yosarianilives

You should bench this for your country in country cup, don't think many countries will show up with 390x2

Damn, I wonder if we'll see a platform record on ln2

That's a fair stance, which also brings up the question of the "semi professional" chinese companies that make the permanently modded to lga 775 server chips although that probably will be left cause it's would be way too confusing with rankings. But ig to go back to my original question, if I mod an AGP or PCI to be natively PCIe it would have a huge advantage, especially on PCI but it would still be an AGP/PCI card because that's how it shipped retail.

If you want to make it relevant then just look at the question as "Should we treat non native AGP/PCI as AGP/PCI". Which if treated like cpus the answer would be no, as when cpus run non native sockets (771 in 775, 479 in 278, haswell bga on lga 1150, etc) they get treated as the native socket not the one which they were actually run on. However historically this has not been the case for non native gpus, personally I think they should be left as AGP or PCI if that's the out of the box configuration as at this point it would piss off too many people to change it to the native socket. But are they really any different than lga 771 cpus that have the pad mod soldered on and edges cut so they fit into a 775 board with zero modifications? Or the haswell chips with the conversion from BGA to LGA soldered on?

So I'm thinking over how the driver seems to handle seeing a PCI card with a bridge chip as just a PCIE card, and it makes me wonder purely hypothetically could you bypass the bridge chip at the hardware level and go directly to PCIe with essentially the PCI or AGP bus hanging to the side as almost a vestigial bus. While I don't expect or plan on seeing this during the comp it could in theory be possible and would raise the question of if it's even still an AGP or PCI card since you've bypassed the PCI/AGP portion of the card. I'd think at first it would be like a cpu adapter where the cpu is still treated as it's original socket, but you're not converting from AGP/PCI to PCIE, you're preventing a conversion from PCIE to AGP/PCI. But if we say that bypassing the pcb level adapter makes it no longer an AGP/PCI card then can we even say it was a AGP/PCI card to begin with as it's just an adapter built into the card to go from PCIE to the intended bus in basically the same way the interposers for mobile haswell chips bring them to desktop sockets. I suppose this has no relevance to the current comp anymore,especially since nobody is going to risk a rare gpu to try this myself included, but reading back through the discussion kinda brought up a head twist. I REPEAT, THIS IS NOT A LOOPHOLE GOD DAMMIT! SO PEOPLE DON'T FUCKING CUSS ME OUT FOR THIS SHIT

So apparently there isn't a vega pro duo but there is a fiji and polaris. The vega dual gpu card is the v340 which doesn't have proper drivers for an OS, it has drivers for esxi and basically is only made for VMs. Lots of other gcn based dual gpu cards from AMD that you haven't mentioned though

Wondered how long before a big S478 score came out on the comp

I think we all fear the day that there are members in the hobby younger than our own experience.

Indeed, not sure how people expect that being a dick will translate to the hobby growing. Of course there is the more sinister possibility that they want OC to be a members only club with basically no allowance for new members. I hope not because history has shown time and time again that mentality benefits exactly zero hobbies.

0.o

Really hoping to see good things from Greece, would be nice to have anyone dethrone AU. Certainly not holding out hope for USA as we seem to have a similar problem of the "big bois" not having much interest in this comp due to lack of modern HW. Even if you guys don't beat AU the more teams that put up a good fight, the more competition, and the more interesting and fun that the competition is.

Indeed, I believe 1440p resolution is the minimum for r20 or r15 extreme for that matter.

I think people are missing the point, for daily OC is pointless, cpus and cards come 95% of the way to max daily at ambient. Hell on newer ryzen cpus most people would actually see a decrease in gaming performance if they tried manually OC. Both NVidia and Radeon cards respond more to simply being cooled better than they do to attempts to OC. However that's not what this thread is about nor is this site really about that, this site is about pushing to the absolute max, where the stability test is can it just pass this bench who gives a toss about 24/7 stability. I think there are plenty of people who want to do that to some degree, many come in by seeing how their daily OC compares, but honestly they catch the bug and try to push to the absolute max or else they lose interest. If we want to preserve or expand the hobby then we need to find a way to make it appeal to more newcomers, plain and simple. Also Chew I think you've been on the site enough to know to not mind ozzie, and I recommend the rest do the same

Is that a hype train for the surprise UK upset of AU country cup dominance?

Damn intel's AVX2 putting in work against a clock deficient vs TR

I mean all hobbies have a sink or swim point I think. A point where you have to decide if you're willing to go whole hog into a hobby and invest heavily or if you're happy where you're at/don't want to continue further. Most people get into OC with seeing how their daily compares to others, then they may decide there's a platform they enjoy playing with etc, after that is where they decide if they want to go XOC or not typically.

@Leeghoofd Notice a bug in the configuration for ref clock stage, lga 1151 v1=v2 is not working. Haven't checked other countries but team USA has both a z270 and a z370 score counting towards our total. https://hwbot.org/search/submissions/permalink?ids=4291784,4293088,4272691,4271794

In the case of iris pro I'd say we can call edram system memory as the cpu still has access to it and benefits it, whereas hades canyon the hbm stack is truly dedicated to the gpu. Also the iris pro gpu is on die with the cpu currently. A further conundrum there's plenty of motherboard igps of yester year where they had basically a dedicated Ddr3 chip on board right next to the chipset so they don't use system memory but they're definitely an igp. So I think the best way may be to loosely define an igp as graphics on a chip that has multiple purposes OR graphics that use system memory in the case of potential chiplet design apus.

Interconnect on Clarksdale is dmi (or whatever the equivalent was then) , on motherboards its FSb or htt. Interconnect on hades canyon is pcie, if you soldered a mobile 1070 instead of using an mxm you wouldn't call that integrated etc

Does this mean that a retail mobile chip that's been put on one of these adapter substrates would be allowed and just not one that registers as ES? Or does that mean all mobile chips on desktop boards are considered non retail?

I mean the theme of this comp is if it's not explicitly disallowed it's probably allowed. And so if the socket that it's registered in on the db is allowed and the arch is allowed I don't see any reason it wouldn't be allowed but leeg would need to confirm.

That's because basically all modern "PCI" cards are actually native PCI-e at the chip level, it then hits a bridge chip to bring it to PCI protocal. Which is part of why any non native pci board will score terribly as you essentially are converting from pcie to pci then back to pcie using 2 bridge chips. To the driver it would probably only see the pcie bus that the bridgechip provides as the bridge chip probably never talks to the driver.

Is it actually slightly faster per clock or just binned harder?

Good lord, almost doubled the previous first.

Yeah I just did a single socket full population with the mem sticks we had for it, didn't even realize how much mem it was until later. Really didn't seem to mind running 3200. If I have time the next time I'm in the lab want to try 1dpc bdie. The bios gives dividers all the way to 4200 although I don't know how many work.