Yeah, the EK TR block is lazy engineering. They reused their existing supremacy block and extended the cold plate. I initially wanted to wait for the Heatkiller TR block but it may be the end of the month before orders go live and I couldn't wait that long. The EK block does a decent job though or maybe it's because I have crazy radiator capacity I'll wait for some reviews on these blocks before thinking of swapping.Yeah that contact plate size looks like a joke... Noctua did this better.
Yeah, their water block looks like it has the best fin surface I've seen. It really looks like a beast :
Looks like the fin surface area is nearly double what i've seen even on the EK block. It looks like the EK block simply modified the base plate w/o modifying the fins. Best waterblock i've seen so far. Question i have for all of these is the flow pattern.. So, it flows from L->R or R->L, wouldn't one die get hotter water above it than the other? not that it will impact things much but still. Very impressed with the phantek block. This is the tiny EK block fin surface area for comparison :
Not yet releasedAnyone using the Swiftech Apogee TR4 block?
Not yet released
Last I checked they opened it to preorders. Don't think they have started shipping as yetI thought they had finished and were starting to ship already. Or were those the other ones?
Test it thoroughly. I can get 3600Mhz boom up and booting and then see stability issues only under hard stress tests. 3600@CL16 is even stable at mild stress tests .. When I pushed it hard, I get full lockups, bit errors, and I sometimes get post loops. I might see if a different kit works but I'm sitting on Samsung B-dies rated for 3600@CL16. Please post your memory details (product link preferred) also include your timings.Well I read all the hype on ddr 3600mhz ram. So I bought the gskill 3600mhz ram. I put the ram in and set the profile in the bios to 3600mhz speed and boom I was up and running. I still have my 64gb 3000mhz corsair kit .. I want to see if it is worth it to have 3600mhz over 3000mhz ram. We shall see.. I will report back
Yeah that contact plate size looks like a joke... Noctua did this better.
Their water block looks like it has the best fin surface I've seen. It really looks like a beast :
Looks like the fin surface area is nearly double what i've seen even on the EK block. It looks like the EK block simply modified the base plate w/o modifying the fins. The Phantek looks like the best waterblock i've seen so far. Question i have for this waterblock is the flow pattern.. So, it flows from L->R or R->L, wouldn't one die get hotter water above it than the other? not that it will impact things much but still. Very impressed with the phantek block. This is the tiny EK block fin surface area for comparison :
Although the water in/out is in the center, you'd still have an up->down or down->up flow having hot water from one die flow over the other and then out.
I was unaware of this, I was wondering what that plate was for... Do you have a link to any animations that depict how the flow works across jet plates vs. other designs? I'm curious and want to know moreEK's jet plate design is still superior to the point that practically everyone has copied it. There's also the matter of fin width and channel spacing which is finer on the EK giving it more surface area. Also, I''m not a fan of the rudimentary inlet to outlet on that Phanteks. Without a jet plate design you will end up with hot and cold sides to the block. That's why all the major true watercooling companies have moved to a similar jet plate design, whether it be an actual jet plate or a integral jet design.
Will try this right now and post back with my resultsALERT : Found root cause of inability to push RAM to 3600@CL16-16-16-36
2x (2x8GB 3600@CL16 kits)..
One kit is not up to snuff and is causing the bit errors.
I bought 2 independent kits of : https://www.newegg.com/Product/Prod...kill_tridentz_rgb_3600-_-20-232-491-_-Product
When all 4 are installed I can run 3200@CL-14-14-14-34 and 3433@CL16-16-16-36.
Unstable bit errors when I run them at 3600@CL-16-16-16-36. So I began investigating today.
@FiLeZz's post about successful 3600 out the box finally caused me to rage hard enough...
3600@CL-16-16-16-36
I had one kit to each side : This was the most unstable. Would get post loops and would get lockups quickly when I tried to run a stress test. In this configuration, each CPU had a mix of sticks. I then tried one kit spread across to the respective slot so there was one kit per CPU. In this configuration, no post loops occurred and the bit error didn't happen until 8min into the stress test. At this point, I figured it was one of the kits that were bad. So, I yanked both kits and ran one kit (16GB - 2x8GB). One of the kits hit a bit error after 7-8min in the stress test.. The other passed it with flying colors.
Tomorrow, I'm going to talk with newegg customer service on how to best handle this. Either RMA one kit or send them both back and get a proper quad-channel kit. Thing is, I better not have this issue with a quad channel stick and have to do QA for G.skill again . It's sort of why I bought 2 just so I could RMA only one kit if there was an issue. Sorta pissed right now but happy I isolated the ram stick kit that is the issue.
I urge anyone having similar issues to run in 2 stick configuration one kit at a time and see if its a particular kit you have that's the issue. Remember to look at the id # on the stick when yanking and replacing them. Matching kits will have sequential id #'s.
I was unaware of this, I was wondering what that plate was for... Do you have a link to any animations that depict how the flow works across jet plates vs. other designs? I'm curious and want to know more
Ok from what I can tell, one kit of RAM absolutely refuses to post at 3600. It will cause the machine to bootloop until the mobo resets it back to 2133. The other kit posts 9/10 times at 3600 and will boot loop once. Both kits work flawlessly at 3466 independently. I did not stress test the one that posts at 3600 to check for stability within windows as yet.
I know I said mine were stable, and they are under hard stress sometimes for days. Then I might get 2 reboots in a day, and the crash analyzer says its driver related. MAYBE mine are marginal also, The next time it reboots, I will try the 3466 setting.Ok from what I can tell, one kit of RAM absolutely refuses to post at 3600. It will cause the machine to bootloop until the mobo resets it back to 2133. The other kit posts 9/10 times at 3600 and will boot loop once. Both kits work flawlessly at 3466 independently. I did not stress test the one that posts at 3600 to check for stability within windows as yet.
RAM clock settings.
Did some poking around on the interwebs to see if this is widespread and it seems so :
https://www.reddit.com/r/Amd/comments/6upe0s/any_threadripper_memory_compatibilityperformance/
https://www.reddit.com/r/Amd/comments/6x7mx9/1950x_and_memory_help/
I hear the number 3466 being expressed as a ceiling with instability at 3600 or having to do memory training at this setting. This coincides with all of the mobo QVL lists, my observations and other users here, as well as g.skills own customized kit :
https://www.gskill.com/en/press/view/g-skill-announces-new-ddr4-specifications-for-amd-ryzen™-threadripper™-processors
They've even loosened the timing on their own Threadripper specific kits from 16-16-16-36 to 16-18-18-38. The reasoning would likely be that it was hit or miss to get ram to truly run at 16-16-16-36 on Threadripper at 3600. For me, one two stick combo ran fine at 3600(16-16-16-36) through stress test. Another had post loops and hit errors. Both were stable at 3466@16-16-16-36 and 3200@14-14-14-34.
Making a decision as to whether or not to RMA tomorrow or just stay at 3466@CL16. I feel like, given that it's a lottery with these kits that I just might get another kit that has the same issue after having gone through an RMA, had an extra charge on my card and dealing with foolishness. So, I'm kind of thinking its not worth the hassle. Word is on other forums that many define 3600 as an unstable setting no matter what the kit. 3600-3466 = 133 .. I like the way 3600 rolls off the tongue but I can't sacrifice stability for it. My aim is a rock solid, no bit errors, high uptime system at mild OC settings. I'm backing off 4.0Ghz to 3.7Ghz so I think i'll back off 3600 to 3466 on the ram too. Lower power/rock solid stability...
Quick thoughts on this logic : If I RMA the 2 stick pack, i might get another with the same 3600ceiling issue... G.skill wont even sell threadripper specific kits at these timings, others report instability at 3600... Oh and, although I didn't hit any errors in hardcore testing on one of the 2stick packs, I did get a boot loop after I reset a couple of times later... So, 3600@CL16 just seems no bueno. Maybe bios updates stabilize it. Maybe it just is unreachable without diminishing timings.. But, for 133mhz more, it seems to require too much
Ok from what I can tell, one kit of RAM absolutely refuses to post at 3600. It will cause the machine to bootloop until the mobo resets it back to 2133. The other kit posts 9/10 times at 3600 and will boot loop once. Both kits work flawlessly at 3466 independently. I did not stress test the one that posts at 3600 to check for stability within windows as yet.
I don't know if there's an animation showing this, sorry. But its easy to follow the flow of water. I'll try to describe is as best I can. In the Glacier block, it looks like cold water travels from one side of the block (inlet) to the other in one direction(outlet). This fundamentally creates a cold side and hot side. The cold side gets fresh cool water, which then warms up as it travels across to the other side of the coldplate. The warmed water will carry less heat since its already carrying heat from the cold side. The major brands have been using central jets since like the mid 2000's. A jet plate design forces cold water over the center of the coldplate. The cold water moves from the center outward so the block is cooled evenly. A peripheral channel allows the warmed water to exit the block. The best blocks today all employ this principle or some variation of it.
RAM clock settings.
Did some poking around on the interwebs to see if this is widespread and it seems so :
https://www.reddit.com/r/Amd/comments/6upe0s/any_threadripper_memory_compatibilityperformance/
https://www.reddit.com/r/Amd/comments/6x7mx9/1950x_and_memory_help/
I hear the number 3466 being expressed as a ceiling with instability at 3600 or having to do memory training at this setting. This coincides with all of the mobo QVL lists, my observations and other users here, as well as g.skills own customized kit :
https://www.gskill.com/en/press/view/g-skill-announces-new-ddr4-specifications-for-amd-ryzen™-threadripper™-processors
They've even loosened the timing on their own Threadripper specific kits from 16-16-16-36 to 16-18-18-38. The reasoning would likely be that it was hit or miss to get ram to truly run at 16-16-16-36 on Threadripper at 3600. For me, one two stick combo ran fine at 3600(16-16-16-36) through stress test. Another had post loops and hit errors. Both were stable at 3466@16-16-16-36 and 3200@14-14-14-34.
Making a decision as to whether or not to RMA tomorrow or just stay at 3466@CL16. I feel like, given that it's a lottery with these kits that I just might get another kit that has the same issue after having gone through an RMA, had an extra charge on my card and dealing with foolishness. So, I'm kind of thinking its not worth the hassle. Word is on other forums that many define 3600 as an unstable setting no matter what the kit. 3600-3466 = 133 .. I like the way 3600 rolls off the tongue but I can't sacrifice stability for it. My aim is a rock solid, no bit errors, high uptime system at mild OC settings. I'm backing off 4.0Ghz to 3.7Ghz so I think i'll back off 3600 to 3466 on the ram too. Lower power/rock solid stability...
Quick thoughts on this logic : If I RMA the 2 stick pack, i might get another with the same 3600ceiling issue... G.skill wont even sell threadripper specific kits at these timings, others report instability at 3600... Oh and, although I didn't hit any errors in hardcore testing on one of the 2stick packs, I did get a boot loop after I reset a couple of times later... So, 3600@CL16 just seems no bueno. Maybe bios updates stabilize it. Maybe it just is unreachable without diminishing timings.. But, for 133mhz more, it seems to require too much
They are working on TR specific tested ram.I am not going to RMA mine. The hassle is not worth it and like you said it's possible you will get another stick that doesn't run 3600. It's likely Gskill only tested these kits on Intel mainstream parts where 3600 is easily achievable.
This is an incorrect understanding on fluid dynamics and what is happening. Yes, technically correct, but we are talking fractions of a degree, a degree tops, on water temp between the in and out on the block, which is found on ANY block, including with jets. Yes, it takes a little longer to transverse, thereby increasing time of contact, picking up more heat, thereby having more heat, but that is how water cooling works! Stop the scare mongering. Yes, there are better designs, but that DOES NOT mean other designs are bad.
Water delta is different from core delta.Scare mongering? What? No brand that matters is still producing blocks like that today. And the difference between a rudimentary block like that from 2005 vs 2017 is not fractions. Are you a CFD guy or something? WTF?
Water delta is different from core delta.
Edit: also, you are attributing to the jet all benefits of other design changes in the past 12 years, which makes no sense. You also do not account for flow rate in the system, etc. So, yes, scare mongering.
AJC9988 is correct. The difference between the water at different points is at most fractions of a degree. A computer system is not like a car that can cause massive deltas in water temperature due to high temperatures. The water moves so quickly in your loop, the differences are negligible. Also, one of the reasons why order of components in your loop doesn't matter (except reservoir to pump).