Can you please summurize to me what's coming next from intel as far as desktop is concerned

sram

[H]ard|Gawd
Joined
Jul 30, 2007
Messages
1,699
Hi,

Yes this maybe a lazy question and I probably need to do my homework but I have a test tomorrow and I need to know a decent answer to tell a friend of mine. I'm mostly interested in knowing the TOP cpu when it comes to mainstream desktop or HEDT platforms. I'm coming from i7 7700K kaby lake, and want the next big thing. I can see that the i9 is coming (i9 9900k?) which is going to have a soldered IHS (No need for delidding)...that is good news actually. I can also see that AMD released their 2990WX and 2950X or whatever and they got nice reviews. I think they are workstation cpus. Will intel release cpus with comparable number of cores? (with HT enabled of course). I actually don't mind switching to AMD if they have the performance crown currently. Sorry this is a very hasty post. Me and one of my friends want to talk about building a box for bragging rights that is.

Thanks
 
Yes. Add to the fact that usually after a company has taken a swift kick to the nuts and they lose some face, they tend to come back with some good stuff. Sorta like Chrysler/Dodge or AMD themselves.
 
Yes. Add to the fact that usually after a company has taken a swift kick to the nuts and they lose some face, they tend to come back with some good stuff. Sorta like Chrysler/Dodge or AMD themselves.
But hopefully it is not ganna take a long time. It has taken AMD what? 10 years? When was the first core processor? I think it was 2006 if I remember correctly.
 
Damn. I just took a minute to scan the other thread in here titled intel 9th generation. I'm quoting somebody:

"Minnesota gonna run out of lakes pretty soon"

It is this bad. Pathetic Intel. So it is like Brian said, it is more of the same. I think I will wait.
 
Last edited:
More 14nm, few more cores, clock bump, more heat, poor efficiency, less clock headroom (they are using headroom you used to take for granted).
Or half defective 10nm, so bad they get duals from quads, the iGPU is dead and wrapped up with a clock regression.
So yeah, AMD with 7nm next year for desktop is looking good right now. I'd wait for that than bother with spectre/meltdown riddled 14nm stuff.
But If I was building right now what you want to do? 2950x
That thing is still a beast no matter what Intel eventually drags out around December or whenever they will be actually available and easily obtained.
 
More 14nm, few more cores, clock bump, more heat, poor efficiency, less clock headroom (they are using headroom you used to take for granted).
Or half defective 10nm, so bad they get duals from quads, the iGPU is dead and wrapped up with a clock regression.
So yeah, AMD with 7nm next year for desktop is looking good right now. I'd wait for that than bother with spectre/meltdown riddled 14nm stuff.
But If I was building right now what you want to do? 2950x
That thing is still a beast no matter what Intel eventually drags out around December or whenever they will be actually available and easily obtained.
You summarized it very well. I thank you for that. 14nm !!! Why? Why not change the process if they can? When they were alone, maybe it is understandable, but now with AMD kicking it???

Anyways, I don't really need a system now. It is just the upgrade bug again. The most tasking thing I do is probably virtualization. I tend to run 7+ virtual machines at the same time and compare linux distributions. So the more cores the better, and I need RAM of course. Does the i9 9900K support higher than 64GB memory?
 
You summarized it very well. I thank you for that. 14nm !!! Why? Why not change the process if they can? When they were alone, maybe it is understandable, but now with AMD kicking it???

Anyways, I don't really need a system now. It is just the upgrade bug again. The most tasking thing I do is probably virtualization. I tend to run 7+ virtual machines at the same time and compare linux distributions. So the more cores the better, and I need RAM of course. Does the i9 9900K support higher than 64GB memory?

Supposedly the i9-9900K maxes out at 64GB.

If you are virtualizing you'll want a lot of cores/threads as well as tons of RAM so the Threadripper2 2950x or 2990wx start looking good as they support 128GB quad-channel RAM (the 9900K is dual-channel) and have a more wallet-friendly core-to-$ ratio. TR2 also supports unbuffered ECC RAM so if you are running something sensitive to data corruption eg. FreeNAS then it starts looking very attractive price-wise vs. Xeons. Mind you apparently Intel is for their bigger customers starting to practically price-match the Epyc so if you know somebody a decent Xeon could be in the cards for you.

Are you using a Type-1 or Type-2 hypervisor? If it is a Type-2 and you plan on using the computer for things like gaming or AVX512-heavy processing then the 9900K is definitely worth a look depending on your budget. The 9900K will own the Ryzen 2700X in everything except ECC RAM and possibly price. If you like to render or do other things that are core/thread-count dependent then the TR2 is probably the way to go.

7nm Zen2 cores won't be out until at least 2019 (I'm a little skeptical about the announced timelines) so if you have a specific schedule in mind that is something to consider. There isn't any reliable available data on the Zen2's performance so no point in speculating a Ryzen3 against the 9900K.
 
What about the heat? I think Coffee Lake and Ryzen are about even in temps? All the benchmarks I found, seemed to indicate they are pretty close - if you delid the CL, it's pretty close to Ryzen and therefore, the real solder on the i9 - the temps on each should be very close.

The Coffee Lake processors seem more power efficient than Ryzen, too. What about that, guys? One thing that delays my decision is that discrepancy. The 2700X doesn't seem as efficient but is that because of the extra two cores and extra threads? The 2600X has 6 cores, too, though and the power consumption is more than i5 Coffee Lake processors.

Just something to think about or it's not that important?
 
What about the heat? I think Coffee Lake and Ryzen are about even in temps? All the benchmarks I found, seemed to indicate they are pretty close - if you delid the CL, it's pretty close to Ryzen and therefore, the real solder on the i9 - the temps on each should be very close.

The Coffee Lake processors seem more power efficient than Ryzen, too. What about that, guys? One thing that delays my decision is that discrepancy. The 2700X doesn't seem as efficient but is that because of the extra two cores and extra threads? The 2600X has 6 cores, too, though and the power consumption is more than i5 Coffee Lake processors.

Just something to think about or it's not that important?

Temp is a function of your cooling. You can make any chip run at (almost) any temp you want to if you apply enough heat transfer to it.

The key metric you are probably looking for is TDP - thermal design power. This is the metric that all the power/temp regulation functions work off of, and what power management is supposed to regulate the chip to for steady state maximum power consumption. It's a semi-marketing term, so it has to be taken with some degree of skepticism (chips can and will exceed it for short periods of time, and any type of overclocks utterly obliterate it).

In layman's terms, TDP is the amount of heat that a chip is going to give off at full rated power -- all cores at 100% at stock speeds. That also happens to be roughly equivalent to the amount of electrical power the CPU will require from the PSU. All things equal - the chip with a higher TDP will either run hotter with a given heat sink, or require a more robust cooling solution to run at the same temperature.

TDP is regardless of core count, frequency, etc. It's just a power rating at full power draw. If a quad core chip has a TDP of 100W, and a 32 core chip has a TDP of 100W - the TDPs are equal - they will give off the same amount of heat, they will require the same cooling, even though one has 8x more cores than the other.

There is a lot more physics that goes into it than that, but for what we are talking about here, it's a start. Solder versus TIM on the IHS - that does affect temperature, but it won't change the TDP. Solder makes it easier to cool for a given TDP (it has a lower thermal resistance), but the key number there is still going to be the TDP - if the heat is generated it has to be removed eventually, no matter if there is TIM or solder.

So... a 9900K is supposed to be a 95W TDP (which sounds legit, Intel is usually in the 75-100W range). The 2700X is a TDP of 105W. That's pretty close, and most non-workstation desktop CPUs run in the 75-100W TDP range.

Just for fun and comparison, the TDP of a 32 core 2990X Threadripper is 250W. High end video cards are usually in the 225-275W range.
 
Yeah, Intel/AMD are pretty close TDP-wise, with intel winning hands-down in idle power and AMD coming close or beating Intel in full load power. All that goes out the window when you overclock, of course. Particularly AMD's multi-die CPUs, the uncore power skyrockets when you overclock, though it's reasonable at stock compared to a couple separate systems running full-load.
 
Consider this, recently intel new CPUs required you to buy a new motherboard/chipset, so you need to sell the z270 or hope you have a board that can be modded for the new series. Or... Go AMD and know the AM4 socket will (or should) last you 2-3 more years, and any new Ryzen based off of AM4 should work with a BIOS update.

Now sounds great till you want PCIE 4.0 or any other new feature that isn’t supported on your mono. Also Most Ryzens lack an iGPU (minus 2200/2400g) so a dGPU is a must.
 
Supposedly the i9-9900K maxes out at 64GB.

If you are virtualizing you'll want a lot of cores/threads as well as tons of RAM so the Threadripper2 2950x or 2990wx start looking good as they support 128GB quad-channel RAM (the 9900K is dual-channel) and have a more wallet-friendly core-to-$ ratio. TR2 also supports unbuffered ECC RAM so if you are running something sensitive to data corruption eg. FreeNAS then it starts looking very attractive price-wise vs. Xeons. Mind you apparently Intel is for their bigger customers starting to practically price-match the Epyc so if you know somebody a decent Xeon could be in the cards for you.

Are you using a Type-1 or Type-2 hypervisor? If it is a Type-2 and you plan on using the computer for things like gaming or AVX512-heavy processing then the 9900K is definitely worth a look depending on your budget. The 9900K will own the Ryzen 2700X in everything except ECC RAM and possibly price. If you like to render or do other things that are core/thread-count dependent then the TR2 is probably the way to go.

7nm Zen2 cores won't be out until at least 2019 (I'm a little skeptical about the announced timelines) so if you have a specific schedule in mind that is something to consider. There isn't any reliable available data on the Zen2's performance so no point in speculating a Ryzen3 against the 9900K.

Thanks for your imput my friend. I'm actually using type-2 virtualization. And it is not me, but the kids may play fortnite or games like it on the machine. And I don't need ECC.
 
Last edited:
If you were buying today, then your options for the best of the best are:

AMD TR 2990WX - 32c/64t - 3.0/4.2 GHz - $1750
Intel i9-7980XE - 18c/36t - 2.6/4.4 GHz - $1900

...I think the clear winner is the Ryzen TR. Your opinion may differ, and that's cool...neither are slouches, so get whichever you feel has what you're looking for.
 
threadripper for hedt. why limit yourself?

I'm not sure what your saying. It's an incomplete phrase with no context.

I can run 5x handbrake encodes and still play TitanFall 2 at 144fps ultra maxed at 1080p. Um what's weak about 2950x?
 
I'm not sure what your saying. It's an incomplete phrase with no context.

I can run 5x handbrake encodes and still play TitanFall 2 at 144fps ultra maxed at 1080p. Um what's weak about 2950x?

congratulations. My 3930K does some cool things too,

back to the original question hedt the best is currently threadripper. the context was clearly set by the OP, What is not clear?

'Im mostly interested in knowing the TOP cpu when it comes to mainstream desktop or HEDT platforms"
 
Last edited:
Your lack of grammar and sentence structure is not clear. Start there!

Stop with the grammar nagging. Everyone else can understand but you apparently.

EDIT - Thanks Teach for adjusting your above post to reflect your outstanding English skills. I'm sure there are thousands of other posts in the forums you have to address now as well - so let me leave you to your work.

EDIT2 - couldnt stick it out so you deleted it. LOL
 
Last edited:
Hi,

Yes this maybe a lazy question and I probably need to do my homework but I have a test tomorrow and I need to know a decent answer to tell a friend of mine. I'm mostly interested in knowing the TOP cpu when it comes to mainstream desktop or HEDT platforms. I'm coming from i7 7700K kaby lake, and want the next big thing. I can see that the i9 is coming (i9 9900k?) which is going to have a soldered IHS (No need for delidding)...that is good news actually.

It is good news, but its news thats somewhat blown out of proportion. Intel's only switching to solder for PR reasons. Intel CPU's were still vastly out clocking their AMD counterparts while kicking the shit out of them in IPC for the bulk of the last decade. Solder or not, it didn't matter. Granted, soldering the heat spreader is good for us but Intel has never been concerned about overclocking. They simply wanted a chip to achieve certain performance levels at a given manufacturing cost. Nothing more, nothing less.

I can also see that AMD released their 2990WX and 2950X or whatever and they got nice reviews. I think they are workstation cpus.

Just what do you think Intel HEDT CPU's are? They are intended for the workstation and enthusiast market just like Threadripper. Intel created the HEDT market. It is responsible for the segregation of parts. Really, the HEDT market was created back in the late 1990's with the Pentium Pro. A CPU which other companies had zero competition for until it was too late. More specifically, Intel's HEDT CPUs have always been Xeon variants. The chipsets are variants of the chipsets used in workstation and server oriented motherboards. For gaming, Intel still has an edge for clock speed and for IPC. However, games are primarily GPU limited and when talking about workstation applications, AMD offers many more cores for the same money.

Also, consider the platform. While X299 is in some ways better than X399, its also got a huge flaw from a storage perspective. You are tied to Intel drives for bootable NVMe RAID arrays using the CPU's PCIe lanes and you have to pay extra for a vROC license key. Intel's being a bunch of greedy fucks with that shit given that everything is built onto the motherboard for this. Its just a cash grab. Its horseshit. AMD's platform has a few extra limitations but they aren't huge and you can do what you want with the storage for no additional charge.

Will intel release cpus with comparable number of cores? (with HT enabled of course). I actually don't mind switching to AMD if they have the performance crown currently. Sorry this is a very hasty post. Me and one of my friends want to talk about building a box for bragging rights that is.

Intel's still faster for games. Games don't leverage anything beyond 8 threads right now and most don't even do that. AMD is more bang for your buck. In certain applications Threadripper is an absolute monster that Intel doesn't really compete well with right now. But that all depends on a variety of factors. You have to look at what the box is being used for and then do your research to determine which is better.
 
Intel's still faster for games. Games don't leverage anything beyond 8 threads right now and most don't even do that. AMD is more bang for your buck. In certain applications Threadripper is an absolute monster that Intel doesn't really compete well with right now. But that all depends on a variety of factors. You have to look at what the box is being used for and then do your research to determine which is better.

I'm currently running dual Xeons with 12C/24T and Doom under Vulkan fully utilizes every core. :)
 
It is good news, but its news thats somewhat blown out of proportion. Intel's only switching to solder for PR reasons. Intel CPU's were still vastly out clocking their AMD counterparts while kicking the shit out of them in IPC for the bulk of the last decade. Solder or not, it didn't matter. Granted, soldering the heat spreader is good for us but Intel has never been concerned about overclocking. They simply wanted a chip to achieve certain performance levels at a given manufacturing cost. Nothing more, nothing less.



Just what do you think Intel HEDT CPU's are? They are intended for the workstation and enthusiast market just like Threadripper. Intel created the HEDT market. It is responsible for the segregation of parts. Really, the HEDT market was created back in the late 1990's with the Pentium Pro. A CPU which other companies had zero competition for until it was too late. More specifically, Intel's HEDT CPUs have always been Xeon variants. The chipsets are variants of the chipsets used in workstation and server oriented motherboards. For gaming, Intel still has an edge for clock speed and for IPC. However, games are primarily GPU limited and when talking about workstation applications, AMD offers many more cores for the same money.

Also, consider the platform. While X299 is in some ways better than X399, its also got a huge flaw from a storage perspective. You are tied to Intel drives for bootable NVMe RAID arrays using the CPU's PCIe lanes and you have to pay extra for a vROC license key. Intel's being a bunch of greedy fucks with that shit given that everything is built onto the motherboard for this. Its just a cash grab. Its horseshit. AMD's platform has a few extra limitations but they aren't huge and you can do what you want with the storage for no additional charge.



Intel's still faster for games. Games don't leverage anything beyond 8 threads right now and most don't even do that. AMD is more bang for your buck. In certain applications Threadripper is an absolute monster that Intel doesn't really compete well with right now. But that all depends on a variety of factors. You have to look at what the box is being used for and then do your research to determine which is better.

This is very informative. I'm at your debt Dan. Thanks.
 
  • Like
Reactions: Dan_D
like this
Personally I believe the push for better multithreaded performance as opposed to outright clock speed is well overdue.

The problem is that games don't always benefit from additional threads. Clock speed and general IPC still count for a lot in games. Not every application lends itself to additional parallelism.
 
The problem is that games don't always benefit from additional threads. Clock speed and general IPC still count for a lot in games. Not every application lends itself to additional parallelism.

Well they definitely wont when the API isn't optimized for it. Eventually multithreaded performance has to improve, silicon technology hit a brick wall years ago. Clock speeds aren't going to continue ramping in a cost effective manner and even at 4k GPUs need to be fed with data.
 
Considering overheads and hyperthreading affecting percentages, that's pretty good multithreaded CPU utilization running Doom under Vulkan:

LguvZJE.png
 
Personally I believe the push for better multithreaded performance as opposed to outright clock speed is well overdue.

Considering overheads and hyperthreading affecting percentages, that's pretty good multithreaded CPU utilization running Doom under Vulkan:

View attachment 105746

Games aren't throughput workloads. Games have one or two master threads and one or more helper threads. You can split helper task into two or more threads and execute them in more cores, but this doesn't improve performance if the master thread is bottlenecked. In above graph we can check one core is at 99% load. This is the bottleneck.
 
Yep, it is not always about an outdated API or "developers being lazy". It is just the nature of the workload. Hell, think of how limited humans are at multitasking!
 
Games aren't throughput workloads. Games have one or two master threads and one or more helper threads. You can split helper task into two or more threads and execute them in more cores, but this doesn't improve performance if the master thread is bottlenecked. In above graph we can check one core is at 99% load. This is the bottleneck.

The whole idea of master/slave multithreading 'is' to improve performance. In the case of a program to calculate distance traveled by a salesman the master thread would delegate slave threads to calculate the distance of every leg to report the results back to the master thread, whereby the master thread then calculates the result. This parallelism definitely increases performance, that's the whole point.
 
The whole idea of master/slave multithreading 'is' to improve performance. In the case of a program to calculate distance traveled by a salesman the master thread would delegate slave threads to calculate the distance of every leg to report the results back to the master thread, whereby the master thread then calculates the result. This parallelism definitely increases performance, that's the whole point.

Master/slave parallelization has its limits and drawbacks, one of the them are the overheads caused by master <--> slave communication. Those overheads introduce a limiting point beyond which further parallelization reduces performance instead increasing it. Moreover you said that Vulkan Doom "fully utilizes every core" but load profiling shows that most slave cores are under 50% utilization. The average is 44.2%. Reducing the cores to one half, the loads will not be 88.4%, because one has to substract the overheads of the eliminated N/2 threads. So an alternative hardware configuration with half the cores would still be far from fully utilizing every core (except the master, which will continue being the bottleneck of the system).
 
You want a summary of what's coming next from Intel?

Another 3-4 month wait to buy a paper launched processor.
 
Master/slave parallelization has its limits and drawbacks, one of the them are the overheads caused by master <--> slave communication. Those overheads introduce a limiting point beyond which further parallelization reduces performance instead increasing it. Moreover you said that Vulkan Doom "fully utilizes every core" but load profiling shows that most slave cores are under 50% utilization. The average is 44.2%. Reducing the cores to one half, the loads will not be 88.4%, because one has to substract the overheads of the eliminated N/2 threads. So an alternative hardware configuration with half the cores would still be far from fully utilizing every core (except the master, which will continue being the bottleneck of the system).

To quote myself:

Considering overheads and hyperthreading affecting percentages, that's pretty good multithreaded CPU utilization running Doom under Vulkan:

What you see there is 'effectively' 100% utilization across all cores on a hyperthreaded processor. I believe that in the case of a hyperthreaded processor HTOP will report both virtual cores at 50% utilization, adding both virtual cores together results in 100% utilization considering one physical core. What this highlights is that Vulkan is actually making use of hyperthreading. Of course nothing is going to be perfect, but that's pretty damn good multithreaded utilization all things considered..

I am getting 150 - 200fps @ 1200p using a 980Ti on a circa 2010 workstation @ 3.06Ghz, nothing overclocked. As far as I'm aware 200fps is the max Doom can render. That's pretty good performance for a system that's not even 4Ghz on a single core at a CPU bottlenecked 1200p.

[EDIT]:

Below we can see typical CPU utilization running CoD MW2 MP, this is the kind of utilization we have been used to in the past. In this scenario more cores will not do anything in relation to performance:

oXhFCON.png
 
Last edited:
What you see there is 'effectively' 100% utilization across all cores on a hyperthreaded processor. I believe that in the case of a hyperthreaded processor HTOP will report both virtual cores at 50% utilization, adding both virtual cores together results in 100% utilization considering one physical core.

So if 1+2 means one core is runing at 100% of load, then 5+6 means one core is running at 147%. Rigth?

Below we can see typical CPU utilization running CoD MW2 MP, this is the kind of utilization we have been used to in the past. In this scenario more cores will not do anything in relation to performance:

View attachment 107465

Same problem. Two cores are fully loaded and generating the bottleneck.
 
Back
Top