AMD really, really wants you to know its chips are doing OK without any help from Intel and its supply issues AMD is touting record quarterly numbers as the chip designer closed out a solid fiscal 2019. Both the quarter and the full year brought high water marks in revenue. For the fourth quarter, ended December 28: Revenues of $2.13bn were up 50 per cent from the year-ago quarter and ahead of analyst estimates at $2.11bn. Net …
Ummmm....what?
Current GPUs are not constrained by PCIe3 x8 outside of multicard setups running triple 4K setups and x16 is sufficient to address these. PCIe4 gives twice as much bandwidth again. The real issue is Intel CPU's lack of PCIe lanes rather than PCIe3.
PCIe4's big advantage is the ability to deliver high bandwidth to multiple storage devices AND GPUs concurrently.
The requirements for PCIe5? In servers/blade servers I can see some benefit but I suspect the power requirements will make desktop/laptop users very unhappy.
Upvoted.
The point you are not getting, though, is speed of each individual PCIe lane, as it augments, you theoretically need fewer of them to push the same anount of data down the pipe. Now, you correctly opined that this is good for multi-GPU, I think it is good as well as you use more and more PCIe lanes for other stuff, I have NVMe SSD's, 10G nic, and a high end GPU, all good, but merely 24 PCIe lanes available.
My point was that you lacked lanes, not speed.
The downside of PCIe4 is power usage (i.e. 5W for X470 vs 12-15W for X570 chipsets).
PCIe5 is likely to see that increase significantly again - potentially in the 20W-35W space depending on number of lanes supported. The high power usage is caused by needing to use powered extenders to increase the distance the PCIe4 signals can travel from the CPU and will have greater demands in PCIe5.
Reducing the number of lanes AND speed to conserve power puts you back in the PCIe3 era of insufficient bandwidth.... I would even speculate that low power chips will retain PCIe3 with more lanes to address performance issues rather than going to PCIe5.
Between Intel's on-going security issues, the high core-count of AMD parts and their superior single core performance I had already decided to make my next upgrade an AMD one, despite having opted for Intel for the last couple of decades. There is no question which is the better CPU at present.
>I had already decided to make my next upgrade an AMD one, despite having opted for Intel for the last couple of decades.
Make that another one, need AMD to release some MB chipsets though that are cheaper than X570 before I jump onboard, where are the B550s & A520s ?
The new Ryzen 3 CPUs will still work in the older boards, but you might need a BIOS upgrade first.
Of course, you'll probably need an older generation CPU to boot the board to upgrade the BIOS in the first place.
Theoretically one could buy a CPU from a big retailer named after a South American river, use it to flash the BIOS, and then return it for a full refund, but that would probably be against their terms of service, so you shouldn't do that...
I disagree somewhat there as most Intel CPU single core performance is still better than AMD. Also Intel have matched core count. However what AMD have done well is match very well on price. Often 20-30% cheaper on comparable chips is really good for the competition in the market and with better more power efficient chips coming from AMD they will continue to improve and match Intel which means cheaper pricing and better choices for us customers.
You are a bit out of date - clock for clock the latest AMD chips beat Intel on IPC. Only because Intel can take advantage of the mature 14nm process by overclocking to 5 and above GHz is this deficit reversed. Hence the inherent problems with the 10nm process - better IPC for intel than 14 nm (add some + here) but can't clock as high.
[disclaimer] I own a Skylake X processor based machine so not shilling for AMD [/disclaimer]
Erm, nope. If you’re looking at a single core then Intel parts can clock higher, but guzzle power when they are doing so. AMD IPC is higher, but their lower clock speed makes it about a wash for most benchmarks (some higher for Intel, some for AMD, but not much in it either way). The EPYC line is however much more power efficient and offers the option of many more cores (64 core, 128 threads in the top-of-the-range 3990X), which Intel can’t get close to, and what they do have has crazy power and cooling requirements (400W for a 56 core Xeon Platinum 9200 vs 280W for the 64 core AMD part). The Xeon part is also so expensive that Intel will only tell OEMs the price (rumoured to be circa $25k each, vs $4k for the AMD).
At end of call Su confirmed wafer supply tight which explains limited growth forecast.
Assuming Servers get priority due to profitability I expect limited marketing of consumer GPU and few price cuts across all products this year.
Good news for Intel they can keep prices high at the cost of losing maybe 5% market share, which they would probably have lost even if they had cut prices. I expect they will lose more server share especially in Cloud.
USA keen for TSMC to build fabs in USA so opportunity for Intel to sell couple of plants to TSMC in return for short term supply contract until (if) Intel gets back on track. Bad news for AMD if happens.
Pure speculation...
AMD likely reduced GPU production and delayed server CPU launch to ensure it could meet demand and avoid a paper launch. There were also suggestions that AMD struggled to produce fully operational CCX's in early batches which delayed the launch of "all core" parts.
I would also suggest that the Global Foundries lawsuit aimed at TSMC may have been to help free up capacity for more AMD CPU's...
My understanding is that TSMC are currently NOT capacity constrained at 7nm HP but that it is a recent change (i.e. December/January) and that is likely caused by other manufacturers moving to smaller nodes in preparation for upcoming launches.
"USA keen for TSMC to build fabs in USA so opportunity for Intel to sell couple of plants to TSMC in return for short term supply contract until (if) Intel gets back on track. Bad news for AMD if happens."
High-end fabs support very customised processes - it's unlikely that TSMC would wish to use an Intel fab OR that Intel would let them. At best, TSMC would get land with the appropriate permits and the shell of a building.
It's likely that TSMC would be able to get that in the time it takes for new fab equipment to be delivered (i.e. 12-18 months).
The USA is keen for TSMC to build in the USA in case Hong Kong accepts Chinese rule and China moves onto Taiwan....
Microsoft's lust for user data and contempt for user feedback, and Intel's decades of complacancy must be punished. I'm really hoping AMD will leverage into desktop market dominance by partnering with an alternative open source OS, killing the unholy Microsoft/Intel hegemony because they both deserve to burn
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
