- Mar 3, 2017
- 1,687
- 6,237
- 136
Those are a given really. Charlie sometimes likes complaining about stuff that's not even worth wasting your breath on. This would be one such case.Ryzen 8020, 8030, 8035, 8040, 8045 incoming?
(same as 7000 but with the first digit changed)
Maybe @igor_kavinski is dreaming of AMD finally applying its proven Radeon fine wine™ tech to CPUs?What ridiculous nonsense is this?
AMD can ask more If their product performs better than Intel, simple. There is no reason to cripple their own product.
The point does not stand. It's not the same at all. The 2080S was NOT a firmware upgrade, it had two more SMs (48 vs 46). It was also mostly ignored because it was a dumb release, and just barely faster than the 2080. Nvidia releases overpriced cards, because it has both the brand power and performance difference at the top. The 2070 and 2060 were IMO overpriced garbage releases (especially the 2070), but they were released at a time when the 2060 was faster than AMD's flagship at the time, Vega 64.The point stands. Nvidia sold sandbagged cards, one way or another.
2080 Super and the above were just firmware updated GPUs to new specs.
The 2080S was firmware upgrade, on Nvidia side. All they had to do was to change the BIOS of the GPU. Thats how they control the SM counts in their GPUs.The point does not stand. It's not the same at all. The 2080S was NOT a firmware upgrade, it had two more SMs (48 vs 46). It was also mostly ignored because it was a dumb release, and just barely faster than the 2080. Nvidia releases overpriced cards, because it has both the brand power and performance difference at the top. The 2070 and 2060 were IMO overpriced garbage releases (especially the 2070), but they were released at a time when the 2060 was faster than AMD's flagship at the time, Vega 64.
AMD CAN release only slower CPUs and then release faster ones (e.g. only release non-x CPUs or skip the 8800 and only release 8700 and 8900 and only later release 8800 and 8950). That's not at all the same as releasing a part and then "unlocking" more performance once competition gets heated. As I've said, it's also dumb from a benchmark perspective - people looking for your product will usually find the launch-day reviews, and reviewers would be unlikely to re-review your product.
Edit: I know that the naming convention in my example uses the older names. I know that AMD changed the numbers to a confusing mess where the digits for some reason include the release year, and Zen version and what-not.
The 2080S also had 16-Gbps rated GDDR6 vs 14-Gbps rated GDDR6 on the 2080.The 2080S was firmware upgrade, on Nvidia side. All they had to do was to change the BIOS of the GPU. Thats how they control the SM counts in their GPUs.
Same thing with all of the GPUs that were above 2080S in Stack.
Lastly, NVIDIA made the memory faster by running it at 15.5 Gbps and using 16 Gbps-rated GDDR6 memory chips compared to the 14 Gbps memory clock of the original RTX 2080.
The 2080S was firmware upgrade, on Nvidia side. All they had to do was to change the BIOS of the GPU. Thats how they control the SM counts in their GPUs.
Same thing with all of the GPUs that were above 2080S in Stack.
FWIW, aren't games more sensitive to RAM latency rather than front-end width? If games in general are programs that have low IPC, widening the front-end without a decrease in latency to system memory would not make a difference in gaming, right?It's specifically made to shred front-end bound apps like video games or cloud native(tm) server-side microservice(r) bloatware.
View attachment 87382
It's not just width.aren't games more sensitive to RAM latency rather than front-end width?
oh god decode != front-endwidening the front-end without a decrease in latency to system memory would not make a difference in gaming, right?
The more instructions you have in flight the larger your oooe window the more latency you can hide..... If you don't have decode / issue width you can't fill your big oooe window.FWIW, aren't games more sensitive to RAM latency rather than front-end width? If games in general are programs that have low IPC, widening the front-end without a decrease in latency to system memory would not make a difference in gaming, right?
Sometimes just raw bandwidth (with simple operations being carried out are large collections of vertices, etc). Depends on the engine, the game, the scene. That's why game benchmarks are all over the place. GPU dependent/CPU dependent, blah, blah, blah.FWIW, aren't games more sensitive to RAM latency rather than front-end width? If games in general are programs that have low IPC, widening the front-end without a decrease in latency to system memory would not make a difference in gaming, right?
I am honestly surprised we haven’t seen another XT chip from AMD. I imagine at least some chips are capable of hitting 6ghz.The point does not stand. It's not the same at all. The 2080S was NOT a firmware upgrade, it had two more SMs (48 vs 46). It was also mostly ignored because it was a dumb release, and just barely faster than the 2080. Nvidia releases overpriced cards, because it has both the brand power and performance difference at the top. The 2070 and 2060 were IMO overpriced garbage releases (especially the 2070), but they were released at a time when the 2060 was faster than AMD's flagship at the time, Vega 64.
AMD CAN release only slower CPUs and then release faster ones (e.g. only release non-x CPUs or skip the 8800 and only release 8700 and 8900 and only later release 8800 and 8950). That's not at all the same as releasing a part and then "unlocking" more performance once competition gets heated. As I've said, it's also dumb from a benchmark perspective - people looking for your product will usually find the launch-day reviews, and reviewers would be unlikely to re-review your product.
Edit: I know that the naming convention in my example uses the older names. I know that AMD changed the numbers to a confusing mess where the digits for some reason include the release year, and Zen version and what-not.
Never thought I'd see the day that people would argue for companies to give us less performance for our money.
That's a better idea. They can restrict the Fmax and then relax it later if the need arises.In terms of performance the worst they will do is not push the last 100Mhz which will just mean far more efficient chips or cooler running chips which is probably a net positive anyway.
Not worth it. Too much juice and GR is around the corner. That and I don’t think ZT chips would move anymore inventory.I am honestly surprised we haven’t seen another XT chip from AMD. I imagine at least some chips are capable of hitting 6ghz.
Should have done that at release. I would have paid 100 maybe 150 EUROs extra to have some kind of binned Black Edition 6GHz version of 7950x, out of the box. Just to indulge myself.Not worth it. Too much juice and GR is around the corner. That and I don’t think ZT chips would move anymore inventory.
Yep. I agree.Should have done that at release. I would have paid 100 maybe 150 EUROs extra to have some kind of binned Black Edition 6GHz version of 7950x, out of the box. Just to indulge myself.
The XT chips were more a product of TSMC's process cadence than anything else. The 3000-series originally taped out on N7, which eventually matured into N7P. So the binnings changed over time and the XT-series happened.I am honestly surprised we haven’t seen another XT chip from AMD. I imagine at least some chips are capable of hitting 6ghz.
They don't care.I am honestly surprised we haven’t seen another XT chip from AMD. I imagine at least some chips are capable of hitting 6ghz.
Hope they don't do the stupid V-cache/frequency CCD thing with the 8950X3D. Either give V-cache to both CCDs or do whatever it takes for the V-cache not to cause frequency regression compared to the non-V-cache version.They don't care.
Refresh cycles are X3D now.
Yeah this would be the holy grail, unfortunately this means putting the cache die under the core and that IMO really doesn't make sense unless they're redoing the packaging anyway - so most likely we'll have to wait for Zen 6 for this limitation to be removed....or do whatever it takes for the V-cache not to cause frequency regression compared to the non-V-cache version.
Yes they will.Hope they don't do the stupid V-cache/frequency CCD thing with the 8950X3D
No.do whatever it takes for the V-cache not to cause frequency regression compared to the non-V-cache version.
Definitely no.so most likely we have to wait for Zen 6 for this limit to be lifted.
Because they're ersatz server CPUs.I must say, I'm a bit disappointed that AMD will not change the packaging of high-end consumer CPUs, even with Zen 5
It always will cause Freq regression IMO, at least as long as the cache is put over the cores.Hope they don't do the stupid V-cache/frequency CCD thing with the 8950X3D. Either give V-cache to both CCDs or do whatever it takes for the V-cache not to cause frequency regression compared to the non-V-cache version.
The XT chips were more a product of TSMC's process cadence than anything else. The 3000-series originally taped out on N7, which eventually matured into N7P. So the binnings changed over time and the XT-series happened.
Ryzen 5000 was already N7P and Ryzen 7000 is already N5P, so no real opportunity for an XT refresh.