Originally Posted by Joa3d43 Part 3 Warning: Spoiler! (Click to show)
It also troubles me that the 'usual suspects' have attacked others, such as the new thread master at Valley, as a 'liar' for posting my more recent Valley runs. As I had already stated, I provided multiple runs (!) and other back-up inof (!) in a pm to him as well as two mods here given the giant previous mess in this thread !!! Only then did he load it. I since then have posted to more (higher) results, not least as I was able to get my Sandy-E to run 2666MHz memory.
What folks here don't know is that there was further PMs behind the scene, as some mods were genuinely interested in accommodating some of the critics who had otherwise contributed a lot - as such, FtW contaced me for some 'AA / no-AA' runs @ Valley.
Neither he nor I really did have an easy time to turn it off at various spots, but I finally managed to do it - providing FtW with 5 Ghz 'control' runs (the early days of the 2666MHz memory which was them still running stock - not anymore ;-) FtW has those 5 GHz comparative runs.
I add that I did provide those somewhat reluctantly because of the personal situation / time pressures I described above, but in the interest of 'Valley, I went ahead with it - even though I predicted (and was proven correct since then) that no matter what other runs I do, it will never be enough. Personally, I think someone pointing to over a hundred verified submissions in various disciplines at HWBot should be enough, but that's just me...
...there is a lot of other nonsense floating around in the latest Valley posts...let's start with the 7990s. I already mentioned that they run 16x 16x 16x 16x on my RIVE mobo, not the 16x 8x 8x 8x 4x 7970 in CF would require. At 5.1 - 5.25 GHz on a 6-core Sandy-E when you are trying to minimize bottle-necking by the CPU and maximizing throughput to the GPUs for their maximum usage, it makes a difference (as does 2666mem, or even better 2666mem with tightened timings).
Ditto for running a total of 2050watts PSUs for that system. FtW and others have also already - on multiple occasions - pointed out that at the bot, 7990s beat 7970 CFs regularly, even when the 7990s run lower clocks...a lot has to do with the on-PCB dedicated PCIe lanes and the latest, low-latency control chips...running a pair of those thus gives you 16x 16x 16x 16x on top of everything else with 2 cards...and apart from the bot, there are other tests (Hardware info) that show 7990s beating CF 7970 GHz edition even with lower clocks (Heaven 4 extreme is one example they gave).
Next, I have seen comments here (and by the same posters in other threads) that ALL 7990s all have high ASIC values, 70+ min. ABSOLUTE NONSENSE by those who don't actually own them. My 2x 7990 4x GPU ASIC values range from 59% to 67% or so - and I like it that way as I have been hunting for - and recently found - the ability to have UNLIMITED GPU voltage control in the BIOS...you want lower ASIC values for higher-v overclocks, not the other way around, as also stated by an AMD engineer in other threads on this.
My latest run - the 165.6 FPS one - came after I had worked with a BIOS specialist at another site and acted as beta tester to get control of the VRM section of 7990s which are utterly different (and IMO nicer) than 7970s as they use multi-phase Volterras. My 165.6 FPS run was just another iteration in a lot of time, dollars and efforts I invested in getting the 7990s to run at their full potential. They are also water-cooled with a dedicated loop that has 880mm x 60 mm rads JUST for the two 7990s, along with two high-flow pumps.
The new BIOS - still a beta with a few bugs - has nevertheless other special settings, including a PowerTune level of 'above 50%" and a wattage max number currently set to more than 2 1/2 that of stock per GPU...shortly, another version I will test for the author will have VRAM voltage control as well. So please, if you don't know anything about 7990s in general or own them, much less these uniquely prepped ones, don't make statements in public which don't apply...this was A LOT of work, not just on the GPUs but making sure they have ample power (per twin PSUs) and all that.
FtW for example knows all this as he has been trying to help me as I kept on bugging him about extra MSI AB control help...he also knows about the hundreds of hours I spent fine-tuning the system memory, and he is no fool (just like the other mods who spoke out here) when they see an Intel XTU sub at 'only' 5.125G which is at 1548 already...s.th like 15th worldwide on any kind of Hexacore, including super-cooled ones...this is about efficiency vs sheer brute speed...the latter which is yet to come from me on this system, but only AFTER I bring all other sub-systems to their max, and take care of an IHS issue...
...by that I mean that my CPU not only has an extremely unusual IMC (current world-record holder @ HWBot for memory frequency for 3970x, 3960x and 3930k hexacore Sandy-Es), I can also run strap125 / 5250GHz with ease at about 1.55v and VCCSA and VTT at 1.2v (their max I'm comfy with on short runs).
The problem is simply that the IHS is extremely concave - more so than any other CPU I have ever encountered in 20 years - and my 'normal' application of liquid metal tim CL-U won't work right, even when painted on double-sided....so for now, I have 'a lot of MX4' in there, helping up to 5.125GHz with another CPU-only massive w-c loop.
The solution will be a major lapping of the IHS with up to 2000 grid paper - but I only got this C$1100 CPU back in March or so, and wiping out stuff important for warranty and such is not an entirely easy decision, especially as I am interested in 4960X if I can find one that exceeds the 3970X's great performance.
But these will be my next steps at Valley - a further updated VID BIOS (this time up to 1.375v though there is no soft limit, just what the VRM and GPUs can handle), and the final speed step to 5250 (which I already have run on shorter test runs which is why I know that temps go above 85C until I fix the IHS).
In closing, this is not the 1st time that I have spent a lot of time drafting something like this. The Valley thread has been damaged by a few individuals for many months now who should stop doing that - and I certainly have the right to tell you to 'cease and desist' with any further insults, attempts to discredit honestly-achieved results, calling 'cheating' and all such unacceptable methods.
Furthermore, when you have a question on Valley, stick to Valley, not your (utterly false) conclusions on other threads, ie perfectly legit 3D11 postings. That is just a cheap high-school trick to amass as many doubts about someone's performance here as you can so that you can argue 'plenty more evidence'...repeating a wrong 15 times doesn't make it anymore right.
The saddest part is that there are some among this group I used to have a lot of respect for, give their technical achievements - but you see fit to destroy your own reputation with this nonsense.
To the rest of you folks, I am sorry you have to witness all this, I am open to answer rational, calmly posted technical questions. To the folks who have been defending me - thank you.