2016+ CPU/GPU News

[revelator]

gah... ok makes sense in a way now 256 bit bus and back then at the chip shortage time micron could only deliver gddr6x with 1 gb pr chip = 32 bit x 8 = 256 bit bus. sadly while the 3090 which was a hotplate in its original configuration with vram on both sides of the card, got 2 gb chips when they became avaliable so they could move them to the front and get rid of the heat issues the 3070 newer got the same chips. to make matters worse the stock 3070 has gddr6 which has only half the bandwidth of gddr6x but atleast the ti version got stocked with gddr6x chips so at high resolutions it should make a dent but at lower resolutions they performed to close to each other to really justify upgrading.

[jaxa]

The original sin of the RTX 3070 is the VRAM amount, not the bandwidth. 7600 XT 16 GB has half the bus width and 64% the bandwidth (288 vs 448 GB/s), and even it can be made to benefit from the doubled VRAM.

You're going to see some RTX 5000 cards with 256-bit bus and 16 GB, 192-bit and 12 GB. Though it will be faster GDDR7.

Edited May 8, 2024 by jaxa

[revelator]

for the original 3070 id agree it was the ammount though the increased bandwidth in the ti models should help in 4k hell even the old AMD fury-x could at the time run 4k with only 4gb due to the hbm memory being that much faster to swap but yeah theres more than one way to skin a rhinocereus . a 3070 with 2 gb gddr6x 16 gb would have been pretty interresting to say the least .

[revelator]

looks like a foregone conclusion but several game studios have now enabled framegen via AMD's FSR3 on nvidia cards from previous generations (it had to come). a bit bad for said framegen mods who will now face obscurity except for a small number of titles who wont patch in support maybe ?. the framegen mod for pre RTX and pre FSR 3 cards like the 5700xt will probably still be needed since none of these cards support DLSS anyway and so are pinned for FSR 2 to allow using raytracing.

[jaxa]

From what I hear, AMD's frame generation implementation is considered largely superior to Nvidia's, while they still lag in upscaling image quality, but have improved that as of FSR 3.1.

[revelator]

aye so i heard ghost of tsuchima was allegedly the first game to add support for AMD's framegen and is said to run really well with it.

there is also talk about lossless image scaling in regards to AMD though from what i heard its still some ways off in quality but better than what was.

nvidia has been actively deleting posts on reddit and several other sites to stem the rumours but shooting themself in the foot instead when it became public knowledge that it was indeed working .

[revelator]

nukems framegen mod has also been updated for FSR3.1 though vulkan support is not completely supported yet (from what i could read on the github site he does not have the knowlegde to get it fully supported and it might newer be added unless someone with a good understanding of vulkan takes over that part).

this unfortunatly means that it might newer come to pass with more and more game companies adding there own support.

i tested it with alan wake 2 and horizon zero dawn and it works well enough atleast.

[jaxa]

https://www.anandtech.com/print/21469/amd-details-ryzen-ai-300-series-for-mobile-strix-point-with-rdna-35-igpu-xdna-2-npu

https://www.tomshardware.com/pc-components/cpus/amd-deep-dives-zen-5-ryzen-9000-and-strix-point-cpu-rdna-35-gpu-and-xdna-2-architectures/

More details about Zen 5 CPUs and APUs are out ahead of release.

+16% IPC increase which will need to be independently verified as usual, possibly +20% in games. Power efficiency improvements mean that you are getting more performance with less power usage and temperature, at the same time, not either-or.

Zen 5 has a full implementation of AVX-512, not "double pumped" 256-bit like Zen 4.

You can expect a 20-30% increase from 890M iGPU over 780M. It can outperform an RX 6400 and other weak discrete GPUs.

Edited July 15, 2024 by jaxa

[jaxa]

Time to talk about the elephant in the room.

Intel Addresses Desktop Raptor Lake Instability Issues: Faults Excessive Voltage from Microcode, Fix Coming in August

Intel finally announces a solution for CPU crashing and instability problems — claims elevated voltages are the root cause; patch coming by mid-August [Updated]

Buggy microcode in 13th and 14th gen Raptor Lake desktop CPUs caused incorrect increases in voltage, which led to permanent and irreversible degradation of the chips. Now all of the impacted CPUs will be suspect on the used market.

[revelator]

the ghost of meltdown strikes again this time it will really melt down

[jaxa]

This could easily become Intel's greatest disaster in the last 10, 15 years. They are signalling that they won't do a recall. I smell a class action lawsuit.

In other news, reviews of AMD's Strix Point for laptops have been coming out:

https://www.anandtech.com/show/21485/the-amd-ryzen-ai-hx-370-review

Zen 5 desktop CPUs in 1-2 weeks. Gains should be relatively minor, but power efficiency is good, and the 7800X3D will still be the best choice for gaming until the 9800X3D comes out.

[revelator]

https://www.techspot.com/news/104010-intel-crashing-cpu-crisis-deepens-more-models-affected.html

whoops... the rabbit hole just got deeper.

[revelator]

tbh silicons role is largely outplayed by the ever increasing demand of smaller and faster, we are allready at the end of how small a silicon wafer can be made because of quantum tunneling and heat. there is some work being done on a new material called cubic boron arsenide which can be made as low as 3 atoms thick and has 10 times the thermal conductivity of silicon, the downside is that this is not a natural resource as it can only be created in a lab so atleast for the initial chips the development price might put of most people. another excellent material is graphene which also is a lab created material but which can allready be made in huge quantities and has only slightly worse thermal conductivity.

carbon nanotubing is also speculated upon.

right now graphene seems like the best candidate but even better materials might be found in the not to distant future and would possibly make quantum computing avaliable to us mere mortals without the need for bulky superconductors.

[jaxa]

I don't think the silicon scaling issues are at play here. We know there's other types of transistors, such as GAAFETs and QFETs, that could be less susceptible to quantum tunneling. They are being deployed or are on the roadmaps or in the lab. Carbon nanotube transistors or other advances will take over in 10 years or so. After we run out of obvious things to improve, there will be a complete shift to 3D scaling.

The problem is the company culture at Intel, and the attitude that they must be #1 at any cost. They rushed Raptor Lake as a refresh product, and got burned by faulty microcode and other issues. They own fabs but have struggled to open them up to customers other than themselves. They are burning billions to get into discrete GPUs, after completely missing the peak of gaming GPU pricing, and may not establish good AI revenue before that bubble pops. All sorts of problems.

BTW the class action is here:

https://abingtonlaw.com/class-action/consumer-protection/Intel-Processor-Issues-class-action-lawsuit.html

[revelator]

probably wont happen tomorrow or the day after but things are stacking up in the wrong direction for silicon so at some point in the next 5 to 10 years it will have to give way. 3d scaling is a good way of stacking more into current chips but also has the downside of being even more heat sensitive as seen with the first am5 processors who literally burned a hole through the metal heat sink because of mainboard producers wanting to skew figures feed them slightly over the top vcore values (incidently this was the first time a cpu considered ready for overclocking had melted from this practice and previous cpu´s tolerated it), strangely this seems to allmost the same problem intel has now at first glance, but since non overclockable intel versions are also hit by it i reckon a rushed release might indeed have resulted in this.

yeah its hard to fall from the top knowing intel they are probably rolling out the big nasties for the competition (legal or not).

[jaxa]

https://wccftech.com/intel-extends-warranty-boxed-14th-13th-gen-cpus-two-years-instability-issues/

“We've been trying to reach you about your CPU's extendead warranty"

[revelator]

a few examples of the am5 debacle https://www.tomshardware.com/news/amd-ryzen-7000-burning-out-root-cause-identified-expo-and-soc-voltages-to-blame luckily a bios update solved it in amd's case, but also makes a point on the heat issue. btw these are not even the worst cases i seen a picture of one am5 chip that literally burnt a hole through the cpu's heat dispenser plate . asus for one has allways had the practice of feeding cpu's unsafe voltages with overclockers having to reign in the wild values on some boards for stability but all the big companies did it to get better benchmarks.

at one time i was partial to MSI boards because they generally worked well, then at the end of life of the board came broken bios updates some even destroying an otherwise perfectly working board (deliberate ? heh) and since then i been on ASUS boards (just dont use the software that comes with it and you will have a your board a long loooong time).

sadly ABIT went boom some time after the p4 (early xp days), they made some of the best and most stable boards back then, i had an earlier model with a celeron 300A overclocked to double of what the 300A was rated for and it still runs today (my sister has it).

[jaxa]

I believe Intel's issues are damaging millions of chips in real time, vs. a small handful of 7000X3D AMD had a problem with on specific motherboard vendors overriding the safe voltage (particularly ASUS). Intel's issue is orders of magnitude worse, still unsolved, and the scope is unclear (if it affects mobile it's truly over).

How about another class action? lulz

https://wccftech.com/second-law-firm-begins-class-action-lawsuit-investigation-on-intel-on-14th-13th-gen-cpu-instability-issues/

[revelator]

in a word YUP they are screwed. the example with the amd chips was more to point out that 3d stacking can also lead to instability (or meltdowns) especially when the board makers do something silly.

[revelator]

and the phoenix flies again lawl i just finished my retro build with the old gigabyte ga-p35-ds3r board.

gfx geforce gtx 560 ti.

harddrives two mechanical 7200 rpm seagates in raid config.

cpu core2 quad 6600.

8 gb ddr2 1600 ram corsair dominator.

asus tiger tv card PCI.

psu some old something i had laying around 450 watt.

case phoenix.

one ata CD RW and one SATA DVD RW (yup they both work).

os multi boot win2k pro winxp pro win7 ultimate 32 bit.

im actually amazed by how well it runs, plays wolfenstein 2009 no problem in highest setting.

even handles the old metro on high and crysis 1  and 2 as well. bioshock infinite also runs fine as well as a series of older games like timeshift. a good deal of dos games also run fine on it though i had to use the vdmsound driver on some of them to emulate old soundblaster cards. as for glide games i use nglide which works quite ok with the windows titles and an old dos glide emu which i cant remember where i got lulz.

browser supermium though it is probably best to keep this thing off the internetz hehe .

these old games actually look better on this machine than on my regular gamer pc go figure might be the old monitor im using.

[jaxa]

Honestly. if supermium is still being updated, maybe it will be fine.

Full-fledged Adblocker (not gimped Manifest V3) will improve performance and block some threats.

Script blocker will do more but it's not as convenient.

Maybe threats inherent to those old OSes make it totally unsafe because it will be scanned and pwned? Not sure.

https://www.nvidia.com/content/geforce-gtx/gtx_560_user_guide.pdf

Quote

Hardware Decode Acceleration: Provides ultra-smooth playback of H.264, VC-1, WMV, DivX, MPEG-2 and MPEG-4 HD and SD movies without the need for a dual or quad-core CPU.

You can decode H.264! Welcome to the future!

[revelator]

Supermium is still updated though the last version does not work on xp / win2k minimum is vista so i use a slightly older build.

Yeah for some quick lookaround the net it might be ok but its getting pretty hard to find an antivirus that works and which gets new signatures. So far for xp / win2k i only have a few choices but comodo seems to work.

Hehe yeah the old gtx 560 ti was pretty ok for its time. No vulkan mind but it does support dx11 .

For xp and lower im not sure it matters greatly as those only supported dx9.

 

[revelator]

for something more recent it seems the new nvidia 4070 super 16 gb is taking the performance cake from the 7900 xt amd. while the 7900 xt has more ram (20 gb) the super is now stocked with the faster gddr6x and more cuda cores and that seems to have done the trick in 4k atleast benchmarks now put it ahead. sadly only a few select models have 16gb the gigabyte windforce 4070 OC edition for one, the standard model only has 12 gb.

with that in mind it would have been fun to see the performance of the 3070 in a super model

[revelator]

well seems latest supermium does indeed still work with XP going to try win2k next. Last updated a month ago.

p.s antivirus might bicker because it lacks certificates groan....

[revelator]

oh as for the new 4070 super it only takes the cake in DLSS games, in raster performance AMD rules still .

well it would have been somewhat akward if the reverse held as the 7900 XT has quite a lot more performance cores, the super is higher clocked but in this case even that cant beat the XT. the DLSS improvements while nice is not a valid performance criteria.