I still think it's impressive how well Ryzen competes in efficiency even built on an objectively inferior process technology. So realistically we are looking at 50-60W usage at all core turbo? Still seems reasonable but should be sold as 65W, it's a very fine line to false advertising.
With Ryzen, both the arch and the node are optimized for efficiency at moderate wattage targets... with Intel, especially with all the +'s added to 14nm, the focus is just raw frequency potential and throughput, density and efficiency be damned.
Posted on Reply
#14
OSdevr
This is backwards. Many things can't be done in parallel, and this limits the performance increase in multi-core CPUs. Fewer, faster cores are always preferable to more, slower ones if the whole CPU has the same throughput and all else being equal. This seems to me like a marketing ploy.
Posted on Reply
#15
lynx29
hat, post: 3982436, member: 32804"
No, 1.7 is the base speed and 3.3 is the all core turbo. TDP is measured at base clock speed, which is how the 9900k gets away with being "95w" and this chip is 35w. Once Turbo is active, kiss those figures goodbye.
ok yeah 1,7ghz is a joke. nevermind. i will pass in full. looks like ryzen 3800x is still my next cpu after all! might as well go all in since it most likely will be my last silicon cpu.
Posted on Reply
#16
notb
ghazi, post: 3982498, member: 175630"
Yeah, you can always configure the BIOS to not allow the chip to exceed TDP. It'll be really funny to see how this chip performs when it actually has to stay within 35W...
It's made for SFF office machines, possibly for the more demanding users. I'm pretty sure it'll be fast enough.
This CPU is about getting the job done, not getting great benchmark results. :)
Nope, not true. AMD's chips stay within TDP.
No, they don't. They'll easily exceed TDP by around 10%. But that's just few Watts, so people don't care very much.
It's quite similar on Intel's side. Or at least it used to be until 8th gen.
8700K can go past TDP by as much as 30%. But it is a K CPU and Intel strongly advises you to get good PSU and cooler for these models.
8700 also seems to get past 65W, but I haven't seen decent tests, because - sadly - non-K CPUs don't get that much attention from reviewers.

Well, it's a price you pay for single-core performance. And Intel's 8th gen has lots of it.

The main problem here is efficiency and heat distribution. Heat increases faster than power draw (i.e. CPUs are less effective at high frequencies). That's the main problem with current lineup of the Blues.
With Ryzen, both the arch and the node are optimized for efficiency at moderate wattage targets... with Intel, especially with all the +'s added to 14nm, the focus is just raw frequency potential and throughput, density and efficiency be damned.
Well, the aim is making a fast CPU, so the high-end desktop CPUs are fast. Stop caring so much about how many "+" there are. :)
Intel can make efficient CPUs when it matters - in notebooks. Zen is not even close and the result is that low-voltage AMD solutions are still using old Excavator chips.

As you said: Zen is designed to be efficient in average situation. But Intel makes more purpose-built CPUs and they'll always have the edge in particular scenarios.
And when Intel catches up on node efficiency, it will be interesting to see AMD's response. We'll see how long the Zen strategy is going to work. :)
Posted on Reply
#17
Darmok N Jalad
notb, post: 3982577, member: 165619"
It's made for SFF office machines, possibly for the more demanding users. I'm pretty sure it'll be fast enough.
This CPU is about getting the job done, not getting great benchmark results. :)

No, they don't. They'll easily exceed TDP by around 10%. But that's just few Watts, so people don't care very much.
It's quite similar on Intel's side. Or at least it used to be until 8th gen.
8700K can go past TDP by as much as 30%. But it is a K CPU and Intel strongly advises you to get good PSU and cooler for these models.
8700 also seems to get past 65W, but I haven't seen decent tests, because - sadly - non-K CPUs don't get that much attention from reviewers.

Well, it's a price you pay for single-core performance. And Intel's 8th gen has lots of it.

The main problem here is efficiency and heat distribution. Heat increases faster than power draw (i.e. CPUs are less effective at high frequencies). That's the main problem with current lineup of the Blues.

Well, the aim is making a fast CPU, so the high-end desktop CPUs are fast. Stop caring so much about how many "+" there are. :)
Intel can make efficient CPUs when it matters - in notebooks. Zen is not even close and the result is that low-voltage AMD solutions are still using old Excavator chips.

As you said: Zen is designed to be efficient in average situation. But Intel makes more purpose-built CPUs and they'll always have the edge in particular scenarios.
And when Intel catches up on node efficiency, it will be interesting to see AMD's response. We'll see how long the Zen strategy is going to work. :)
My 2400G is rated for 65W, and best I can tell from monitoring tools, it might hit 70W for a few seconds, but it does average 65W or less over a long load, and that’s with it bumping the rated 3.9ghz max boost on all cores. Granted, I’m using a dGPU in it now, but it’s still performing very near the advertised speeds and TDP. I think most modern semi’s will boost past the norm for a little bit before they heat up, as even my rx480 will report a max TBP of 180W, but it usually averages under 110W in games (I undervolt). I think it’s one thing for a semi to boost past the rated TDP for a few seconds, it’s entirely another to just blow the rating out of the water with no regard. It seems more like false advertising at that point. I suppose intel wants you to think you’re getting bonus performance by overclocking for you? It’s not like the way the chip behaves is inherently bad—it’s just a bad point on the spec sheet.

And no arguement on Zen Mobile. AMD needs a better answer there. I think it’s the Infinity Fabric that kills them on mobile—it just can’t power down enough like the cores and GPU can.
Posted on Reply
#18
ghazi
notb, post: 3982577, member: 165619"
It's made for SFF office machines, possibly for the more demanding users. I'm pretty sure it'll be fast enough.
This CPU is about getting the job done, not getting great benchmark results. :)
That's not the point... the point is that these chips either won't sustain their boost clocks or won't stay within TDP. Reviewers might test these CPUs using a CLC and a motherboard that couldn't care less about the TDP and give consumers an inaccurate picture of the performance.
No, they don't. They'll easily exceed TDP by around 10%. But that's just few Watts, so people don't care very much.
It's quite similar on Intel's side. Or at least it used to be until 8th gen.
8700K can go past TDP by as much as 30%. But it is a K CPU and Intel strongly advises you to get good PSU and cooler for these models.
8700 also seems to get past 65W, but I haven't seen decent tests, because - sadly - non-K CPUs don't get that much attention from reviewers.
You do realize that TDP is thermal design power, not power consumption, right? The amount of heat dissipated by the CPU is necessarily lower than the amount of power it draws. Additionally, TDP is a sustained value, not accounting for spikes.

The i7-8700 when used with adequate cooling will use the exact same amount of power as the i7-8700K, about ~120-125W according to Anandtech. If you try using the 8700 with the stock cooler, which is rated for 73W (as opposed to the chip's 65W), it hits 100C and throttles within 5 seconds under any kind of serious load. This causes a very serious impact to performance that is generally not reflected in reviews.
The main problem here is efficiency and heat distribution. Heat increases faster than power draw (i.e. CPUs are less effective at high frequencies). That's the main problem with current lineup of the Blues.

Well, the aim is making a fast CPU, so the high-end desktop CPUs are fast. Stop caring so much about how many "+" there are. :)
Intel can make efficient CPUs when it matters - in notebooks. Zen is not even close and the result is that low-voltage AMD solutions are still using old Excavator chips.
You seem to be taking this as emotionally charged when I'm simply analyzing why the chips compare to each other the way they do. I would hope Zen 2 is using relatively less dense libraries for TSMC 7nm and is geared (from a manufacturing standpoint) towards high transistor performance and high yield at the expense of efficiency, similarly to how Intel tweaked 14nm to be focused on high-performance compute instead of laptops.

As for why AMD isn't using Zen in the ultra-ultra-low-voltage chips yet: Raven Ridge is a 210mm^2 die which is unsuitable for these markets and wattage ranges, and I assume AMD doesn't see these markets as worth the money that'd have to go into designing a new die. AMD has indicated on roadmaps if I'm not mistaken that Zen will be going down to the 4-5W range by the end of 2020.
As you said: Zen is designed to be efficient in average situation. But Intel makes more purpose-built CPUs and they'll always have the edge in particular scenarios.
And when Intel catches up on node efficiency, it will be interesting to see AMD's response. We'll see how long the Zen strategy is going to work. :)
Let's not forget that Zen's clock potential was limited by its node, which AMD had no control over and was originally designed for cell phones. In the future AMD will be able to utilize the best-performing HPC nodes for high power products and the most efficient low power nodes for low power products.
Posted on Reply
#19
Mussels
Moderprator
Coming soon to an overheating dell/mac near you
Posted on Reply
#20
eidairaman1
The Exiled Airman
Ok so a cpu for laptop huh? ;)
Posted on Reply
#21
TheGuruStud
NicklasAPJ, post: 3982475, member: 136240"
Thats how they both are doing, AMD are doing the same thing.
Wrong. Dead wrong.

TDP and consumption aren't 1:1, but they should be close unless you're just plain lying. Depending on your test app the 2700X uses 10W less than this (magically inline with TDP...).

78% higher consumption vs direct TDP comparison....just lol. Basically, 9900T is a joke, too, but on a grander scale. OEMs will have to limit it or it'll burn up whatever SFF they stick it in.
Posted on Reply
#22
notb
TheGuruStud, post: 3982792, member: 42692"
OEMs will have to limit it or it'll burn up whatever SFF they stick it in.
So they will. Why worry so much? :-)
You can choose between slower 8 cores or faster 4/6. Choice is good, right? :-)

I'm really looking forward to reviews.
Based on clocks, single-thread performance of this CPU should be around that of Intel's mobile CPUs. Which means it should be faster than 1700X (and not far behind 2700)
1700X was released just 2 years ago and sucked 100W in full load. 35W or 65W - 9900T would be a fantastic achievement.
Posted on Reply
#23
TheGuruStud
notb, post: 3982833, member: 165619"
So they will. Why worry so much? :)
You can choose between slower 8 cores or faster 4/6. Choice is good, right? :)

I'm really looking forward to reviews.
Based on clocks, single-thread performance of this CPU should be around that of Intel's mobile CPUs. Which means it should be faster than 1700X (and not far behind 2700)
1700X was released just 2 years ago and sucked 100W in full load. 35W or 65W - 9900T would be a fantastic achievement.
Yeah....mmkay. Lay off the drugs. 2700 uses 62 watts LOL. Intel is a joke. If 1.7 = 35 tdp, then what what does 3(+) ghz equal? It sure isn't beating AMD.

This something of a 1700 nonX competitor if you keep consumption in check-ish. If they wanted low TDP, then all they had to do was disable HT.
Posted on Reply
#24
ArbitraryAffection
notb, post: 3982577, member: 165619"
Well, the aim is making a fast CPU, so the high-end desktop CPUs are fast. Stop caring so much about how many "+" there are. :)
Intel can make efficient CPUs when it matters - in notebooks. Zen is not even close and the result is that low-voltage AMD solutions are still using old Excavator chips.
Huh, the 2500U in my HP Envy X360 is pretty efficient at 15W. According to HWINFO64 it doesnt go over 12W even when playing games. Though I admit the battery life isn't as great as I hoped for when in use. Something I've heard a lot about with the Ryzen mobile low power parts. Thinking it could be a software firmware issue?

notb, post: 3982833, member: 165619"
So they will. Why worry so much? :)
You can choose between slower 8 cores or faster 4/6. Choice is good, right? :)

I'm really looking forward to reviews.
Based on clocks, single-thread performance of this CPU should be around that of Intel's mobile CPUs. Which means it should be faster than 1700X (and not far behind 2700)
1700X was released just 2 years ago and sucked 100W in full load. 35W or 65W - 9900T would be a fantastic achievement.
I had a 1700 on launch and in full load it was 3.3 GHz all core and stuck pretty close to the 65W TDP: around 70-75W peak in Prime95 iirc. My friend has a Laptop with a socketed 2700 and at stock he is almost bang on 65W at 3.4GHz all core.
Posted on Reply
#25
Vayra86
notb, post: 3982480, member: 165619"
This CPU is made mainly for OEMs and they will know how to tune it to keep under certain limit they need (be it 35W or 45W). Don't worry too much. :)
Sorry what. Razer would like a word with you. And Asus. And GB. And MSI. And literally everybody else.
Posted on Reply
Add your own comment