Who's buying Skylake-X? (You may now change your vote)

[Zucker2k]

Does your brilliant discovery also follow for Ryzen processors? If a Ryzen can rune P95 stable at 3.9Ghze, then is it able to run any other workload at 4.3Ghz?

I guess it also means that any Skylake - X that runs P95 stable at 4.4Ghz is capable of 4.8Ghz stable? Or does the magic frequency increase only apply to processors running P95 at 4.2-4.3Ghz?


I'm just trying to gain some of that great knowledge you have about processors, LOL!

He gave specific figures. An overclock of 4.6ghz that throttles to 4.2ghz under prime95(avx) loads. I gave him an answer based on the results I've seen so far. I was specific but I'll say it once more, a SKL-X chip running P95 (AVX) at 4.2ghz should draw more power and produce more heat than it would at 4.6ghz running typical everyday software. The 400mhz extra clocks may require subtle adjustments be made to settings/voltages in bios. His 7800x is in for a treat, 4.6ghz, rock stable, easy.

 

[Zucker2k]

With the new kit, my little 8 core broke 2100...

Impressive!

Edit: Could you disable HT, touch nothing else, and run that AIDA memory benchmark again for comparison? Pretty please. Nice numbers, by the way.

 

[TahoeDust]

4.9GHz @ 1.325v

Impressive!

Edit: Could you disable HT, touch nothing else, and run that AIDA memory benchmark again for comparison? Pretty please. Nice numbers, by the way.

Yeah. I'll do it in a bit.

 

[MarkPost]

With a 7800x, 4.6ghz should be a breeze. You should even be stable with the non-avx version of P95 at those clocks. The AVX-512 version of P95, however, I'm not so sure. So, unless you plan to run anything with AVX-512 code besides stress-testing, I wouldn't worry about being stable at 4.6ghz for the 7800x. That chip is clocking as high as 4.8ghz stable.

Hope you are right, I do a lot of x264 encoding. It will upgrade my second main rig, a beloved 2600k oced to 4500.

Unfortunately for the moment this rig will use a dual channel configuration. My concerns are with cooling and PSU... It will be cooled with a Prolimatech Megahalems (I think its compatible), using a 550W PSU (Silverstone Platinum). I'm not sure if both will be enough to do the job.

 

[Topweasel]

Does your brilliant discovery also follow for Ryzen processors? If a Ryzen can rune P95 stable at 3.9Ghze, then is it able to run any other workload at 4.3Ghz?

I guess it also means that any Skylake - X that runs P95 stable at 4.4Ghz is capable of 4.8Ghz stable? Or does the magic frequency increase only apply to processors running P95 at 4.2-4.3Ghz?


I'm just trying to gain some of that great knowledge you have about processors, LOL!

This is a bit of an oversimplification of what is going on but I think it will help you understand. A Ryzen CPU for example doesn't need an AVX offset for two major reasons. Ryzen hits its limits mostly in silicon. Without active refrigeration it just plainly won't run at higher levels than the what most people are seeing on overclocks (3.8-4.1 GHz). Whereas most of the AVX clocking intricacies of SL-X are mostly power and thermal limitations meaning if you had the power and the cooling you could overcome the limits.

So to get an idea of how SIMD instructions work, basically they are shortcuts. A CPU might have to go through let's say 11 stages in a pipeline for normal X86 instructions. You send a piece of code it takes 11Hz for the CPU/Core to process it. AVX, MMX, and SSE and 3dnow basically exist as single or near single stage pipeline specifically with that code and the answer sheet all in the CPU/Core. You send a piece of code that instead of saying "do this calculation" it instead says "pull up answer 1A of the answersheet". Assuming you understand that lets move on. With MMX, 3dnow, SSE and the like, the pipeline and the answer sheet were basically in line with the width of the normal pipeline for the CPU. AVX changes that. Without changing much else in the layout of the CPU AVX, AVX2, AVX-512 pipelines are 128bit, 256bit, and 512 bit respectively. In the past whatever clock speed and power usage that was OK on the normal pipeline also worked just as well as lets say on SSE. That meant outside the increased die size these SIMD instructions were free.

So here is where it changes with AVX. Since these are much larger width pipelines for handling these much wider chunks of code. There can be higher limit on clockspeed and when measured at the same clock speed it will use much more power. Just to give you an example, there was a statement at one point that the answer sheet for AVX-512 was the size of an Atom core within SL-X. That 512 support on SL-X accounts for almost 12% of the die size. AMD doesn't really have this issue because they only use 128 bit pipelines and combines them to do AVX2 (256 bit code) so in AVX2 it's half as productive as AVX. This means the AVX pipelines are closer in line with the the rest of the CPU design and power and clockspeeds don't have to be changed. On the 256b and 512b pipelines it gets farther and farther away causing the core to run hotter and hotter.

So Intel to make sure they were running as fast as possible on regular X86 instructions created and offset that we can now change. That means when the core is running X86 instructions it runs at the speed you set. When it starts to process AVX code it clocks down by whatever you have the offset set to. This means that if you can clock up 5GHz when running a burn in test without AVX, but it always crashes at 4.6GHz or higher with an AVX burn in test, you know that it is the AVX pipeline that is the limit and can set and offset to -500MHz. This means you will run at 5GHz when running x86 and as soon as it starts to do an AVX piece of code it will clock down to 4.5GHz for that core.

 

[TheF34RChannel]

Hope you are right, I do a lot of x264 encoding. It will upgrade my second main rig, a beloved 2600k oced to 4500.

Unfortunately for the moment this rig will use a dual channel configuration. My concerns are with cooling and PSU... It will be cooled with a Prolimatech Megahalems (I think its compatible), using a 550W PSU (Silverstone Platinum). I'm not sure if both will be enough to do the job.

For X299 that's not nearly enough imo. I'd get a higher wattage PSU (850W+) and at least an AIO - I wouldn't put an air cooler on it for a million bucks. It's just asking for trouble.

 

[MarkPost]

For X299 that's not nearly enough imo. I'd get a higher wattage PSU (850W+) and at least an AIO - I wouldn't put an air cooler on it for a million bucks. It's just asking for trouble.

850w? ummm my other main system (1700X oced to 3.8 or 3.9 with a Radeon RX480) runs happily with the same PSU, so keeping in mind it is a 7800x and that the graphic card will be just a Radeon R7 250 (I dont play games at all), a 550W PSU with a high efficiency (its the Platinum model) should be enough... i guess 😕

If not, I have other PSU without use right now, its a Seasonic 650W Gold... The Silverstone is the PSU installed with my 2600k.. so all cable management is done with nice cablemod cables btw 🙂 It would sucks if i have to change it 🙁

About cooling, I have a SilentLoop 280 too, but problem is that when i do typical light tasks I'm a silentpc fanatic xD and pump buzz annoys me. Well I'll try air cooling first and see if it works.. If not, i will have to install the AIO...

 

[GoNavy1776]

Ive toyed with the idea but the 10 core and up is just retarded in pricing. There is no reason an 18 core should cost 2000 for consumers. I get it ... enthusiast will... market is like sub 1% blah blah. It just sucks.

 

[IndyColtsFan]

Ive toyed with the idea but the 10 core and up is just retarded in pricing. There is no reason an 18 core should cost 2000 for consumers. I get it ... enthusiast will... market is like sub 1% blah blah. It just sucks.

I'd love the 10 core, but pricing isn't my obstacle - the temps are. I guess I'll end up going for the 8 core or waiting for Coffee Lake, even though it looks like it is being delayed.

 

[jmelgaard]

So i received a Core i9 7900x this weekend, and now it's just laying on my desk mocking me as the ASUS X299 Prime Deluxe was out of stock and the 8x16GB as well, so I don't even have a system to install it into yet, and I don't currently know how fast they can get the board, the RAM should be max 5 days... 🙁

 

[RichUK]

So i received a Core i9 7900x this weekend, and now it's just laying on my desk mocking me as the ASUS X299 Prime Deluxe was out of stock and the 8x16GB as well, so I don't even have a system to install it into yet, and I don't currently know how fast they can get the board, the RAM should be max 5 days... 🙁

Cancel your memory and motherboard order and buy elsewhere?

Simples, no?

 

[RichUK]

Ive toyed with the idea but the 10 core and up is just retarded in pricing. There is no reason an 18 core should cost 2000 for consumers. I get it ... enthusiast will... market is like sub 1% blah blah. It just sucks.

To be fair, I'd have thought 18 cores is really for professional use, not enthusiast.

Therefore the cost is more palatable, though appreciate threadripper should be a lot cheaper albeit only 16 cores.

 

[jmelgaard]

Cancel your memory and motherboard order and buy elsewhere?

Simples, no?

Nope, apparently 4x16 (64) GB kits is not in stock anywhere (locally), I guess they don't sell to many of those >.<... So unless I order abroad I have to wait 🙁...

I can find the board at one other dealer, it would cost more though but depending on the answer from where I ordered on how fast they will get it, I might just get it from there instead.

Edit: Ended up canceling the board and ordering it else where, so should have all by the weekend...

 

[GoNavy1776]

To be fair, I'd have thought 18 cores is really for professional use, not enthusiast.

Therefore the cost is more palatable, though appreciate threadripper should be a lot cheaper albeit only 16 cores.

Well anything over 4 cores could be argued to be professional since there is no need for more than 4 cores for anything else, games included. However, Intel is marketing this as Enthusiast. Not discounting your point of view in the least of course.

 

[GoNavy1776]

So i received a Core i9 7900x this weekend, and now it's just laying on my desk mocking me as the ASUS X299 Prime Deluxe was out of stock and the 8x16GB as well, so I don't even have a system to install it into yet, and I don't currently know how fast they can get the board, the RAM should be max 5 days... 🙁

I would return tour CPU. Wait on Tripper to drop and that chip might drop in price if Threadripper poses a threat.

 

[ManyThreads]

You're comparing by price. I checked Newegg and there were only 3 X299 boards cheaper then than Crosshair and 1 X370 that was more expensive (entry-level X299 vs top end X370). I think if you wanted a better picture of the price premium, you would compare motherboards in the same family, for example X370 Taichi, Z270 Taichi, X299 Taichi.

I see what you're saying, but I am comparing what the premium would be *for me* because that's the only thing I care about price-wise, so it was CH6 vs the X299 selection. If you wanted to compare entry level to entry level, the gap would be larger in favor of Ryzen, no doubt, but I was not interested in those $90 B350 boards. What I really wanted is the better multi and especially single core performance for my specific usage (heavy photo editing with a mixture of single and multi threaded plug-ins). Since the premium to go X299/7820X wasn't that much, it was an easy decision for me given the performance advantage I would see for my work.

 

[Topweasel]

I see what you're saying, but I am comparing what the premium would be *for me* because that's the only thing I care about price-wise, so it was CH6 vs the X299 selection. If you wanted to compare entry level to entry level, the gap would be larger in favor of Ryzen, no doubt, but I was not interested in those $90 B350 boards. What I really wanted is the better multi and especially single core performance for my specific usage (heavy photo editing with a mixture of single and multi threaded plug-ins). Since the premium to go X299/7820X wasn't that much, it was an easy decision for me given the performance advantage I would see for my work.

But even within that there is still a big difference. Like what's the difference that would make a C6H a better purchase over a Taichi? Or what does either of those two do that you would have needed a an X370 board over a B350?

I am not saying you did wrong or that your performance on a 7820x isn't worth it. There is a great chance that it is. I just see a lot of people comparing 1800x+C6H pricing vs. the 7820x and cheapest x299 setups to show the little difference. Yet talk about overclocking (not saying you have) on the 7820x. It's just comes off a disengenous and bothers me a bit.

Here you don't name why the B350 wouldn't work and choose the most expensive board. But then talk about needing cores and speed and nothing else. Where in theory if that was the real concern a person could do a 1700, clock it up to 3.8 on a B350. Get 85-90 the performance for almost half the price.

If that 10% makes all the difference in the world. It sounds like it might. Then why not just state that instead of fudging the numbers to make it seem like a small upgrade price? I mean in the end it probably shouldn't matter if you want the fastest 8c CPU the 7820x is by far the best option.

 

[GoNavy1776]

Im anticipating Thread ripper 16 core at $895 or heck even less haha. I hope the performance is better than Intels 10 core 7900x or otherwise there is no reason to waste money on the AMD part. I am talking multithreaded. I never and still do not expect single thread to surpass Intel. Maybe in 3 mote gens of stellar Ryzen like comeback engineering.

I really hope the performance surpasses Intel to the point of Intel having another price war. It's good for we the people.

According to leaked benches it is far slower. But that may all just be negative propaganda generated by Intel. When your dealing with billions In sales wars it might as well be a damn James Bond plot between the two.

 

[Topweasel]

Not sure this is the thread for that post.

 

[Zucker2k]

But even within that there is still a big difference. Like what's the difference that would make a C6H a better purchase over a Taichi? Or what does either of those two do that you would have needed a an X370 board over a B350?

I am not saying you did wrong or that your performance on a 7820x isn't worth it. There is a great chance that it is. I just see a lot of people comparing 1800x+C6H pricing vs. the 7820x and cheapest x299 setups to show the little difference. Yet talk about overclocking (not saying you have) on the 7820x. It's just comes off a disengenous and bothers me a bit.

Here you don't name why the B350 wouldn't work and choose the most expensive board. But then talk about needing cores and speed and nothing else. Where in theory if that was the real concern a person could do a 1700, clock it up to 3.8 on a B350. Get 85-90 the performance for almost half the price.

If that 10% makes all the difference in the world. It sounds like it might. Then why not just state that instead of fudging the numbers to make it seem like a small upgrade price? I mean in the end it probably shouldn't matter if you want the fastest 8c CPU the 7820x is by far the best option.

Yeah right! Nice straw man you got going there, buddy.

 

[Topweasel]

Yeah right! Nice straw man you got going there, buddy.

I am not sure what I said or did wrong. I was just asking for some clarification because when the topic comes up it is always C6H+1800 and not any other alternatives. Even when comparing other high end X370 boards the C6H is still $100 more expensive. I would be like me saying I am not interested in SL-X because between a 7900x and Asus Prime Deluxe it's 4 times more expensive than a 1700 and a Biostar B350GT5.

Look I am not trying to rain on anyone's parade. If it's a better buy because of workload. It's a better buy. But I just don't understand without knowing what the actual limitations were that required the most expensive motherboard and CPU combination if Ryzen was ever a possible solution outside the stated " was not interested in those $90 B350" and therefore would have had to get the most expensive (while not doing the same with SL-X).

 

[GoNavy1776]

Not sure this is the thread for that post.

Sure it is. The question is who is buying the x299 platform. My response is simple. If threadripper fails to deliver the goods then yes I am buying x299. Unless you were referring to someone else. You didn't quote anyone.

Anyways if you or anyone goes with the 10 core I highly recommend full blown water cooling. These apparently run smoking hot. All in one's may not have the grunt necessary.

 

[IEC]

Anyone seen any BOINC results for SKL-X? I haven't found any takers. Apparently no one wants to (ab)use their shiny new hardware and take one for the TeAm... 🙁

 

[tamz_msc]

Anyone seen any BOINC results for SKL-X? I haven't found any takers. Apparently no one wants to (ab)use their shiny new hardware and take one for the TeAm... 🙁

What project(s) do you run?

 

[IEC]

What project(s) do you run?

A 72-hour sample across different projects such as LHC@home, Rosetta@home, Yoyo@home, etc with power consumption #s would give me a pretty good idea of PPD and relative performance/efficiency versus adding more R7 1700 nodes. I'm planning on expanding my DC farm in the fall when I'm not as thermally limited (due to GPU farm not in sig + hot summer). Having a better idea of how Skylake-X compares to Ryzen for DC/crunching work would be nice. For new BOINC CPU hardware the R7 1700 is hard to beat for both efficiency and perf/$.