the ultimate $3500 or less vr-friendly PC build

[Alonzo]

7 hours ago, mpdugas said:

You ask "Do you have a link for this info?"  Sorry, I can't provide a link to my mind.  The following analysis is simply the result of my personal logic and thinking, however flawed it may be.  I am quite willing to be wrong about this, because I may very well have been laboring under some horribly wrong concepts about PC, etc., and I stand ready to accept being corrected about how I've thought this whole mess has worked all along.

 

As I understand, here's the fundamental issue: the CPU itself has only so many lanes, electrically, for receiving and transmitting information.  The MB does not "process" anything, so no matter how many lanes you fill it (the MB) with peripherals, no matter how many communication channels to the CPU that the MB offers, the CPU information input/output bandwidth limit is the choke-point.

 

The info that I got from Googling was that the motherboard uses something called DMI 3.0 to multiplex peripherals into the CPU, and it has something like PCI-e x4 worth of bandwidth to do so. So depending on peripheral, it's not as good as dedicated PCIe lanes directly into the CPU, but it's not too bad either. And it avoids the very weird situation of someone plugging in a USB device like a fan controller and suddenly their GPU is running at x8 instead of x16. That seems very bizarre to me, but certainly not beyond the realms of possibility, so I was curious about your info and thinking and wanted to explore it with an open mind (instead of being some dickhead on the internet just arguing with someone and telling them they're wrong -- there's plenty of that already I happen to think people who love flight sims like I do should be treated with some respect).

 

6 hours ago, ZachariasX said:

Socket1151 just use an equivalent of 4x PCIe3.0 to connect ALL your periphery if you have those 16 lanes in use for your GPU. (You have.)

 

This was my conclusion from Googling. Basically the motherboard bridges stuff into that PCIe x4 connection to the CPU, sharing that much bandwidth between devices (which is fine for USB junk, not fine for a schwack of NVMe drives).

[mpdugas]

7 hours ago, Alonzo said:

 

The info that I got from Googling was that the motherboard uses something called DMI 3.0 to multiplex peripherals into the CPU, and it has something like PCI-e x4 worth of bandwidth to do so. So depending on peripheral, it's not as good as dedicated PCIe lanes directly into the CPU, but it's not too bad either. And it avoids the very weird situation of someone plugging in a USB device like a fan controller and suddenly their GPU is running at x8 instead of x16. That seems very bizarre to me, but certainly not beyond the realms of possibility, so I was curious about your info and thinking and wanted to explore it with an open mind (instead of being some dickhead on the internet just arguing with someone and telling them they're wrong -- there's plenty of that already I happen to think people who love flight sims like I do should be treated with some respect).

 

 

This was my conclusion from Googling. Basically the motherboard bridges stuff into that PCIe x4 connection to the CPU, sharing that much bandwidth between devices (which is fine for USB junk, not fine for a schwack of NVMe drives).

 

I'm trying to see how a CPU, with limited electrical connections, gets more of them just because the MB has these additional pci-e x 4 channels into which peripherals can connect.  In the case of my MB, I know that if I plug certain pci-e slots with gear, the system reduces the adjacent slot's capacity, because the two slots share pci-e lanes to the CPU.  Maybe the MB uses these x4 pci-e channels to somehow let the peripherals communicate with the CPU by taking turns, like the old expanded memory model of DOS.

 

Simply put, I don't think a CPU can somehow increase the amount of pci-e connections its capable of, electrically, by happening to be put into a particular MB.

 

That's where I come down, and that's why I tried to advise the OP to be careful of how many peripherals are connected to the MB, because the system automatically creates new lanes by robbing Peter to pay Paul.  You can easily find the GPU reduced-down to a x4 pci-e connection.

 

Too, that's why I think the new AMD processors are such a big deal: they have (relative to Intel products) a very large number of electrical channels with which they can communicate to the MB.  The 3990X has, on the TRX40 chipset, up to 88 pci-e lanes; awesome stuff.

 

My OC i9-7960x system, which runs on the X299 chipset, has far more pci-e lanes than does any of the x370/x390 chipsets.  So I know I can run two x16 GPU and still have pci-e lanes to serve a couple of nice m.2 SSD and a wireless card or two.  The 9900KS simply can't do that and keep the x16 GPU slot going; it only has 24 pci-e slots to begin with.

 

See: https://ark.intel.com/content/www/us/en/ark/products/133293/intel-z390-chipset.html  

 

I do believe that AMD is eventually going to stomp Nvidia into the graphics dust, too, but until then, an AMD system, with a couple of the EVGA GeForce RTX 2080 Ti K|NGP|N GAMING cards plugged into the abundant pci-e lanes of a gaming TRX40 MB, is going to make gaming resolution restrictions a thing of the past.

 

See: https://www.amd.com/en/chipsets/str40

 

"...wanted to explore it with an open mind...": thanks for that.  That's why I tried to be clear that I wasn't being sarcastic or condescending to you.  I genuinely wanted to explore this problem in a civilized manner.  And so we have.

[ZachariasX]

58 minutes ago, mpdugas said:

I'm trying to see how a CPU, with limited electrical connections, gets more of them just because the MB has these additional pci-e x 4 channels into which peripherals can connect.

That is not how this works. You have (as you state 16 PCIe lanes that the CPU can use and they go to the PCIe slots where you put the GPU(s).

 

In case of Socket1151, you can use one slot for the GPU, dedicating all 16 lanes to it to maximize I/O between CPU and GPU. If you put in a second GPU, you can only allocate 8 lanes per GPU, effectively halving transfer speed between GPU and CPU.

You could also use 4 lanes to directly connect an NVMe SSD. This will also result in connecting your GPU with only 8 lanes. It will then allocate another 4 lanes to a third PCIe slot, where you could install another NVMe SSD.

 

On top of this, Intel uses DMI3.0 to connect the CPU to the Chipset. This is about equivalent to 4 PCIe lanes in terms of transfer speed. But this is exclusively the communication between the chipset and the CPU. DMI3.0 stops at the chipset. From there on, the chipset mediates the communication to USB, SATA, audio, network, etc. This means that there is no direct communication between a SATA drive (or the network) with the CPU. If all these devices want to talk to the CPU, they choke DMI3.0 very quickly. It takes one fast SSD to copy files, and DMI3.0 is saturated, meaning your network audio, USB communication is chocked.

 

On top of that, since the chipset has to hand everything to and from the CPU, you're adding latency. This is not so bad, as Windows tends to fetch one datum after the other from an SSD, rather than everything at once, where SSD's would actually be fastest. SSD's are much less good in just giving one datum after the other than handing all data at once, hence in real work conditions, you feel hardly differences between fast and much less fast SSD's, because they all can't live up to their ideals. It's obnly when you copy large files, that the world changes.

 

The X299 platform now has the advantage in providing 40 lanes (or so, depending on the revision), meaning you can plug in TWO GPU's and still have 8 lanes to spare to hook up NVMe SSD's directly to the CPU. On top of that, you have DMI3.0, that hooks up all the rest of your components.

 

However, it is now up to the mainboard maker to decide, whether he should hook up network, USB (USB3.1 is fast....) etc. with dedicated lanes. Usually, they decide against such reasonable layouts, as Intel is screwing everybody over. How does Intel do that: They vary the number of PCIe lanes between different Socket2066 CPU's. Hence, the mobo designer might actually confronted with Intel deciding to sell a Socket2066 CPU with ONLY 16 lanes. If now the network chips are hooked up with 4 lanes, the USB with four lanes as well, then this CPU would only be able to provide 8 lanes to a GPU and none to a dedicated SSD. Just to be on the safe side and not further losing business over Intel's upselling scheme, they play safe. This means even the powerful and expensive X299 platform is on purpose crippled and less than it could be.

 

Hooking up periphery directly to the CPU helps a great lot keeping a "stutter free" platform when doing the heavy work. Thus, Intels upselling scheme hurts the client directly.

 

 

 

AMD now hooks up always all lanes to the CPU's, meaning you can build much more powerful systems than it is financially reasonable to do with Intel parts. For free! It makes AMD so much more affordable than Intel, much beyond the CPU pricing. You are not required to use Intel peripherals should you want to hook them up to the CPU.

 

You really need to understand how fundamentally obliterated Intel’s whole platform scheme has become after the introduction of Ryzen CPU's. The "single thread performance lead" (that doesn't exist anymore) is irrelevant when you have to arrange your product such that an equal platform costs many times as much as the competitors platform. The CPU price is just one part of the puzzle. What Intel did was something like, e.g. "Apple computers ONLY can connect Apple USB drives". On the server front, this scheme was even worse. It is there, where this platform lock-in matters most and where it obliterates any offering Intel can make, effectively putting every new offer from Intel TCO under water.

 

We will not just see "better products" from Intel (they can last out 2 years of misery selling uncompetitive products), what we should see is Intel putting an end to much of his way of actively and deliberately hurting customers with their upselling schemes and vendor lock in practices. Single core performance is just one note in the chorus of bragging rights. Intel is hurt in the whole way it is selling products.

 

 

[CanadaOne]

That was a good post.

[SeaW0lf]

1 hour ago, ZachariasX said:

That is not how this works. You have (as you state 16 PCIe lanes that the CPU can use and they go to the PCIe slots where you put the GPU(s).

 

In case of Socket1151, you can use one slot for the GPU, dedicating all 16 lanes to it to maximize I/O between CPU and GPU. If you put in a second GPU, you can only allocate 8 lanes per GPU, effectively halving transfer speed between GPU and CPU.

You could also use 4 lanes to directly connect an NVMe SSD. This will also result in connecting your GPU with only 8 lanes. It will then allocate another 4 lanes to a third PCIe slot, where you could install another NVMe SSD.

 

On top of this, Intel uses DMI3.0 to connect the CPU to the Chipset. This is about equivalent to 4 PCIe lanes in terms of transfer speed. But this is exclusively the communication between the chipset and the CPU. DMI3.0 stops at the chipset. From there on, the chipset mediates the communication to USB, SATA, audio, network, etc. This means that there is no direct communication between a SATA drive (or the network) with the CPU. If all these devices want to talk to the CPU, they choke DMI3.0 very quickly. It takes one fast SSD to copy files, and DMI3.0 is saturated, meaning your network audio, USB communication is chocked.

 

On top of that, since the chipset has to hand everything to and from the CPU, you're adding latency. This is not so bad, as Windows tends to fetch one datum after the other from an SSD, rather than everything at once, where SSD's would actually be fastest. SSD's are much less good in just giving one datum after the other than handing all data at once, hence in real work conditions, you feel hardly differences between fast and much less fast SSD's, because they all can't live up to their ideals. It's obnly when you copy large files, that the world changes.

 

The X299 platform now has the advantage in providing 40 lanes (or so, depending on the revision), meaning you can plug in TWO GPU's and still have 8 lanes to spare to hook up NVMe SSD's directly to the CPU. On top of that, you have DMI3.0, that hooks up all the rest of your components.

 

However, it is now up to the mainboard maker to decide, whether he should hook up network, USB (USB3.1 is fast....) etc. with dedicated lanes. Usually, they decide against such reasonable layouts, as Intel is screwing everybody over. How does Intel do that: They vary the number of PCIe lanes between different Socket2066 CPU's. Hence, the mobo designer might actually confronted with Intel deciding to sell a Socket2066 CPU with ONLY 16 lanes. If now the network chips are hooked up with 4 lanes, the USB with four lanes as well, then this CPU would only be able to provide 8 lanes to a GPU and none to a dedicated SSD. Just to be on the safe side and not further losing business over Intel's upselling scheme, they play safe. This means even the powerful and expensive X299 platform is on purpose crippled and less than it could be.

 

Hooking up periphery directly to the CPU helps a great lot keeping a "stutter free" platform when doing the heavy work. Thus, Intels upselling scheme hurts the client directly.

 

 

 

AMD now hooks up always all lanes to the CPU's, meaning you can build much more powerful systems than it is financially reasonable to do with Intel parts. For free! It makes AMD so much more affordable than Intel, much beyond the CPU pricing. You are not required to use Intel peripherals should you want to hook them up to the CPU.

 

You really need to understand how fundamentally obliterated Intel’s whole platform scheme has become after the introduction of Ryzen CPU's. The "single thread performance lead" (that doesn't exist anymore) is irrelevant when you have to arrange your product such that an equal platform costs many times as much as the competitors platform. The CPU price is just one part of the puzzle. What Intel did was something like, e.g. "Apple computers ONLY can connect Apple USB drives". On the server front, this scheme was even worse. It is there, where this platform lock-in matters most and where it obliterates any offering Intel can make, effectively putting every new offer from Intel TCO under water.

 

We will not just see "better products" from Intel (they can last out 2 years of misery selling uncompetitive products), what we should see is Intel putting an end to much of his way of actively and deliberately hurting customers with their upselling schemes and vendor lock in practices. Single core performance is just one note in the chorus of bragging rights. Intel is hurt in the whole way it is selling products.

 

Your post seems to be incorrect.

 

From this discussion below, looks like the Intel 1151 chipset talks with the CPU by a different bus, not the 16 PCIe lanes, which usually are just for the PCIe lanes.

 

---

 

The CPU's have 16 PCIe lanes, and a "DMI" interface, which is basically a 4x PCIe link between the CPU and chipset, for example DMI 3.0 is roughly "4x PCIe 3.0".

 

---

 

Two slots are usually directly connected to the CPU via the 16 lanes you mentioned, the rest of the devices are wired to the chipset (up to 24 additional lanes), but the chipset itself is connected to the CPU through a thing called DMI 3.0, which is like a 4x lane PCIe bus. You can choose to reduce the lanes of your video card or go through the chipset.

 

---

 

If you want a direct to CPU connection for your NVMe then you must use a Hyper M.2 card which you must purchase separately. And doing so will limit your GPU to x8 lanes, which won't affect it unless you have some thing like a 1080Ti or 2080Ti and by only 1% or less. The Hyper M.2 card supports two M.2 drives if used in the second PCIe x16 slot.

 

If connecting directly to the CPU doesn't matter to you then you can use the motherboard's M.2 slot which connects to the chipset then to the CPU. Doing so means the Chipset<=>CPU link which is already x4 PCIe link will share both the NVMe drive plus all other things connected to the chipset including SATA, USB, Audio, some other PCIe slots.

It will only matter if you're using a high-end NVMe drive as it alone can saturate the Chipset<=>CPU link in some cases.

If you still want x16 lanes to the GPU while having the NVMe to the CPU directly then the only two solutions is to move to Intel HEDT platform or jump to AMD Ryzen CPU (not APU).

 

--

Edited December 14, 2019 by SeaW0lf

[ZachariasX]

12 minutes ago, SeaW0lf said:

Your post seems to be incorrect.

 

From this discussion below, looks like the Intel 1151 chipset talks with the CPU by a different bus, not the 16 PCIe lanes, which usually are just for the PCIe lanes.

That is what I meant to say. It has 16 lanes, PLUS 4 "lanes" that are made as DMI3.0 connection. You have both. (Also in Socket2066 you have DMI3.0 plus the PCIe lanes.)

 

Traditionally, when you put an SSD in a Socket1151 mobo (at least all the ones I took a closer look at), this SSD is then hooked up to the chipset and not directly to the CPU. I gave the example of a user choosing to hook up the SSD directly to the CPU, which you can in a manner I described.

 

Even on the Socket2066 boards, the onboard slots where you directly stick in the SSD is hooked up to the chipset. If you want to hook it up directly to the CPU, then you have to use specific adapter bards and place then in either dedicated slots or the PCIe slots.

 

[SeaW0lf]

1 hour ago, ZachariasX said:

In case of Socket1151, you can use one slot for the GPU, dedicating all 16 lanes to it to maximize I/O between CPU and GPU. If you put in a second GPU, you can only allocate 8 lanes per GPU, effectively halving transfer speed between GPU and CPU.

 

You could also use 4 lanes to directly connect an NVMe SSD. This will also result in connecting your GPU with only 8 lanes. It will then allocate another 4 lanes to a third PCIe slot, where you could install another NVMe SSD.

 

When you posted the above, the layman [myself included] will assume that you are saying that you can't use an NVMe SSD and use a GPU running at 16X, which is the same to imply that NVMes takes part of the PCIe lanes, which is not true, especially because to use the PCIe lanes you need to buy a PCIe adapter or use a PCIe SSD.

 

The posts that I quoted are clear in saying that M.2s usually talk to the CPU by the DMI interface, which is not misleading. I just went to spend time researching because you post implied that I could not use an M2 and a GPU running at 16X, that's all.

 

Honestly, I was a regular in a hardware forum for years, and the beef / bias / AMD / Intel is a pain in the but. The objective of these posts are to relay clear / impartial information.

 

At least I learned a bit more about the DMI interface elsewhere ???

[ZachariasX]

2 minutes ago, SeaW0lf said:

I just went to spend time researching because you post implied that I could not use an M2 and a GPU running at 16X, that's all.

It shouldn't have. Sorry for that.

 

2 minutes ago, SeaW0lf said:

The objective of these posts are to relay clear / impartial information.

I don't think I was "partial" here because out of some bias. If that is what you suggest. I mean, I'm typing this on an expensive Intel box. All I tried to do was explaining how this product works in terms of selling it.

 

I surely am not only partial about Intel’s way of doing it. I am actually infuriated. Not because I'm taking AMD's side. Or if I "prefer AMD". I'm taking my side, the side of the consumer. Your side actually. They screwed you too. I don't need to be "pro AMD" when I say that AMD does not follow suit in specific extortion schemes as a sales practice. AMD has now the more competitive offer, but this is (if I had to throw a guess) subject to change. I can't think of anything that AMD is not able fail on as well. (Boy they did many times!) AMD not neutering their products as the mean of choice for customer segmentation and vendor lock-in makes them actually even more competitive in the server market than the performance lead they have now. Just look what the price differences add up to. We're talking here about three, for fold prices for a similarly performing box. How happy would you be if you had to amortize your server like that?

 

But the actual situation we are in as consumers is not just "due to technical difficulties that we also soon will overcome" to finally get that great rig for flight sims. It is a result of blowing fuses to neuter a product such that you can just do one thing with your rig and not more. For everything more, you gotta pay again. This makes the average user end up with a significantly reduced feature set. Now why on earth should a game developer open the can of worms going all multithread when world and their dog run four cores? Now there's more than four cores, but think again why Intel did that. It's just because someone (could be anyone) did and made Intel look bad.

[SeaW0lf]

3 minutes ago, ZachariasX said:

I surely am not only partial about Intel’s way of doing it. I am actually infuriated.

 

I can see that, but it is always better to keep the posts impartial and complete, because people do take decisions after reading such things that appear to be coming from an educated opinion.

 

No fault, perhaps you did not understood how the DMI interface worked and your post ended up being misleading. At least it forced me to go look for information. I even opened a thread in a specialized forum, although that Reddit thread has the information laid out nicely.

[ZachariasX]

There's not much to not understand about it, but I should have posted that image earlier. It was clearly not a good idea from my side just explaining how you hook up an NVMe to your Socket1151 CPU as start.

 

Here's all there's about it:

 

The "three independent display support" is something you don't need. And where GFX are, there you can plug an NVMe SSD for direct connection. Set it in BIOS accordingly. But normally that goes through IRST/chipset.

 

Edited December 14, 2019 by ZachariasX

[simfan2015]

People who care about keeping their rigs as long as possible (in other words will invest in PC upgrades over the years) should try to evaluate and decide which platform has the best chance to stay relevant TODAY.

Important considerations are abundant (and surely not only socket, PCI version and market share).

I, personally, and of course YMMV, today I would go for AMD, but I am not really sure *any* choice we can make now will prove the right one when we could look back on it say ... 3 years from now !

Intel seems dormant now, but they sure came back in the past ! (AMD64)

I hope AMD can return with superb GPU offerings, further improve price/performance,  and thus make life for Nvidia ... miserable :-).

 

 

 

Edited December 14, 2019 by simfan2015

[mpdugas]

This discussion about DMI has failed to show me how its function increases the electrical connectivity of the CPU to the MB. I'm very interested in learning how, when the CPU is limited to 16 pci-e lanes, the function of the mysterious DMI can increase the CPU's electrical connectivity to the MB?  If it (CPU) has 16 electrical communication lanes available though pci-express, how can the DMI/MB increase that?  I'm not being sarcastic with this question; I genuinely want to know the answer.

 

As I understand PC connectivity, the 9900KS, for example, only has 16 pci-e lanes to communicate to the MB with.  Intel's data sheet shows the possible pci-e lane configurations with this chipset, which is here: https://ark.intel.com/content/www/us/en/ark/products/192943/intel-core-i9-9900ks-processor-16m-cache-up-to-5-00-ghz.html  and shows that it uses the LGA1151 CPU socket, connecting to the MB through an Z390 chipset : https://ark.intel.com/content/www/us/en/ark/products/133293/intel-z390-chipset.html.   This is the likely build-out for an enthusiast Intel build focused on single-core performance.

 

The LGA1151 is the large pin connector for the CPU/MB interface.  The socket type is inextricably linked to CPU type.  It has nothing to do with pci-e count.

 

The Z390 is the main chipset installed on the motherboard by the manufacturer of the main board.  The CPU has a limited number of pci-e lanes available to communicate with the MB;  in the case of the i9-9900KS, whose specifications are shown above, it can be configured as no more than 1x16, 2x8, or 1x8+2x4 pci-e lanes.  This is the primary reason why SLI/Crossfire is not a good choice for this CPU: each GPU would only get 4x pci-e lanes, shown in the last configuration possibility above.

 

My original point, for the OP's consideration, is that pci-e lane creep (meaning the PC BIOS automatically re-allocates pci-e lane connections when more peripherals are added) occurs without warning if peripherals are added to a PC, without being mindful of the total pci-e lane capacity of the CPU.

 

For instance, you can put a GPU in the x16 slot, in the case of the i9-9900KS, but as soon as you add a boot drive, which requires 4 pci-e lanes, the system automatically reduces the electrical connectivity of the GPU to x8 pci-e lanes, freeing up the four pci-e lanes needed for the SSD and leaving the unused four for another peripheral.  This happens in the background, without notice.

 

When I use my overclocked i9-7960X, I can connect two x16 GPUs in SLI/Crossfire and still have 12 pci-e lanes left for peripherals:  https://www.intel.com/content/www/us/en/products/processors/core/x-series/i9-7960x.html?wapkw=i9-7960x.

 

Thus, the CPU determines, in large part, what kinds of peripheral connectivity is available to the PC user.

 

My only contribution to his question was to forewarn him of that automatic pci-e re-allocation function.

 

[SeaW0lf]

9 minutes ago, mpdugas said:

This discussion about DMI has failed to show me how its function increases the electrical connectivity of the CPU to the MB. I'm very interested in learning how, when

 

From what I understood, it is a separate bus / controller that have nothing to do with the PCIe lane's bus. That's what the threads are saying. It would not make sense otherwise, which obviously left me scratching my head.

 

We could get to the bottom of it and see is the DMI bus can be bottlenecked and how bad it would be, but at least for gaming I don't think it is relevant, since you would have to really push both bus channels at the same time. I never streamed, so I'm not sure you overload the M2 while streaming, but I think not.

[mpdugas]

I appreciate your answer; my guess?  It is nothing we can directly influence, but we can be mindful of how adding peripherals affect the available pci-e count.

 

Thanks for the civil conversation; it's been very informative.

[simfan2015]

Very interesting conversation, Thank you!!! 

The new TR40 mobos should eliminate most of these issues!? But I fully agree that for gaming primarily much of these potential bottlenecks will not come into Play. 

[von_Michelstamm]

On 12/12/2019 at 8:17 PM, mpdugas said:

 

Let me think about that delicious prospect for just a little bit, ok?

 

I'll get back to you with something...

Have you had a chance to look into this? Thanks!

[mpdugas]

4 hours ago, von_Michelstamm said:

Have you had a chance to look into this? Thanks!

 

I actually have posted quite a few times, above, about your original question; do you require more?

[von_Michelstamm]

11 hours ago, mpdugas said:

 

I actually have posted quite a few times, above, about your original question; do you require more?

 

Perhaps i missed it, I was asking for your ideal component list to easily build a pc within a budget of $3500, which you said you would get back to me on.

[mpdugas]

2 hours ago, von_Michelstamm said:

 

Perhaps i missed it, I was asking for your ideal component list to easily build a pc within a budget of $3500, which you said you would get back to me on.

 

 

Check this out: https://pcpartpicker.com/list/MQq8Qq

This system is for the case and components only.

Be sure to read the comments I've made, above, for cautionary notes.

Edited December 18, 2019 by mpdugas
missing material

[Rei-sen]

So is Ryzen 7 PRO 3700 is the new king now (well, when it's actually becomes available)?
https://www.cpubenchmark.net/singleThread.html

Specifically for IL-2 GB VR I mean.

Edited December 21, 2019 by Arthur-A

[Dakpilot]

Not really because intel chips overclock, my 9600k spanks that passmark ST score at 5.2ghz, Ryzen does not really have much overclock headroom so it is not so simple to compare K series intel at stock clocks, the whole point of K series is its overclock 

 

Cheers, Dakpilot 

[mpdugas]

2 hours ago, Dakpilot said:

Not really because intel chips overclock, my 9600k spanks that passmark ST score at 5.2ghz, Ryzen does not really have much overclock headroom so it is not so simple to compare K series intel at stock clocks, the whole point of K series is its overclock 

 

Cheers, Dakpilot 

 

Nice article by Hardware Unboxed telling why finding over-clocking silicon in the 9900K series is going to be pretty poor pickin's; likely applies across Big Blue's ever-shrinking performance margins.

 

 

[SeaW0lf]

You have to contextualize. For simulators, the best bet is to get a 9th generation i5 or i7 and overclock it to the limit. Or even an i9 if you have the means. Silicon Lottery says that 100% of the i7-9700K gets to 4.9Ghz with 1.337v on SSE frequency. So if you don't have a bad day of shopping, you can get a chip clocked at 5Ghz with 1.35v or less. Or you can even buy a binned chip. They have i9-9900KS' clocked at 5.2Ghz and 1.32v with AVX Offset 2. Then you pair it with top of the line RAM and a good cooling solution and you are good to go for another 4+ good years of gaming.

 

 

Edited December 22, 2019 by SeaW0lf

[Jaws2002]

What's the difference between Ryzen 3700 pro and the standard Ryzen 3700x?   Cache, TDP, Overclock?  

3 hours ago, Arthur-A said:

So is Ryzen 7 PRO 3700 is the new king now (well, when it's actually becomes available)?
https://www.cpubenchmark.net/singleThread.html

Specifically for IL-2 GB VR I mean.

 

  The saddest part about this whole thing is that single thread performance is considered as the deciding factor in gaming performance. 

[dburne]

21 minutes ago, SeaW0lf said:

You have to contextualize. For simulators, the best bet is to get a 9th generation i5 or i7 and overclock it to the limit. Or even an i9 if you have the means. Silicon Lottery says that 100% of the i7-9700K gets to 4.9Ghz with 1.337v on SSE frequency. So if you don't have a bad day of shopping, you can get a chip clocked at 5Ghz with 1.35v or less. Or you can even buy a binned chip. They have i9-9900KS' clocked at 5.2Ghz and 1.32v with AVX Offset 2. Then you pair it with top of the line RAM and a good cooling solution and you are good to go for another 4+ good years of gaming.

 

 

 

Have my i9 9900k at 5.2 GHz on all 8 cores, no AVX offset, and 1.318v set it bios.

Performance for me in IL-2 VR is very good.

Sure hope I can get 4+ years out of it. I got 5 on my previous 4820k build. 

[SeaW0lf]

5 minutes ago, Jaws2002 said:

The saddest part about this whole thing is that single thread performance is considered as the deciding factor in gaming performance. 

 

For simulators I think this notion is not disputed. From all I read, Il-2 is mainly a single threaded application. The dserver, netcode is single threaded, then even if you are building a server, better to buy one of those unlocked anniversary i3's and get it to 5.2Ghz.

 

Here and there I hear people saying that they are working on a new netcode / dserver, but that might take years or never be implemented. Even if they get a four threaded netcode, you are still better off with an i5-9600K overclocked past 5Ghz.

[ZachariasX]

8 hours ago, SeaW0lf said:

Il-2 is mainly a single threaded application.

It is absolutely NOT. It spawns several threads. What happens is that one main thread uses most CPU time and this one will hit 100% core load much before the other do. How much the other threads are using CPU time depends on what is going on in your actual scene to be rendered. The fact that IL2 can spawn several threads makes it like many core CPU's more that "few core" CPU's. Not that you get much higher FPS, but you get less stutter and it becomes tolerant to drawing many things in your scene without the same degradation in performance.

 

An unnamed other simulation that I shall not mention shows this in most drastic manner as well. There, you take a "simple" plane and just fly without much else set in the scene and fly over towns and trees, then you will see that one thread is used 100% and all 15 remaining (or whatever threads you still might have available) are somewhere near 0% CPU load. Now do the same and fly a "complicated" plane and much other things that do things around and you suddenly see other threads using significant CPU time.

 

What you're seeing is that IL2 is limited by a single thread, same as in essence P3D and FSX, both of which are not single thread by any mean. P3D can even give you almost 100% CPU time on ALL threads while still being limited by the main thread!

[SeaW0lf]

6 hours ago, ZachariasX said:

It is absolutely NOT. It spawns several threads. What happens is that one main thread uses most CPU time and this one will hit 100% core load much before the other do. How much the other threads are using CPU time depends on what is going on in your actual scene to be rendered. The fact that IL2 can spawn several threads makes it like many core CPU's more that "few core" CPU's. Not that you get much higher FPS, but you get less stutter and it becomes tolerant to drawing many things in your scene without the same degradation in performance.

 

An unnamed other simulation that I shall not mention shows this in most drastic manner as well. There, you take a "simple" plane and just fly without much else set in the scene and fly over towns and trees, then you will see that one thread is used 100% and all 15 remaining (or whatever threads you still might have available) are somewhere near 0% CPU load. Now do the same and fly a "complicated" plane and much other things that do things around and you suddenly see other threads using significant CPU time.

 

What you're seeing is that IL2 is limited by a single thread, same as in essence P3D and FSX, both of which are not single thread by any mean. P3D can even give you almost 100% CPU time on ALL threads while still being limited by the main thread!

 

That's not what most people say. That Windows spreads the work around the threads, not the game.

[ZachariasX]

54 minutes ago, SeaW0lf said:

That's not what most people say. That Windows spreads the work around the threads, not the game.

 

IL-2 spawns several worker threads. Regardless of what "they" say. It is always the application itself that has to spawn the worker threads, Windows doesn't do that. Windows then juggles them around the available cores as you (or "they") say. And Windows sucks at doing that. It has only become slightly better with the latest version of Win10. Other than that it is so bad at it that in the HPC scene, they wouldn't touch Windows with a stick.

 

That said, IL2 spawns several threads, you can see then, even in the task manager. What Windows does with them is up to Windows, but as said, it has gotten better. The last addition to "multicore" was distant buildings. Still, you see them distributed on different cores.

 

It is often the case that threaded applications are limited by a main thread. Compare it to you running a factory. If you are alone (single thread) then you have to go and fetch each item and assemble that to a product that you hand out. If you multicore that, you can spawn workers that each fetch you an item and bring it to you. That makes you faster, more products per second. The more items you need for assembly, the more time you save for fetching them. But it does not scale that well. You always have to wait for the last worker until you can assemble the product. on top of that, in the end everything still depends on you, the main worker to get the product done ("scene assembly"). If you are saturated with the assembly, then more workers won't help. This is why every simulation really love one obscenely fast core and may use many cores, but normally put relatively little load on that one as it is usually the simple task that you are outsourcing.

 

You can see that well in P3D, where the max. FPS is only dependent on "single core speed". One main worker thread has to assemble the scene and that takes almost the same work, regardless of how many details you plan on drawing. (The GPU loads and draws then anyway.) The speed of the core running that thread gives your max. FPS. More cores allow you to send out more workers, meaning autogen can compute many more objects without impacting the main worker thread. But at some point the main thread is saturated. This is when "they" accuse your sim being "single threaded".

 

it is also the reason why "Turbo Boost" is a total dud for a sim. As you pointed out, Windows juggles threads around the cores. It always sends one task to a free core. As it happens, it relatively evenly uses the cores and loading them with very little work; just simple tasks. But the net result is Turbo Boost sees usage on all cores and clocks according to an "all core use" scenario. As long as you cannot cammand Windows to assign a specific thread (say, the main and limiting IL2 thread to a specific core, preferably with diversity and this single core is clocked highest, whle all the other cores are running 20% below that speed, which would be enough to cater the spawned workers. Not having to clock all cores to obscene frequencies would make it possible to run your 9900 at 5GHz+ for IL2 with an air cooler.

 

IL2 is not the only culpit here. Windows just lacks basic functionality to make efficient use of many cores. While "Turbo Boost" is great for marketing, is nice that you can clock your CPU dynamically to save power, but for games where it matters for performance, it fails for reasons described.  "Pros" are deactivating that functtion to permanently OC all cores evenly for best results.

[SeaW0lf]

Ryzen should smoke the Intel CPUs then, which is not the case. The fact that an Intel CPU overclocked past 5Ghz is the best CPU for all these simulators, Il-2 included, means that the application is still bound to very few cores and clock, or that one thread is overloaded, bottlenecking the flow. And the dserver / netcode is single threaded, mentioned for a few server owners that an unlocked i3 is enough.

 

I'm not an engineer, all I say is that the best CPU for these games is an 1151 Intel if you care about extra performance.

[ZachariasX]

50 minutes ago, SeaW0lf said:

Ryzen should smoke the Intel CPUs then, which is not the case.

Absolutely not. They suffer from the same. The only way to override that is having all cores collectively overclocked, this is what makes the Intel chips great. They either have very few cores with frequency headroom (not thermal) or there are some unicorn chips, known as 9900 series that allow that game up to 8 cores.

 

I have yet to test disabling cores on my 7900X, going at only 4 cores, I should be able to hit 4.8 GHz. I would bet I get better FPS in IL2. But I'd also expect better FPS at 4 cores forced 4.6 GHz than I have now.

[SeaW0lf]

I guess we are saying the same thing. Get an Intel and overclock it to play simulators.

[dburne]

44 minutes ago, SeaW0lf said:

I guess we are saying the same thing. Get an Intel and overclock it to play simulators.

 

[Rei-sen]

So no alternative to Intel for IL-2 in the near future?

[SeaW0lf]

17 minutes ago, Arthur-A said:

So no alternative to Intel for IL-2 in the near future?

 

You could look for some Ryzen 3000 series benches here in this forum or in DCS, but as far as I know, especially for VR, the best option is to get an i5/i7/i9 and overclock it. The i5-9600K has a good cost benefit. Somewhat cheap, six cores, which is enough for everything other than professionally work with rendering or photo / video editing, no hyper threading (I don't think Il-2 uses HT) and good temperatures, so if you live in a temperate country, you have a pretty good chance to overclock it to 5Ghz with a regular 240mm AIO and keep it below 70ºC in stress tests.

Edited December 22, 2019 by SeaW0lf

[mpdugas]

Be mindful of resources; the Intel 9900 series is quite limited.

Lots of links above about heeding this precautionary note...no free lunch, even with Intel single-core performance.

[SeaW0lf]

On 12/22/2019 at 6:08 AM, ZachariasX said:

It is absolutely NOT. It spawns several threads.

 

I just found this today: the game only uses two cores [the link is broken, they might have removed it]. I've found other posts saying that the cores usage is misleading. From this guy, and he's done bench posts before, the bottleneck is the main core being overloaded in VR, hence why an Intel at 5Ghz or more is the best option.

 

 

[mpdugas]

Perhaps running a VR benchmark, like the various UL VR Mark suites, and monitoring both CPU and GPU performance during the run will give a better indication of which components are most critical to good VR performance, rather than speculate. 

 

I'd suggest base-lining the hardware VR performance (with standard frequency timing settings) and then overclock the CPU and GPU separately while re-running the benchmarks.

 

The resulting scores can then be compared:

 

1) baseline hardware at stock settings, then

2) O/C GPU + standard CPU, then

3) standard GPU + O/C CPU.

 

Comparing these scores should give you a pretty good idea of whether the GPU or the CPU has the greater effect on over-all VR performance.

Edited January 4, 2020 by mpdugas
clarity

[Tonester]

The PC in my sig came in at just over 3200 AUD