• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Question 14-core i9-10940X and the 12-core i9-10920X. Performance Leak on Geekbench.

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
N

nicalandia

Diamond Member
amrnuke said:
Fair points. The 2950X is a better price to performance chip than either 2990WX or 9908XE. The 9980XE is no slouch, and isn't a particularly horrible value either, of course depending on workload. Everything about TR3 and 10xxxXE performance is all speculation. NUMA issues are important, and I look forward to seeing the performance of TR3. But... they're still issues that hamper the 2990WX.
Click to expand...
What hampers the 2990WX? Can't be nebulous. Because as I pointed out we know how performance translates from the parts we do have (Ryzen 3k and Epyc 2) and can tell you whether it exists.

The big point is neither of the these CPU's is a real mystery. There are minor details ones that can create variance here and there. But with Cascade Lake -X we have the SL-X refresh (9980XE) to map out performance and on TR3 we have Ryzen 3k and EPYC 2. The only thing comparing to the TR2 (2990WX) is good for is pricing and outside exact clock speeds that is the only aspect that can have a real affect on comparisons and honestly pretty much clock speeds we can extrapolate against Ryzen 3 and it's power usuage the question is does AMD use 180w, 200w, or 250w as its power template. If it (32c TR3) uses 200w its pretty much going to perform exactly as a 3900x with 20 more cores.
 
maddie said:
Why this, 'They might want to hold off on that one though to have it be a 4x0*4, so need stock of bad chips'?

These tiny die are probably yielding so high and of such low cost that AMD has no reason to leave any SKU off the table because of die harvesting delays. The margins on a 4+1 die TR 16C at $800 would be fantastic, even with the use of cut down more functional die.
Click to expand...

Because they would rather have 4x0*4 then 2x2*4. 2x2 chips are probably more annoying to find then even one that lets say has issues with the IF connection between the CCX or bad L3 cache. Also having a couple Single CCX dies SKU's means they have a place to bin them. Third if AMD is selling every die that can be used it more important to maximize the binning for highest margins possible.

What AMD doesn't want to do is have two different configurations. You either do 2x2 Dies or or 4x0 dies. Even with the performance leveling aspects of the IO die they won't want to introduce product variance into a single SKU like that.
 
Topweasel said:
What hampers the 2990WX? Can't be nebulous. Because as I pointed out we know how performance translates from the parts we do have (Ryzen 3k and Epyc 2) and can tell you whether it exists.

The big point is neither of the these CPU's is a real mystery. There are minor details ones that can create variance here and there. But with Cascade Lake -X we have the SL-X refresh (9980XE) to map out performance and on TR3 we have Ryzen 3k and EPYC 2. The only thing comparing to the TR2 (2990WX) is good for is pricing and outside exact clock speeds that is the only aspect that can have a real affect on comparisons and honestly pretty much clock speeds we can extrapolate against Ryzen 3 and it's power usuage the question is does AMD use 180w, 200w, or 250w as its power template. If it (32c TR3) uses 200w its pretty much going to perform exactly as a 3900x with 20 more cores.
Click to expand...
I'd argue that minor details can make a real difference. Original Zen release had a lot of release issues, and those chips came into their own a little later.

I agree that we do have a template and know how the Zen 2 chiplets work at 6-12 cores per chip at high clocks on Ryzen, and at lower clocks but more cores on Epyc, but we don't know how things will work with more cores and at higher clocks combined. I think it's at least prudent, even if we expect great things (which, to be clear, I do), to also expect speedbumps.
 
amrnuke said:
I'd argue that minor details can make a real difference. Original Zen release had a lot of release issues, and those chips came into their own a little later.

I agree that we do have a template and know how the Zen 2 chiplets work at 6-12 cores per chip at high clocks on Ryzen, and at lower clocks but more cores on Epyc, but we don't know how things will work with more cores and at higher clocks combined. I think it's at least prudent, even if we expect great things (which, to be clear, I do), to also expect speedbumps.
Click to expand...
Threadripper has one thing going for it that Ryzen 3900x does NOT have. A HUGE surface to draw out all the heat it produces. With a good 360 AIO, or maybe even good air (the Noctura NH-14S with dual fans), the threadripper is more capable. I know, I have all of the above. I think we could see a t6 core TR3 that does 4.5 turbo on all cores ! Not that might even require custom water, but reghardless, I think clocks will be better than the Ryzen 3000 series due to this.

So, to stay on-topic, I think the Cascade-lake X CPUs will be EOL compared to Ryzen 3000 Threadrippers.
 
amrnuke said:
I'd argue that minor details can make a real difference. Original Zen release had a lot of release issues, and those chips came into their own a little later.

I agree that we do have a template and know how the Zen 2 chiplets work at 6-12 cores per chip at high clocks on Ryzen, and at lower clocks but more cores on Epyc, but we don't know how things will work with more cores and at higher clocks combined. I think it's at least prudent, even if we expect great things (which, to be clear, I do), to also expect speedbumps.
Click to expand...
What speed bumps? What minor details. I get that you may not think, probably imho you haven't looked at actual performance numbers between the 3 configurations we have seen (1 Die, 2 Die, 8 die Zen 2) to be able to speculate and measure the performance levels of a TR3. But even if you look at cooling. AMD is using the same config at 200-280w on the server end so we know power wise and cooling capability wise the socket and heatspreader have the ability to handle a 200w load. We know what it takes to keep a 3900x cool and in boost. So not even getting into overclocking, as much as we look at numbers on the 10980xe and can project its performance we can do the same for Threadripper. The only things we don't know is actual boost and base clocks because we don't know power limit and we don't know price. We can probably guess within 100 MHz of Speed, so the price is the only real unknown and we do have somewhat of a model for that.

What that equals is if I had to guess on the 24c CPU. A 3.5 GHz base, 4.7 GHz boost, a 4.0 GHz all core boost. If it does that it will be faster in clock, much faster in IPC, and will win just about every metric you can measure and it will come in $1200-$1250. Giving an extra 6 core of compute, for the extra cost of 1 3600x. In fact let me put is prediction in, I will even put it in my signature, changing it for the first time in a decade. The only CPU in the whole Cascade Lake - X lineup that has a chance of being a better gaming CPU is the 10900x, even then it will only matter in games so basic that all the performance is about Frame calls, because as something needs even more then the most basic CPU work, the superior IPC will win.
 
Atari2600 said:
I take it you measured system power draw at idle? and were not using the display heads at the time?

There were users elsewhere running a 3900X at 3.1 GHz fixed to prove that it got 65W - it did. Can't remember the workload - may not have been using AVX units.

It also doesn't tally up with results from reviewers.

y-Cruncher: 115W
Handbrake x264: 139W
Handbrake x265: 132W

Which is entirely in line with AMD's default PPT setting:


Do you have PBO enabled? [As at this point, stock TDP goes out the window, as can be observed within the review.]


Oh and I note you are inferring that TDP == Power consumed. It does not. It never has, but it is commonly misunderstood as being the same thing.
Click to expand...
not at idle, full load handbrake, no PBO no thank you that regulation isnt for me
I know those reviews with pure package power displaying only CPU power, but I couldnt care less about them, since the important is system power
https://static.techspot.com/articles-info/1869/bench/Power.png (this is blender, which uses a little less power than handbrake)
and here https://techreport.com/review/34672/amd-ryzen-7-3700x-and-ryzen-9-3900x-cpus-reviewed/13/ and other when you look at whole system vary from 220 to 250W

while measuring mine I did look at the power meter the whole time and power fluctuated from 215 to 250W with average of 240W, which is completely in line with the reviews (and with other users on this forum)
I know TDP isnt power, but the whole f.g web is using it suddenly as power..and you know since when?....since Intel released 9900K
I am pissed by the double standards used on the internet (and couldnt care less that WCFtech and other roflcopter fanbased sites use it), but I am more pissed that the double standards are used on the tech forum like anandtech, I expected more

tdp doesnt mean power, so why are users pasting package power comparing to listed tdps of amd and intel and even ignoring their own definiton...biased lie nothing else

since the 7900X whole system consumes about the same as 3900X, I expect the 10920X (+2 cores vs 7900X) to consume the same with some improvements Intel made

about that pricing- interesting and since in Intel they are masters of market deception we should analyze that pretty much shock move- my point of view
  • intel knows amd has demand problems and the 3900X price rises (as was 9900K before)
  • intel uses this moment to enter the core wars (up to 18C) with aggresive pricing
  • that way they put pressure on AMD to release their 16C with aggresive pricing- that is the reason they cancelled 16C variant
  • they are adressing AMD profit- as I like my 3900X much, it has to cost a fortune with its 70MB cache to manufacture at 7nm...while Intel uses their fine tuned 14nm
  • joe user will see with intel 4 important marketing things from Intel- coarz number, extreme processor, 4,8GHz turbo OMG, 4mem channels (not 2, not that it matters), moar pcie lines I want that
  • so amd will be competing with 3950x with less lanes, less channels and lower turbo frequency and with that clockgate they didnt earn much trust ..
  • intel is master of marketing......with worse product, since 3950x will except pure mem bandwitch tasks tear the cascade lake x lineup
age of stupidity comes again...
 
Atari2600 said:
I think its safe to say that the bottom of the TR3 line will come in considerably lower than $980.
Click to expand...

I doubt it. Form the horses mouth TR3 starts at 24 cores and if it costs the same per core as 3950x it would be $1125. 24 cores means four 6-core chiplets. I assume due to heat distribution and heatsink pressure anything less than 4 chiplets doesn't make much sense and there probably are not enough defect to warrant an sku with only 4 cores per chiplet.

But yeah starting at maybe $999 will make TR3 very, very niche. And at $999 it would be a bargain compared to the 3950x.
 
Topweasel said:
What speed bumps? What minor details. I get that you may not think, probably imho you haven't looked at actual performance numbers between the 3 configurations we have seen (1 Die, 2 Die, 8 die Zen 2) to be able to speculate and measure the performance levels of a TR3. But even if you look at cooling. AMD is using the same config at 200-280w on the server end so we know power wise and cooling capability wise the socket and heatspreader have the ability to handle a 200w load. We know what it takes to keep a 3900x cool and in boost. So not even getting into overclocking, as much as we look at numbers on the 10980xe and can project its performance we can do the same for Threadripper. The only things we don't know is actual boost and base clocks because we don't know power limit and we don't know price. We can probably guess within 100 MHz of Speed, so the price is the only real unknown and we do have somewhat of a model for that.

What that equals is if I had to guess on the 24c CPU. A 3.5 GHz base, 4.7 GHz boost, a 4.0 GHz all core boost. If it does that it will be faster in clock, much faster in IPC, and will win just about every metric you can measure and it will come in $1200-$1250. Giving an extra 6 core of compute, for the extra cost of 1 3600x. In fact let me put is prediction in, I will even put it in my signature, changing it for the first time in a decade. The only CPU in the whole Cascade Lake - X lineup that has a chance of being a better gaming CPU is the 10900x, even then it will only matter in games so basic that all the performance is about Frame calls, because as something needs even more then the most basic CPU work, the superior IPC will win.
Click to expand...
how did you come to a conclusion 10900X is gaming competitive?
Intel has nothing competitive above 8C now because the product lines
up to 8C there is fast ringbus 9900K which still has no competition when oced, as has 8700K, intel doesnt have problem even if their CPUs are more expensive
the real damage comes from 3900X and above, since the competition is cascade lake x, which has bad gaming IPC
my 3900x has no weaknesses, except that shi..y power
about other IPC, how did you come to the superior IPC conclusion of the r3x lineup?
looking at the numbers, its pretty much tied
even the stilt made some measurements https://www.overclock.net/forum/10-amd-cpus/1728758-strictly-technical-matisse-not-really.html
I consider the products tied
higher MT yield coefficient produces higher absolute performance when running MT loads ofc
 
TheGiant said:
I know those reviews with pure package power displaying only CPU power, but I couldnt care less about them, since the important is system power
Click to expand...

Ye can't really blame the CPU for other components consuming 100W!

Maybe you can blame AMD's chipset (whats the mobo?).
 
Atari2600 said:
Ye can't really blame the CPU for other components consuming 100W!

Maybe you can blame AMD's chipset (whats the mobo?).
Click to expand...
mate you are not stupid.....sorry mods
100W for which components
RAM, board, SSD reads and writes? GPU idling? pretty much the same setup does with the 6600k undervolted 110W wall power at 4,4 GHz, which shows 42-47W in intel extreme utility as package power while stress testing (undervolted)
so with all the losses of VRMs and PSU, I doubt that more than 50W is the power of ram+vrm+ssd+gpu idling+chipset summed together
there is no way my 3900X TDP wattage (yes, TDP is ruined by W it should be some rating number unitless) consumes 105 or 140W, no way with b450 board or there is simply something fundamentally wrong that the CPU itself has the power rated and the power delivery has the losses .....
it could explain why my b450 died after 48hrs of constant handbrake load......
anyway my statement stays...the system power is important, not the CPU power
 
TheGiant said:
100W for which components
RAM, board, SSD reads and writes? GPU idling? pretty much the same setup does with the 6600k undervolted 110W wall power at 4,4 GHz, which shows 42-47W in intel extreme utility as package power while stress testing (undervolted)
so with all the losses of VRMs and PSU, I doubt that more than 50W is the power of ram+vrm+ssd+gpu idling+chipset summed together
Click to expand...

So your saying Toms didn't measure the CPU power at all?

It literally can't be both! 1 + 1 does not equal 3.


If the CPU has a max power draw of 142W (PPT) with PBO disabled - and that has been substantiated via independent review - then your power consumption is coming from elsewhere.

PSU, fans, RAM, board, GPU whatever.

If your PSU is 80% efficient, then that is 35W. [If you are pulling <30% load through a PSU, don't expect to hit the top efficiency ratings]
DDR4 consumes about 3W per 8GB stick. 12W.
5700XT idle power consumption is around 54W.

That is 101W. Without including the VRMs, SDD, chipset and other motherboard ancillaries.

101 + 142 = 243W. Which is right in the middle of your 240-250W.

You want to pull wall power down - you need a better idling GPU and a better matched PSU to your load rating.
 
Atari2600 said:
So your saying Toms didn't measure the CPU power at all?

It literally can't be both! 1 + 1 does not equal 3.


If the CPU has a max power draw of 142W (PPT) with PBO disabled - and that has been substantiated via independent review - then your power consumption is coming from elsewhere.

PSU, fans, RAM, board, GPU whatever.

If your PSU is 80% efficient, then that is 35W. [If you are pulling <30% load through a PSU, don't expect to hit the top efficiency ratings]
DDR4 consumes about 3W per 8GB stick. 12W.
5700XT idle power consumption is around 54W.

That is 101W. Without including the VRMs, SDD, chipset and other motherboard ancillaries.

101 + 142 = 243W. Which is right in the middle of your 240-250W.

You want to pull wall power down - you need a better idling GPU and a better matched PSU to your load rating.
Click to expand...
you must be smoking something
5700XT total system idle https://images.anandtech.com/graphs/graph14618/111343.png is 54W
no, the components add with all the losses no more than 50W and that is an exxageration IMO
the power is in the CPU
AMD is in the same as Intel for some years, catching frequency and with AVX2 muscles the power goes up
my guess is my 3900X consumes 170W, no less


to the thread point

I hope techspot does some really honest testing of the new cascade lake x without hiding something that shows AMD in worse light
I am pissed of this biased -Redacted- of latest

You already know better than to use profanity in the Tech forums.

Daveybrat
AT Moderator
 
Last edited by a moderator:
amrnuke said:
Well, it's workload dependent right? If we're audio processing, you could get a cheap XE and beat the 2990WX. If you're rendering, you want a TR for the most part.
Click to expand...

That's Intel's marketing angle: compare Cascade Lake-X to last-gen chips.

Cascade Lake-X is already going to struggle against AM4 chips like the 3950X. Against TR3 it will be a massacre. They would have to charge prices close to what they charge for the 9900K (and upcoming 9900KS) to win on price/performance.
 
Atari2600 said:
Really?

Whose mouth exactly - as 16C 4Ch unmarked AMDs have been spotted in the wild...?
Click to expand...

Official AMD slides couple days ago stated: "starting at 24 cores"

Yeah maybe later there is a 16-core. But I still doubt due to my previous reasoning and also from the horses mouth about TR3: "core count going up, up". (vs just up for Ryzen3).
 
beginner99 said:
Official AMD slides couple days ago stated: "starting at 24 cores"

Yeah maybe later there is a 16-core. But I still doubt due to my previous reasoning and also from the horses mouth about TR3: "core count going up, up". (vs just up for Ryzen3).
Click to expand...

It states "premiering with 24 cores" not starting - I don't see how that excludes lower end models even at the same time.
 
Topweasel said:
What speed bumps? What minor details. I get that you may not think, probably imho you haven't looked at actual performance numbers between the 3 configurations we have seen (1 Die, 2 Die, 8 die Zen 2) to be able to speculate and measure the performance levels of a TR3. But even if you look at cooling. AMD is using the same config at 200-280w on the server end so we know power wise and cooling capability wise the socket and heatspreader have the ability to handle a 200w load. We know what it takes to keep a 3900x cool and in boost. So not even getting into overclocking, as much as we look at numbers on the 10980xe and can project its performance we can do the same for Threadripper. The only things we don't know is actual boost and base clocks because we don't know power limit and we don't know price. We can probably guess within 100 MHz of Speed, so the price is the only real unknown and we do have somewhat of a model for that.

What that equals is if I had to guess on the 24c CPU. A 3.5 GHz base, 4.7 GHz boost, a 4.0 GHz all core boost. If it does that it will be faster in clock, much faster in IPC, and will win just about every metric you can measure and it will come in $1200-$1250. Giving an extra 6 core of compute, for the extra cost of 1 3600x. In fact let me put is prediction in, I will even put it in my signature, changing it for the first time in a decade. The only CPU in the whole Cascade Lake - X lineup that has a chance of being a better gaming CPU is the 10900x, even then it will only matter in games so basic that all the performance is about Frame calls, because as something needs even more then the most basic CPU work, the superior IPC will win.
Click to expand...
The speed bumps (having to run memory at stock JEDEC due to the IMC, a few motherboard issues) were minor, as I said. When running at stock it was fine (save for the FMA issues and segfault issues with Linux - resolved via micro-code and chip replacement IIRC). Even with Zen 2 we have seen some speed bumps and minor issues to the point that Anandtech re-ran their benchmarks with BIOS updates (and some fairly major ones, like advertised boost not matching real boost).

So I am just, out of an abundance of caution, being cautious with my expectations. BUT deep down inside, I am very excited to see how these perform.
 
Atari2600 said:
I must have smoked glasses.

I didn't notice it was system - I just assumed it was isolated to GPU.




Again, isolated measurements do not indicate this.

You are assuming that none of your other components are doing what they should.

What is system power consumption at idle?
Click to expand...
Hello back
Isolated, non isolated, doesnt matter
There are no components to burn more power. its the CPU/vrm...
.idle power is 59W

there are other cl-x threads so i will post my opinions there
this was off topic
 
I saw this today: https://www.guru3d.com/news-story/intel-core-i9-10980xe-cascade-lake-x-benchmarks.html
AMD stays close alright. If you focus on the tests it becomes obvious that AMD is able to keep their turbo bins running longer than Intel. For gaming, Intel takes the lead, albeit the 999 USD Cofre i9-10980XE is not a gaming processor, for that the 9900K shines.
Click to expand...
Of course we have to wait for a full review, but from these benchmarks, the $500 3900X really holds its own against a $1k CPU.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top